[go: up one dir, main page]
Previous Issue
Volume 15, March-1
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
5.3
2.5
Journals
Applied Sciences
Volume 15
Issue 6
applsci-logo

Journal Menu

► Journal Menu

Journal Browser

► Journal Browser
share announcement

Appl. Sci., Volume 15, Issue 6 (March-2 2025) – 72 articles

  • Issues are regarded as officially published after their release is announced to the table of contents alert mailing list.
  • You may sign up for e-mail alerts to receive table of contents of newly released issues.
  • PDF is the official format for papers published in both, html and pdf forms. To view the papers in pdf format, click on the "PDF Full-text" link, and use the free Adobe Reader to open them.
Order results
Result details
Section
Show export options expand_more Show export options expand_less
Select all
Export citation of selected articles as:
35 pages, 1085 KiB  
Article
Multi-Channel Speech Enhancement Using Labelled Random Finite Sets and a Neural Beamformer in Cocktail Party Scenario
by Jayanta Datta, Ali Dehghan Firoozabadi, David Zabala-Blanco and Francisco R. Castillo-Soria
Appl. Sci. 2025, 15(6), 2944; https://doi.org/10.3390/app15062944 (registering DOI) - 8 Mar 2025
Abstract
In this research, a multi-channel target speech enhancement scheme is proposed that is based on deep learning (DL) architecture and assisted by multi-source tracking using a labeled random finite set (RFS) framework. A neural network based on minimum variance distortionless response (MVDR) beamformer [...] Read more.
In this research, a multi-channel target speech enhancement scheme is proposed that is based on deep learning (DL) architecture and assisted by multi-source tracking using a labeled random finite set (RFS) framework. A neural network based on minimum variance distortionless response (MVDR) beamformer is considered as the beamformer of choice, where a residual dense convolutional graph-U-Net is applied in a generative adversarial network (GAN) setting to model the beamformer for target speech enhancement under reverberant conditions involving multiple moving speech sources. The input dataset for this neural architecture is constructed by applying multi-source tracking using multi-sensor generalized labeled multi-Bernoulli (MS-GLMB) filtering, which belongs to the labeled RFS framework, to obtain estimations of the sources’ positions and the associated labels (corresponding to each source) at each time frame with high accuracy under the effect of undesirable factors like reverberation and background noise. The tracked sources’ positions and associated labels help to correctly discriminate the target source from the interferers across all time frames and generate time–frequency (T-F) masks corresponding to the target source from the output of a time-varying, minimum variance distortionless response (MVDR) beamformer. These T-F masks constitute the target label set used to train the proposed deep neural architecture to perform target speech enhancement. The exploitation of MS-GLMB filtering and a time-varying MVDR beamformer help in providing the spatial information of the sources, in addition to the spectral information, within the neural speech enhancement framework during the training phase. Moreover, the application of the GAN framework takes advantage of adversarial optimization as an alternative to maximum likelihood (ML)-based frameworks, which further boosts the performance of target speech enhancement under reverberant conditions. The computer simulations demonstrate that the proposed approach leads to better target speech enhancement performance compared with existing state-of-the-art DL-based methodologies which do not incorporate the labeled RFS-based approach, something which is evident from the 75% ESTOI and PESQ of 2.70 achieved by the proposed approach as compared with the 46.74% ESTOI and PESQ of 1.84 achieved by Mask-MVDR with self-attention mechanism at a reverberation time (RT60) of 550 ms. Full article
(This article belongs to the Special Issue Feature Paper Collection in the Section ‘Electrical, Electronics and Communications Engineering’)
27 pages, 12878 KiB  
Article
A New Extensible Feature Matching Model for Corrosion Defects Based on Consecutive In-Line Inspections and Data Clustering
by Mohamad Shatnawi and Péter Földesi
Appl. Sci. 2025, 15(6), 2943; https://doi.org/10.3390/app15062943 (registering DOI) - 8 Mar 2025
Abstract
Corrosion is considered a leading cause of failure in pipeline systems. Therefore, frequent inspection and monitoring are essential to maintain structural integrity. Feature matching based on in-line inspections (ILIs) aligns corrosion data across inspections, facilitating the observation of corrosion progression. Nonetheless, the uncertainties [...] Read more.
Corrosion is considered a leading cause of failure in pipeline systems. Therefore, frequent inspection and monitoring are essential to maintain structural integrity. Feature matching based on in-line inspections (ILIs) aligns corrosion data across inspections, facilitating the observation of corrosion progression. Nonetheless, the uncertainties of inspection tools and corrosion processes present in ILI data influence feature matching accuracy. This study proposes a new extensible feature matching model based on consecutive ILIs and data clustering. By dynamically segmenting the data into spatially localized clusters, this framework enables feature matching of isolated pairs and merging defects, as well as facilitating more precise localized transformations. Moreover, a new clustering technique—directional epsilon neighborhood clustering (DENC)—is proposed. DENC utilizes spatial graph structures and directional proximity thresholds to address the directional variability in ILI data while effectively identifying outliers. The model is evaluated on six pipeline segments with varying ILI data complexities, achieving high recall and precision of 91.5% and 98.0%, respectively. In comparison to exclusively point matching models, this work demonstrates significant improvements in terms of accuracy, stability, and managing the spatial variability and interactions of adjacent defects. These advancements establish a new framework for automated feature matching and contribute to enhanced pipeline integrity management. Full article
►▼ Show Figures

Figure 1

Figure 1
<p>Fragments of an internally corroded pipe, illustrating metal loss and wall thinning caused by corrosion. These images are sourced from the research conducted by Beben and Steliga [<a href="#B9-applsci-15-02943" class="html-bibr">9</a>].</p> Full article ">Figure 2
<p>Illustration of affine transformation on a two-dimensional plane, demonstrating how translation, scaling, and rotation facilitate correspondence between moving and reference sets.</p> Full article ">Figure 3
<p>Pipeline segmentation as proposed by Dann and Dann [<a href="#B13-applsci-15-02943" class="html-bibr">13</a>].</p> Full article ">Figure 4
<p>Pipeline unrolling and moving set double unrolling as proposed by Dann and Dann [<a href="#B13-applsci-15-02943" class="html-bibr">13</a>]—example problem.</p> Full article ">Figure 5
<p>Identification of mixed nearest neighbors using Voronoi tessellations as proposed by Amaya-Gómez et al. [<a href="#B10-applsci-15-02943" class="html-bibr">10</a>] using <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">δ</mi> </mrow> </semantics></math> = 0.11 m—example problem.</p> Full article ">Figure 6
<p>Two-dimensional presentation of features (length and width), illustrating how interaction between adjacent defects and corrosion variable growth challenge feature matching and influence defect positioning across inspections—example problem.</p> Full article ">Figure 7
<p>Illustration of the matching problems the proposed framework aims to solve, demonstrating how clustering should facilitate isolated correspondence matching, merging defect matching, and localized transformation problems across ILIs.</p> Full article ">Figure 8
<p>Illustration of the proposed model’s workflow and extensibility, highlighting its parameters, data clustering using DENC and DBSCAN, cluster classification into categories, distance-based filtering, point matching using the Voronoi model, and the process of identifying matching and outlier features.</p> Full article ">Figure 9
<p>Establishing adjacency relationships in DENC based on boundaries defined by the directional proximity thresholds <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">x</mi> </mrow> </msub> </mrow> </semantics></math> = 0.300 m and <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">y</mi> </mrow> </msub> </mrow> </semantics></math> = 0.150 m—example problem.</p> Full article ">Figure 10
<p>Graphical representation of the directed edges and binary adjacency matrix in DENC using <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">x</mi> </mrow> </msub> </mrow> </semantics></math> = 0.300 m and <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">y</mi> </mrow> </msub> </mrow> </semantics></math> = 0.150 m—example problem.</p> Full article ">Figure 11
<p>Clusters (represented by distinct colors) and outliers obtained using DENC with <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">x</mi> </mrow> </msub> </mrow> </semantics></math> = 0.300 m and <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">y</mi> </mrow> </msub> </mrow> </semantics></math> = 0.150 m—example problem.</p> Full article ">Figure 12
<p>Outlier and cluster classification, illustrating the four density-based categories: (1) one-to-one, (2) one-to-many, (3) many-to-one, and many-to-many—example problem.</p> Full article ">Figure 13
<p>Feature matching results obtained by the proposed model using <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">x</mi> </mrow> </msub> </mrow> </semantics></math> = 0.300 m, <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">y</mi> </mrow> </msub> </mrow> </semantics></math> = 0.150 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">λ</mi> </mrow> </semantics></math> = 0.250 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">δ</mi> </mrow> </semantics></math> = 0.110 m, α = 0.010, and τ = 0.001—example problem.</p> Full article ">Figure 14
<p>Feature matching results obtained by the Voronoi model [<a href="#B10-applsci-15-02943" class="html-bibr">10</a>] using <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">δ</mi> </mrow> </semantics></math> = 0.110 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">α</mi> </mrow> </semantics></math> = 0.020, and <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">τ</mi> </mrow> </semantics></math> = 0.001—example problem.</p> Full article ">Figure 15
<p>Illustration of the pipeline inspection setup from the manned wellhead platform to the unmanned wellhead platform.</p> Full article ">Figure 16
<p>Two-dimensional presentation (length and width) of all features across the six pipeline segments, S1 to S6—case study.</p> Full article ">Figure 17
<p>Sensitivity of the Voronoi model [<a href="#B7-applsci-15-02943" class="html-bibr">7</a>] to outlier proportion parameter <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">α</mi> </mrow> </semantics></math> using <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">δ</mi> </mrow> </semantics></math> = 0.110 m and <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">τ</mi> </mrow> </semantics></math> = 0.001—case study.</p> Full article ">Figure 18
<p>Sensitivity of the Voronoi model [<a href="#B7-applsci-15-02943" class="html-bibr">7</a>] to defect’s position uncertainty threshold <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">δ</mi> </mrow> </semantics></math> using <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">α</mi> </mrow> </semantics></math> = 0.080 and <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">τ</mi> </mrow> </semantics></math> = 0.001—case study.</p> Full article ">Figure 19
<p>Sensitivity of the proposed model to DENC’s directional proximity thresholds <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">x</mi> </mrow> </msub> </mrow> </semantics></math> and <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">y</mi> </mrow> </msub> </mrow> </semantics></math> using <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">λ</mi> </mrow> </semantics></math> = 0.250 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">δ</mi> </mrow> </semantics></math> = 0.110 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">α</mi> </mrow> </semantics></math> = 0.080, and <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">τ</mi> </mrow> </semantics></math> = 0.001—case study.</p> Full article ">Figure 20
<p>Sensitivity of the proposed model to outlier proportion parameter <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">α</mi> </mrow> </semantics></math> using <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">x</mi> </mrow> </msub> </mrow> </semantics></math> = 0.220 m, <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">y</mi> </mrow> </msub> </mrow> </semantics></math> = 0.110 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">λ</mi> </mrow> </semantics></math> = 0.250 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">δ</mi> </mrow> </semantics></math> = 0.110 m, and <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">τ</mi> </mrow> </semantics></math> = 0.001—case study.</p> Full article ">Figure 21
<p>Sensitivity of the proposed model to the merging distance threshold <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">λ</mi> </mrow> </semantics></math> using <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">x</mi> </mrow> </msub> </mrow> </semantics></math> = 0.220 m, <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">y</mi> </mrow> </msub> </mrow> </semantics></math> = 0.110 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">δ</mi> </mrow> </semantics></math> = 0.110 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">α</mi> </mrow> </semantics></math> = 0.043 m, and <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">τ</mi> </mrow> </semantics></math> = 0.001—case study.</p> Full article ">Figure 22
<p>Sensitivity of the proposed DBSCAN-based alternative model to the proximity threshold <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> </semantics></math> using <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">λ</mi> </mrow> </semantics></math> = 0.250 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">δ</mi> </mrow> </semantics></math> = 0.110 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">α</mi> </mrow> </semantics></math> = 0.080, and <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">τ</mi> </mrow> </semantics></math> = 0.001—case study.</p> Full article ">Figure 23
<p>Sensitivity of the proposed DBSCAN-based alternative model to outlier proportion parameter <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">α</mi> </mrow> </semantics></math> using <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> </semantics></math> = 0.250 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">λ</mi> </mrow> </semantics></math> = 0.250 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">δ</mi> </mrow> </semantics></math> = 0.110 m, and <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">τ</mi> </mrow> </semantics></math> = 0.001—case study.</p> Full article ">Figure 24
<p>Sensitivity of the proposed DBSCAN-based alternative model to the merging distance threshold <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">λ</mi> </mrow> </semantics></math> using <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> </semantics></math> = 0.250 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">δ</mi> </mrow> </semantics></math> = 0.110 m, <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">α</mi> </mrow> </semantics></math> = 0.055, and <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">τ</mi> </mrow> </semantics></math> = 0.001—case study.</p> Full article ">Figure 25
<p>Feature clustering (represented by distinct colors) using DBSCAN (top) and DENC (bottom) using <math display="inline"><semantics> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> </semantics></math> = 0.250 m, <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">x</mi> </mrow> </msub> </mrow> </semantics></math> = 0.220 m, and <math display="inline"><semantics> <mrow> <msub> <mrow> <mi mathvariant="sans-serif">ε</mi> </mrow> <mrow> <mi mathvariant="normal">y</mi> </mrow> </msub> </mrow> </semantics></math> = 0.110 m—case study for segment S5.</p> Full article ">
28 pages, 477 KiB  
Review
Leveraging Digital Twin Technology for Sustainable and Efficient Public Transportation
by Babin Manandhar, Kayode Dunkel Vance, Danda B. Rawat and Nadir Yilmaz
Appl. Sci. 2025, 15(6), 2942; https://doi.org/10.3390/app15062942 (registering DOI) - 8 Mar 2025
Abstract
Public transportation systems face numerous challenges like traffic congestion, inconsistent schedules, and variable passenger demand. These issues lead to delays, overcrowding, and reduced patron satisfaction. Digital twin (DT) technology is a promising innovation for improving public transportation systems by offering real-time virtual representations [...] Read more.
Public transportation systems face numerous challenges like traffic congestion, inconsistent schedules, and variable passenger demand. These issues lead to delays, overcrowding, and reduced patron satisfaction. Digital twin (DT) technology is a promising innovation for improving public transportation systems by offering real-time virtual representations of physical systems. By integrating real-time data from various sources, digital twins can enable predictive analytics, optimize operations, and improve the overall performance of public transportation networks. This work explores the potential of digital twins to optimize operational efficiency, enhance passenger experiences, and support sustainable urban mobility. A comprehensive review of the existing literature was conducted by analyzing case studies, theoretical models, and practical implementations to assess the effectiveness of DTs in transit systems. While the benefits of DTs are significant, their successful implementation in bus transportation systems is impeded by several challenges like scalability limitations, interoperability issues, and technical complexities involving data integration and IT infrastructure. This paper discusses ways to overcome these challenges, which include using modular designs, microservices, blockchain for security, and standardized communication for better integration. It emphasizes the importance of collaboration in research and practice to effectively apply digital twin technology to public transit systems. Full article
13 pages, 963 KiB  
Article
Responsiveness to the Context: Information–Task–Situation Decisional Strategies and Electrophysiological Correlates
by Angelica Daffinà, Carlotta Acconito and Michela Balconi
Appl. Sci. 2025, 15(6), 2941; https://doi.org/10.3390/app15062941 (registering DOI) - 8 Mar 2025
Abstract
Decision-making, defined as a cognitive process involving the selection of a course of action among several alternatives, is pivotal in personal and professional life and is founded on responsiveness to the context of decisional strategies—in terms of pieces of contextual features collected, evaluated, [...] Read more.
Decision-making, defined as a cognitive process involving the selection of a course of action among several alternatives, is pivotal in personal and professional life and is founded on responsiveness to the context of decisional strategies—in terms of pieces of contextual features collected, evaluated, and integrated. This study explored the behavioral and electrophysiological (EEG) correlates of individual tendencies to rely on three distinct decisional strategies: Information (I-ds), Situation (S-ds), or Task (T-ds). A total of 51 individuals performed a decision-making task that required participants to face real-life decision-making situations, during which an unexpected event prompted them to appraise the situation and rely on different sources of contextual features to make the best decision and manage the problem. The behavioral data and EEG frequency bands (delta, theta, alpha, beta, and gamma) were collected during the decision-making task. The results evidenced a general predisposition to adopt a T-ds. In addition, EEG findings reported a higher increase in theta band power in the right frontal area (AF8) compared to the left temporoparietal site (TP9). Moreover, for the gamma band, higher activity was found in the T-ds compared to the I-ds in AF8. Overall, responsiveness to the context was closely linked to the assignment’s requirements. Additionally, adopting a T-ds requires high levels of multilevel attention control systems and a significant workload on human performance. Nevertheless, the T-ds remain the most employed type of responsiveness to the context approach, when compared to situational and contextual aspects. Full article
(This article belongs to the Special Issue Decision-Making and Decision Support Systems: Methods and Applications)
►▼ Show Figures

Figure 1

Figure 1
<p>Behavioral results. The bar chart shows significant differences in Strategy, with higher scores in S-ds compared to I-ds and in T-ds compared to I-ds and S-ds. Bars represent ±1 Standard Error and stars (*) mark statistically significant comparisons.</p> Full article ">Figure 2
<p>EEG results: theta band. The bar chart shows significant differences for the theta band in Electrodes, with higher activity in AF8 compared to TP9. Bars represent ±1 Standard Error and stars (*) mark statistically significant comparisons. The more intense color in the rendering of the head (on the right) represents the increase in EEG power at specific EEG electrodes.</p> Full article ">Figure 3
<p>EEG results: gamma band. The bar chart shows significant differences for the gamma band in Strategy × Electrodes, with higher activity in T-ds compared to I-ds in AF8. Bars represent ±1 standard error and stars (*) mark statistically significant comparisons. The more intense color in the rendering of the head (below) represents the increase in EEG power at the specific EEG electrode for each strategy.</p> Full article ">
37 pages, 13572 KiB  
Article
Integrating Biological Domain Knowledge with Machine Learning for Identifying Colorectal-Cancer-Associated Microbial Enzymes in Metagenomic Data
by Burcu Bakir-Gungor, Nur Sebnem Ersoz and Malik Yousef
Appl. Sci. 2025, 15(6), 2940; https://doi.org/10.3390/app15062940 (registering DOI) - 8 Mar 2025
Abstract
Advances in metagenomics have revolutionized our ability to elucidate links between the microbiome and human diseases. Colorectal cancer (CRC), a leading cause of cancer-related mortality worldwide, has been associated with dysbiosis of the gut microbiome. This study aims to develop a method for [...] Read more.
Advances in metagenomics have revolutionized our ability to elucidate links between the microbiome and human diseases. Colorectal cancer (CRC), a leading cause of cancer-related mortality worldwide, has been associated with dysbiosis of the gut microbiome. This study aims to develop a method for identifying CRC-associated microbial enzymes by incorporating biological domain knowledge into the feature selection process. Conventional feature selection techniques often evaluate features individually and fail to leverage biological knowledge during metagenomic data analysis. To address this gap, we propose the enzyme commission (EC)-nomenclature-based Grouping-Scoring-Modeling (G-S-M) method, which integrates biological domain knowledge into feature grouping and selection. The proposed method was tested on a CRC-associated metagenomic dataset collected from eight different countries. Community-level relative abundance values of enzymes were considered as features and grouped based on their EC categories to provide biologically informed groupings. Our findings in randomized 10-fold cross-validation experiments imply that glycosidases, CoA-transferases, hydro-lyases, oligo-1,6-glucosidase, crotonobetainyl-CoA hydratase, and citrate CoA-transferase enzymes can be associated with CRC development as part of different molecular pathways. These enzymes are mostly synthesized by Eschericia coli, Salmonella enterica, Klebsiella pneumoniae, Staphylococcus aureus, Streptococcus pneumoniae, and Clostridioides dificile. Comparative evaluation experiments showed that the proposed model consistently outperforms traditional feature selection methods paired with various classifiers. Full article
(This article belongs to the Section Computing and Artificial Intelligence)
►▼ Show Figures

Figure 1

Figure 1
<p>Enzyme commission (EC) nomenclature involves seven main enzyme groups with many subclasses, each related to a specific enzyme activity.</p> Full article ">Figure 2
<p>The proposed EC-nomenclature-based G-S-M workflow for analyzing enzyme abundance values obtained from disease-associated metagenomics relative enzyme abundance datasets.</p> Full article ">Figure 3
<p>Detailed description of the grouping, scoring, and modeling components in the proposed EC-nomenclature-based G-S-M approach.</p> Full article ">Figure 4
<p>(<b>A</b>) Top 10 important enzyme groups and (<b>B</b>) top scored enzyme in each enzyme group identified by the EC-nomenclature-based G-S-M method applied to the CRC-associated metagenomic datasets, including relative abundance values of the enzymes. The −log 10 <span class="html-italic">p</span>-values indicate the significance values assigned by the robust rank aggregation method. Each color represents the enzyme commission (EC) activity.</p> Full article ">Figure 5
<p>Performance metrics of the EC-nomenclature-based G-S-M method when applied to population-specific CRC-associated metagenomic dataset, including relative abundance values of the enzymes. (<b>A</b>) AUC values and (<b>B</b>) # of enzymes (features) selected for population-specific datasets.</p> Full article ">Figure 6
<p>AUC values of traditional feature selection methods, including XGB, SKB, IG, MRMR, CMIM, and FCBF, when coupled with different classifiers, including Adaboost, DT, LogitBoost, RF, SVM_opt, Stack_Logitboost_Kmenas, Stack_SVM_Kmeans, and XGBoost, compared with the AUC metrics of the EC-nomenclature-based G-S-M approach when coupled with RF, XGBoost, and DT classifiers and tested on the CRC-associated metagenomic dataset.</p> Full article ">Figure 7
<p>Top 10 most important enzymes detected by (<b>A</b>) XGB and (<b>B</b>) SKB feature selection methods. The colors represent different enzyme functions, as defined by the enzyme commission.</p> Full article ">Figure 8
<p>Top 3 scoring enzyme groups and top 10 scoring enzymes included in these groups, which are identified by the EC-nomenclature-based G-S-M model when applied to the CRC-associated metagenomic dataset. Each color represents the related enzyme commission (EC) activity.</p> Full article ">Figure 9
<p>GO MF terms for the top 3 scoring enzymes that belong to the top 3 scoring enzyme groups, which were identified by the EC-nomenclature-based G-S-M model applied to the CRC-associated metagenomic dataset. GO hierarchy was obtained from the Quick-GO annotations provided by NCBI.</p> Full article ">Figure 10
<p>Top 100 scoring enzymes obtained by the EC-nomenclature-based G-S-M method applied to the CRC-associated metagenomic dataset and their related KEGG pathways.</p> Full article ">Figure 11
<p>The enzymes in the glycosidases (EC: 3.2.1) group and their metabolic pathways, i.e., (<b>A</b>) starch and sucrose metabolism, (<b>B</b>) sphingolipid metabolism, (<b>C</b>) galactose metabolism, (<b>D</b>) <span class="html-italic">N</span>-glycan biosynthesis, and (<b>E</b>) glucuronate interconversions. All pathway information is excerpted from the KEGG database.</p> Full article ">Figure 12
<p>Metabolic pathways of top 3 scoring enzyme groups found using the EC-nomenclature-based G-S-M approach applied to the CRC-associated metagenomic dataset, including relative abundance values of the enzymes. EC: 4.2.1 and EC: 2.8.3 groups of enzymes perform activities in (<b>A</b>) styrene degradation (KEGG database), in (<b>B</b>) butanoate metabolism (KEGG database), in (<b>C</b>) the citric acid cycle (BRENDA database), and in (<b>D</b>) carnitine metabolism (BRENDA database). Each color represents the related enzyme commission (EC) activity.</p> Full article ">Figure 13
<p>Network of top 100 enzymes that have been identified by the EC-nomenclature-based G-S-M on the CRC-associated metagenomic data and associated organisms. A total of 268 species that synthesize the top scoring 100 enzymes are presented. Node size is associated with betweenness centrality.</p> Full article ">Figure 14
<p>Network of top 16 scoring enzymes that were identified either by the XGB or SKB feature selection methods as part of their top 10 scoring lists and their associated organisms. A total of 85 species that synthesize the top scoring enzymes are presented. Node size is associated with betweenness centrality.</p> Full article ">Figure 15
<p>Correlations among the top 10 enzyme groups that were selected by the EC-nomenclature-based G-S-M for different CRC-associated metagenomic datasets, including the relative abundance values of the enzymes obtained from the samples belonging to different populations. The Jaccard index is used to calculate the correlation between the identified EC groups among two populations.</p> Full article ">Figure 16
<p>Commonalities among the top 10 enzyme groups that were selected by the EC-nomenclature-based G-S-M for different CRC-associated metagenomic datasets, including the relative abundance values of the enzymes obtained from the samples belonging to different populations.</p> Full article ">Figure 17
<p>Performance metrics of different feature selection methods (XGB, SKB, IG, MRMR, FCBF, and CMIM) coupled with the RF classifier, RCE with the SVM classifier, and the EC-nomenclature-based G-S-M approach with RF when tested on the CRC-associated metagenomic dataset, including relative abundance values of the enzymes.</p> Full article ">Figure 18
<p>Correlations among the top 100 features that are selected by different feature selection algorithms and the EC-nomenclature-based G-S-M approach when tested on the CRC-associated metagenomic dataset, including relative abundance values of the enzymes.</p> Full article ">Figure 19
<p>Commonalities among the top 100 features that were selected by different feature selection algorithms and the EC-nomenclature-based G-S-M approach when tested on CRC-associated metagenomic dataset, including relative abundance values of the enzymes.</p> Full article ">
25 pages, 6330 KiB  
Article
Post-Filtering of Noisy Images Compressed by HEIF
by Sergii Kryvenko, Volodymyr Rebrov, Vladimir Lukin, Vladimir Golovko, Anatoliy Sachenko, Andrii Shelestov and Benoit Vozel
Appl. Sci. 2025, 15(6), 2939; https://doi.org/10.3390/app15062939 (registering DOI) - 8 Mar 2025
Abstract
Modern imaging systems produce a great volume of image data. In many practical situations, it is necessary to compress them for faster transferring or more efficient storage. Then, a compression has to be applied. If images are noisy, lossless compression is almost useless, [...] Read more.
Modern imaging systems produce a great volume of image data. In many practical situations, it is necessary to compress them for faster transferring or more efficient storage. Then, a compression has to be applied. If images are noisy, lossless compression is almost useless, and lossy compression is characterized by a specific noise filtering effect that depends on the image, noise, and coder properties. Here, we considered a modern HEIF coder applied to grayscale (component) images of different complexity corrupted by additive white Gaussian noise. It has recently been shown that an optimal operation point (OOP) might exist in this case. Note that the OOP is a value of quality factor where the compressed image quality (according to a used quality metric) is the closest to the corresponding noise-free image. The lossy compression of noisy images leads to both noise reduction and distortions introduced into the information component, thus, a compromise should be found between the compressed image quality and compression ratio attained. The OOP is one possible compromise, if it exists, for a given noisy image. However, it has also recently been demonstrated that the compressed image quality can be significantly improved if post-filtering is applied under the condition that the quality factor is slightly larger than the one corresponding to the OOP. Therefore, we considered the efficiency of post-filtering where a block-matching 3-dimensional (BM3D) filter was applied. It was shown that the positive effect of such post-filtering could reach a few dB in terms of the PSNR and PSNR-HVS-M metrics. The largest benefits took place for simple structure images and a high intensity of noise. It was also demonstrated that the filter parameters have to be adapted to the properties of residual noise that become more non-Gaussian if the compression ratio increases. Practical recommendations on the use of compression parameters and post-filtering are given. Full article
(This article belongs to the Special Issue Trends and Prospects in Computer Vision and Pattern Recognition Technology)
►▼ Show Figures

Figure 1

Figure 1
<p>Test single-channel (grayscale) images Frisco (<b>a</b>), Fr03 (<b>b</b>), and Diego (<b>c</b>), all 512 × 512 pixels.</p> Full article ">Figure 2
<p>Dependences of MSE<sub>tc</sub> (calculated between the true and compressed images) for images Diego (<b>a</b>,<b>c</b>) and Frisco (<b>b</b>,<b>d</b>) for <math display="inline"><semantics> <mrow> <msup> <mrow> <mi>σ</mi> </mrow> <mrow> <mn>2</mn> </mrow> </msup> <mo>=</mo> <mn>25</mn> <mo> </mo> </mrow> </semantics></math>(<b>a</b>,<b>b</b>) and <math display="inline"><semantics> <mrow> <msup> <mrow> <mi>σ</mi> </mrow> <mrow> <mn>2</mn> </mrow> </msup> <mo>=</mo> <mn>196</mn> </mrow> </semantics></math> (<b>c</b>,<b>d</b>) for five coders.</p> Full article ">Figure 3
<p>Visualized <math display="inline"><semantics> <mrow> <msub> <mrow> <mo>∆</mo> </mrow> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> </mrow> </semantics></math>, <math display="inline"><semantics> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>,</mo> <msub> <mrow> <mi>I</mi> </mrow> <mrow> <mi>I</mi> <mi>m</mi> </mrow> </msub> <mo>,</mo> <mo> </mo> <mi>j</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>,</mo> <msub> <mrow> <mi>J</mi> </mrow> <mrow> <mi>I</mi> <mi>m</mi> </mrow> </msub> </mrow> </semantics></math> for the test images Frisco (<b>a</b>) and Diego (<b>b</b>) corrupted by AWGN with σ<sup>2</sup> = 100 compressed by HEIF with QF = 50.</p> Full article ">Figure 4
<p>Dependences of the output metrics on QF and β for two test images and three metrics, σ<sup>2</sup> = 50. Plots of PSNR (<b>a</b>,<b>b</b>), PSNR-HVS (<b>c</b>,<b>d</b>) and FSIM (<b>e</b>,<b>f</b>) for the test images Frisco (<b>a</b>,<b>c</b>,<b>e</b>) and Fr03 (<b>b</b>,<b>d</b>,<b>f</b>).</p> Full article ">Figure 4 Cont.
<p>Dependences of the output metrics on QF and β for two test images and three metrics, σ<sup>2</sup> = 50. Plots of PSNR (<b>a</b>,<b>b</b>), PSNR-HVS (<b>c</b>,<b>d</b>) and FSIM (<b>e</b>,<b>f</b>) for the test images Frisco (<b>a</b>,<b>c</b>,<b>e</b>) and Fr03 (<b>b</b>,<b>d</b>,<b>f</b>).</p> Full article ">Figure 5
<p>Dependences of the output metrics on QF and β for two test images and three metrics, σ<sup>2</sup> = 100. Plots of PSNR (<b>a</b>,<b>b</b>), PSNR-HVS (<b>c</b>,<b>d</b>) and FSIM (<b>e</b>,<b>f</b>) for the test images Frisco (<b>a</b>,<b>c</b>,<b>e</b>) and Fr03 (<b>b</b>,<b>d</b>,<b>f</b>).</p> Full article ">Figure 5 Cont.
<p>Dependences of the output metrics on QF and β for two test images and three metrics, σ<sup>2</sup> = 100. Plots of PSNR (<b>a</b>,<b>b</b>), PSNR-HVS (<b>c</b>,<b>d</b>) and FSIM (<b>e</b>,<b>f</b>) for the test images Frisco (<b>a</b>,<b>c</b>,<b>e</b>) and Fr03 (<b>b</b>,<b>d</b>,<b>f</b>).</p> Full article ">Figure 6
<p>Dependences of the output metrics on QF and β for two test images and three metrics, σ<sup>2</sup> = 200. Plots of PSNR (<b>a</b>,<b>b</b>), PSNR-HVS (<b>c</b>,<b>d</b>) and FSIM (<b>e</b>,<b>f</b>) for the test images Frisco (<b>a</b>,<b>c</b>,<b>e</b>) and Fr03 (<b>b</b>,<b>d</b>,<b>f</b>).</p> Full article ">Figure 6 Cont.
<p>Dependences of the output metrics on QF and β for two test images and three metrics, σ<sup>2</sup> = 200. Plots of PSNR (<b>a</b>,<b>b</b>), PSNR-HVS (<b>c</b>,<b>d</b>) and FSIM (<b>e</b>,<b>f</b>) for the test images Frisco (<b>a</b>,<b>c</b>,<b>e</b>) and Fr03 (<b>b</b>,<b>d</b>,<b>f</b>).</p> Full article ">Figure 7
<p>Dependences of PSNR<sub>p pf</sub> (CR) and ∆PSNR<sub>p tc</sub> (CR) (<b>a</b>–<b>c</b>) and ∆PSNR-HVS-M<sub>p pf</sub> (CR) and ∆PSNR-HVS-M<sub>p tc</sub> (CR) (<b>d</b>–<b>e</b>) for the test images Fr03 (<b>a</b>,<b>c</b>,<b>d</b>,<b>f</b>) and Frisco (<b>b</b>,<b>e</b>) for σ<sup>2</sup> = 100 (<b>a</b>,<b>b</b>,<b>d</b>,<b>e</b>) and σ<sup>2</sup> = 200 (<b>c</b>,<b>f</b>).</p> Full article ">Figure 8
<p>Test images Med5 (<b>a</b>), Med7 (<b>b</b>), and Med8 (<b>c</b>) used in the additional experiments.</p> Full article ">Figure 9
<p>Noisy image Frisco (<b>a</b>), the results of its compression in the OOP by HEIF (<b>b</b>), compressed with QF = 36 and post-filtered with the optimal β after decompression (<b>c</b>), noisy image Diego (<b>d</b>), the results of its compression in the OOP by HEIF (<b>e</b>), compressed with QF = 36 and post-filtered with the optimal β after decompression (<b>f</b>).</p> Full article ">
30 pages, 22071 KiB  
Article
Analysis of Optical Errors in Joint Fabry–Pérot Interferometer–Fourier-Transform Imaging Spectroscopy Interferometric Super-Resolution Systems
by Yu Zhang, Qunbo Lv, Jianwei Wang, Yinhui Tang, Jia Si, Xinwen Chen and Yangyang Liu
Appl. Sci. 2025, 15(6), 2938; https://doi.org/10.3390/app15062938 (registering DOI) - 8 Mar 2025
Abstract
Fourier-transform imaging spectroscopy (FTIS) faces inherent limitations in spectral resolution due to the maximum optical path difference (OPD) achievable by its interferometer. To overcome this constraint, we propose a novel spectral super-resolution technology integrating a Fabry–Pérot interferometer (FPI) with FTIS, termed multi-component joint [...] Read more.
Fourier-transform imaging spectroscopy (FTIS) faces inherent limitations in spectral resolution due to the maximum optical path difference (OPD) achievable by its interferometer. To overcome this constraint, we propose a novel spectral super-resolution technology integrating a Fabry–Pérot interferometer (FPI) with FTIS, termed multi-component joint interferometric hyperspectral imaging (MJI-HI). This method leverages the FPI to periodically modulate the target spectrum, enabling FTIS to capture a modulated interferogram. By encoding high-frequency spectral interference information into low-frequency interference regions through FPI modulation, an advanced inversion algorithm is developed to reconstruct the encoded high-frequency components, thereby achieving spectral super-resolution. This study analyzes the impact of primary optical errors and tolerance thresholds in the FPI and FTIS on the interferograms and spectral fidelity of MJI-HI, along with proposing algorithmic improvements. Notably, certain errors in the FTIS and FPI exhibit mutual interference. The theoretical framework for error analysis is validated and discussed through numerical simulations, providing critical theoretical support for subsequent instrument development and laying a foundation for advancing novel spectral super-resolution technologies. Full article
(This article belongs to the Special Issue Spectral Detection: Technologies and Applications—2nd Edition)
►▼ Show Figures

Figure 1

Figure 1
<p>Schematic diagram of FPI principle [<a href="#B26-applsci-15-02938" class="html-bibr">26</a>].</p> Full article ">Figure 2
<p>Diagram of MJI-HI system structure.</p> Full article ">Figure 3
<p>Schematic diagram of combined interference principle: (<b>a</b>) object spectrum <span class="html-italic">B</span>(<span class="html-italic">ν</span>) and (<b>b</b>) its interferogram <span class="html-italic">I</span><sub>0</sub>(Δ) and (<b>c</b>) FPI-modulated spectrum <span class="html-italic">B</span>(<span class="html-italic">ν</span>)<span class="html-italic">T<sub>FPI</sub></span>(<span class="html-italic">ν</span>) and (<b>d</b>) its interferogram <span class="html-italic">I</span><sub>2</sub>(Δ).</p> Full article ">Figure 4
<p>(<b>a</b>) The schematic diagram of the FPI non-parallelism error in the direction perpendicular to the optical axis and a locally magnified schematic diagram of the flatness error. (<b>b</b>) The schematic diagram of the distribution of the non-parallelism error along the optical axis.</p> Full article ">Figure 5
<p>(<b>a</b>) The comparison and local magnification of the FPI transmittance spectra under ideal conditions (<span class="html-italic">T<sub>FPI</sub></span>) and with parallelism errors (<span class="html-italic">T<sub>FPI</sub></span><sub>-</sub><span class="html-italic"><sub>para</sub></span>); <span class="html-italic">T<sub>FPI</sub></span> and <span class="html-italic">T<sub>FPI</sub></span><sub>-</sub><span class="html-italic"><sub>para</sub></span> are decomposed into (<b>b</b>) the DC component, (<b>c</b>) fundamental frequency component, (<b>d</b>) second harmonic component, and (<b>e</b>) third harmonic component, with corresponding comparisons and local magnifications. Here, <span class="html-italic">z</span> is defined as 2<span class="html-italic">πν</span>2<span class="html-italic">r</span><sub>0</sub><span class="html-italic">β<sub>FPI</sub></span> in Equation (12), and the curves ±2<span class="html-italic">a<sub>m</sub>J</span><sub>1</sub>(<span class="html-italic">mz</span>)/(<span class="html-italic">mz</span>) can be regarded as the envelopes of the fundamental frequency and higher harmonic components of <span class="html-italic">T<sub>FPI</sub></span><sub>-</sub><span class="html-italic"><sub>para</sub></span>.</p> Full article ">Figure 6
<p>(<b>a</b>) The comparison and local magnification of the FPI transmittance spectra under ideal conditions (<span class="html-italic">T<sub>FPI</sub></span>) and with parallelism errors (<span class="html-italic">T<sub>FPI</sub></span><sub>-<span class="html-italic">flat</span></sub>); <span class="html-italic">T<sub>FPI</sub></span> and <span class="html-italic">T<sub>FPI</sub></span><sub>-<span class="html-italic">flat</span></sub> are decomposed into (<b>b</b>) the DC component, (<b>c</b>) fundamental frequency component, (<b>d</b>) second harmonic component, and (<b>e</b>) third harmonic component, with corresponding comparisons and local magnifications. Here, the curves ±2<span class="html-italic">a<sub>m</sub></span>exp[−(<span class="html-italic">mπσ</span>Δ<span class="html-italic"><sub>D</sub>ν</span>)<sup>2</sup>] can be regarded as the envelopes of the fundamental frequency and higher harmonic components of <span class="html-italic">T<sub>FPI</sub></span><sub>-<span class="html-italic">flat</span></sub>.</p> Full article ">Figure 7
<p>The curves of <span class="html-italic">a′</span><sub>0</sub> to <span class="html-italic">a′</span><sub>3</sub> as functions of <span class="html-italic">R</span><sub>0</sub>. The zero points of <span class="html-italic">a′<sub>m</sub></span> with respect to <span class="html-italic">R</span><sub>0</sub> are indicated by arrows, representing that at these points, <span class="html-italic">a<sub>m</sub></span> is less affected by changes in <span class="html-italic">R</span><sub>0</sub>.</p> Full article ">Figure 8
<p>(<b>a</b>) The comparison of the FPI transmittance spectrum under ideal conditions (TFPI) and under reflectance variation errors (<span class="html-italic">T<sub>FPI-R</sub></span>); TFPI and <span class="html-italic">T<sub>FPI-R</sub></span> decomposed into (<b>b</b>) the DC component, (<b>c</b>) the fundamental frequency component, (<b>d</b>) the second harmonic component, and (<b>e</b>) the third harmonic component for comparison.</p> Full article ">Figure 9
<p>Beams with different divergence angles produce different OPDs in the FTIS and FPI. Here, <span class="html-italic">θ</span><sub>0</sub> is the maximum divergence half-angle, and ψ is the azimuth angle. The red arrows indicate the phase differences caused by the central aperture light in the FTIS and FPI, respectively, and the black arrows indicate the phase differences caused by the edge aperture light in the FTIS and FPI, respectively.</p> Full article ">Figure 10
<p>Assuming a divergence angle of 1°, (<b>a</b>) FTIS with an OPD of 1 mm, (<b>b</b>) FTIS with an OPD of 2 mm, and (<b>c</b>) FTIS with an OPD of 3 mm: the overall comparison and local magnification of the FTIS double-beam interferometric spectrum.</p> Full article ">Figure 11
<p>(<b>a</b>) In FTIS, the mirror tilt error causes the wavefront (blue surface) to deviate from the ideal wavefront (yellow surface), tilting in the <span class="html-italic">φ<sub>FTIS</sub></span> direction with a tilt angle of 2<span class="html-italic">β<sub>FTIS</sub></span>; in FPI, the non-parallelism error causes the wavefront (blue surface) to deviate from the ideal wavefront (yellow surface), tilting in the <span class="html-italic">φ<sub>FPI</sub></span> direction with a tilt angle of 2<span class="html-italic">β<sub>FPI</sub></span>, where the Z-axis represents the optical axis. (<b>b</b>) The relative tilt direction along the optical axis between the FTIS mirror tilt error and the FPI non-parallelism error.</p> Full article ">Figure 12
<p>(<b>a</b>) The schematic diagram of the geometric relationship in polar coordinates between the vector (<span class="html-italic">β<sub>m</sub></span>,<span class="html-italic">φ<sub>m</sub></span>) and the vectors <math display="inline"><semantics> <mrow> <msub> <mrow> <mover accent="true"> <mrow> <mi>x</mi> </mrow> <mo>→</mo> </mover> </mrow> <mrow> <mi>F</mi> <mi>T</mi> <mi>I</mi> <mi>S</mi> </mrow> </msub> </mrow> </semantics></math> and <math display="inline"><semantics> <mrow> <msub> <mrow> <mover accent="true"> <mrow> <mi>x</mi> </mrow> <mo>→</mo> </mover> </mrow> <mrow> <mi>F</mi> <mi>P</mi> <mi>I</mi> </mrow> </msub> </mrow> </semantics></math>; (<b>b</b>) the schematic diagram of the geometric relationship between the vector (<span class="html-italic">β<sub>−m</sub></span>,<span class="html-italic">φ<sub>−m</sub></span>) and the vectors <math display="inline"><semantics> <mrow> <msub> <mrow> <mover accent="true"> <mrow> <mi>x</mi> </mrow> <mo>→</mo> </mover> </mrow> <mrow> <mi>F</mi> <mi>T</mi> <mi>I</mi> <mi>S</mi> </mrow> </msub> </mrow> </semantics></math> and <math display="inline"><semantics> <mrow> <msub> <mrow> <mover accent="true"> <mrow> <mi>x</mi> </mrow> <mo>→</mo> </mover> </mrow> <mrow> <mi>F</mi> <mi>P</mi> <mi>I</mi> </mrow> </msub> </mrow> </semantics></math>.</p> Full article ">Figure 13
<p>(<b>a1</b>) Hitran data input spectrum and (<b>a2</b>) the error-free interferograms <span class="html-italic">I</span><sub>1</sub> and <span class="html-italic">I</span><sub>2</sub> obtained using MJI-HI; (<b>b1</b>) Gaussian function input spectrum and (<b>b2</b>) the error-free interferograms <span class="html-italic">I</span><sub>1</sub> and <span class="html-italic">I</span><sub>2</sub>; (<b>c1</b>) cosine function input spectrum and (<b>c2</b>) the error-free interferograms <span class="html-italic">I</span><sub>1</sub> and <span class="html-italic">I</span><sub>2</sub>; (<b>d1</b>) <span class="html-italic">B</span><sub>0</sub> = 1 input spectrum and (<b>d2</b>) the error-free interferograms <span class="html-italic">I</span><sub>1</sub> and <span class="html-italic">I</span><sub>2</sub>.</p> Full article ">Figure 14
<p>(<b>a1</b>–<b>a4</b>). Broadening of the displacement components of orders <span class="html-italic">m</span> = 0, 1, 2, and 3 caused by different non-parallelism; (<b>b1</b>–<b>b4</b>). broadening of the displacement components of orders <span class="html-italic">m</span> = 0, 1, 2, and 3 caused by different non-flatness.</p> Full article ">Figure 15
<p>(<b>a1</b>) Interferogram reconstruction result when the <span class="html-italic">FWHM<sub>D</sub></span> of <span class="html-italic">Defect</span><sub>1</sub> is λ<sub>min</sub>/4; (<b>a2</b>–<b>a4</b>) are its local enlargements; (<b>b1</b>) interferogram reconstruction result when the <span class="html-italic">FWHM<sub>D</sub></span> of <span class="html-italic">Defect</span><sub>1</sub> is λ<sub>min</sub>/2; (<b>b2</b>–<b>b4</b>) are its local enlargements; (<b>c1</b>) interferogram reconstruction result when the <span class="html-italic">FWHM<sub>D</sub></span> of <span class="html-italic">Defect</span><sub>1</sub> is λ<sub>min</sub>; (<b>c2</b>–<b>c4</b>) are its local enlargements. Here, <span class="html-italic">I</span><sub>0</sub> is the ideal interferogram, <span class="html-italic">I<sub>sup</sub></span> is the MJI-HI interferogram reconstruction result without using the improved method, and <span class="html-italic">I<sub>sup−imp</sub></span> is the MJI-HI interferogram reconstruction result using the improved method.</p> Full article ">Figure 16
<p>(<b>a1</b>) Interferogram reconstruction results when the <span class="html-italic">FWHM<sub>tilt</sub></span> of <span class="html-italic">g<sub>FTIS</sub></span> is λ<sub>min</sub>/4; (<b>a2</b>–<b>a5</b>) are its local enlargements; (<b>b1</b>) interferogram reconstruction results when the <span class="html-italic">FWHM<sub>tilt</sub></span> of <span class="html-italic">g<sub>FTIS</sub></span> is λ<sub>min</sub>/2; (<b>b2</b>–<b>b5</b>) are its local enlargements. Here, <span class="html-italic">I</span><sub>0</sub> represents the ideal interferogram, <span class="html-italic">I<sub>sup</sub></span> denotes the MJI-HI reconstructed interferogram under mirror tilt errors, and <span class="html-italic">I<sub>0-tilt</sub></span> is the interferogram obtained using FTIS with the same mirror tilt errors.</p> Full article ">Figure 17
<p>(<b>a</b>) When <span class="html-italic">B</span>(ν) = 1 is used as the input spectrum, the overall situation of the interferogram <span class="html-italic">I</span><sub>2</sub>; (<b>b–e</b>). When the tilt direction angle differences are 0°, 30°, 60°, and 90°, the response function convolution broadening of <span class="html-italic">I</span><sub>2</sub> occurs at ±<span class="html-italic">m</span>Δ<span class="html-italic"><sub>FPI</sub></span>; (<b>b1</b>–<b>b4</b>). When the tilt direction angle difference is <span class="html-italic">φ<sub>FTIS</sub></span> − <span class="html-italic">φ<sub>FPI</sub> =</span> 0°, the convolutional broadening of the response functions of the <span class="html-italic">m</span> = 0, ±1, ±2, ±3 order caused by the combined errors; (<b>c1</b>–<b>c4</b>). When <span class="html-italic">φ<sub>FTIS</sub></span> − <span class="html-italic">φ<sub>FPI</sub> =</span> 30°, the convolutional broadening of the response functions of the <span class="html-italic">m</span> = 0, ±1, ±2, ±3 order; (<b>d1</b>–<b>d4</b>). When <span class="html-italic">φ<sub>FTIS</sub></span> − <span class="html-italic">φ<sub>FPI</sub> =</span> 60°, the convolutional broadening of the response functions of the <span class="html-italic">m</span> = 0, ±1, ±2, ±3 order; (<b>e1</b>–<b>e4</b>). When <span class="html-italic">φ<sub>FTIS</sub></span> − <span class="html-italic">φ<sub>FPI</sub> =</span> 90°, the convolutional broadening of the response functions of the <span class="html-italic">m</span> = 0, ±1, ±2, ±3 order.</p> Full article ">Figure 18
<p>(<b>a1</b>) Interferogram reconstruction result under combined errors when <span class="html-italic">φ<sub>FTIS</sub></span> − <span class="html-italic">φ<sub>FPI</sub></span> = 0°, with (<b>a2</b>–<b>a4</b>) showing local magnifications; (<b>b1</b>) interferogram reconstruction result under combined errors when <span class="html-italic">φ<sub>FTIS</sub></span> − <span class="html-italic">φ<sub>FPI</sub></span> = 30°, with (<b>b2</b>–<b>b4</b>) showing local magnifications; (<b>c1</b>) interferogram reconstruction result under combined errors when <span class="html-italic">φ<sub>FTIS</sub></span> − <span class="html-italic">φ<sub>FPI</sub></span> = 60°, with (<b>c2</b>–<b>c4</b>) showing local magnifications; (<b>d1</b>) interferogram reconstruction result under combined errors when <span class="html-italic">φ<sub>FTIS</sub></span> − <span class="html-italic">φ<sub>FPI</sub></span> = 90°, with (<b>d2</b>–<b>d4</b>) showing local magnifications; here, <span class="html-italic">I</span><sub>0</sub> represents the ideal interferogram, and <span class="html-italic">I<sub>sup</sub></span> denotes the MJI-HI reconstructed interferogram under joint errors.</p> Full article ">Figure 19
<p>(<b>a</b>) When the input spectrum is a cosine function with a periodic frequency of <span class="html-italic">k</span><sub>0</sub>, the overall interferogram <span class="html-italic">I</span><sub>2</sub> is shown, indicating the specific positions of <span class="html-italic">I</span><sub>2</sub>(<span class="html-italic">m</span>Δ<span class="html-italic"><sub>FPI</sub></span>), <span class="html-italic">I</span><sub>2</sub>(<span class="html-italic">m</span>Δ<span class="html-italic"><sub>FPI</sub></span> − <span class="html-italic">k</span><sub>0</sub>), and <span class="html-italic">I</span><sub>2</sub>(<span class="html-italic">m</span>Δ<span class="html-italic"><sub>FPI</sub></span> + <span class="html-italic">k</span><sub>0</sub>) within <span class="html-italic">I</span><sub>2</sub>. For divergence angles of 0°, 0.5°, 1.0°, 1.5°, and 2.0°, the convolution broadening of <span class="html-italic">δ</span> functions of the <span class="html-italic">m</span> = 0, 1, 2, 3 frequency-shift components due to collimation errors is observed at (<b>b1</b>–<b>b4</b>). Δ = <span class="html-italic">m</span>Δ<span class="html-italic"><sub>FPI</sub></span>, at (<b>c1</b>–<b>c4</b>). Δ = <span class="html-italic">m</span>Δ<span class="html-italic"><sub>FP</sub></span> − <span class="html-italic">k</span><sub>0</sub>, at (<b>d1</b>–<b>d4</b>). Δ = <span class="html-italic">m</span>Δ<span class="html-italic"><sub>FPI</sub></span> + <span class="html-italic">k</span><sub>0</sub>.</p> Full article ">Figure 20
<p>(<b>a</b>) Super-resolution spectra obtained using MJI-HI with divergence angles of 0°, 1°, and 2°; (<b>b</b>) local magnification at wavenumber 1.5 × 10<sup>4</sup> cm<sup>−1</sup>. Here, <span class="html-italic">B</span><sub>0</sub> represents the input spectrum, and <span class="html-italic">B<sub>sup-</sub><sub>θ</sub></span> denotes the super-resolution spectra obtained under different divergence angles using MJI-HI.</p> Full article ">Figure 21
<p>(<b>a1</b>–<b>d1</b>) The FPI reflectance spectra generated by curves 1–4 under PVs of 0%, 2%, 6%, and 10%, respectively. (<b>a2</b>–<b>a5</b>) Correspond to curve 1, (<b>b2</b>–<b>b5</b>) to curve 2, (<b>c2</b>–<b>c5</b>) to curve 3, and (<b>d2</b>–<b>d5</b>) to curve 4, illustrating the changes in the response functions of <span class="html-italic">m</span> = 0, 1, 2, 3 frequency-shift components due to reflectance variations.</p> Full article ">Figure 22
<p>Using <span class="html-italic">B</span><sub>0</sub> = 1 as the input spectrum, the super-resolution spectral results obtained under reflectance variation (<b>a</b>) curve 1, (<b>b</b>) curve 2, (<b>c</b>) curve 3, and (<b>d</b>) curve 4 with PV = 2%, 6%, and 10% are shown. Here, <span class="html-italic">B<sub>sup</sub></span><sub>-PV</sub> represents the super-resolution spectra obtained using MJI-HI under different FPI reflectance error levels.</p> Full article ">
15 pages, 5492 KiB  
Article
Classification of OCT Images of the Human Eye Using Mobile Devices
by Agnieszka Stankiewicz, Tomasz Marciniak, Nina Budna, Róża Chwałek and Marcin Dziedzic
Appl. Sci. 2025, 15(6), 2937; https://doi.org/10.3390/app15062937 (registering DOI) - 8 Mar 2025
Abstract
The aim of this study was to develop a mobile application for Android devices dedicated to the classification of pathological changes in human eye optical coherence tomography (OCT) B-scans. The classification process is conducted using convolutional neural networks (CNNs). Six models were trained [...] Read more.
The aim of this study was to develop a mobile application for Android devices dedicated to the classification of pathological changes in human eye optical coherence tomography (OCT) B-scans. The classification process is conducted using convolutional neural networks (CNNs). Six models were trained during the study: a simple convolutional neural network with three convolutional layers, VGG16, InceptionV3, Xception, Joint Attention Network + MobileNetV2 and OpticNet-71. All of these models were converted to TensorFlow Lite format to implement them into a mobile application. For this purpose, three models with the best parameters were chosen, taking accuracy, precision, recall, F1-score and confusion matrix into consideration. The Android application designed for the classification of OCT images was developed using the Kotlin programming language within the Android Studio integrated development environment. With the application, classification can be performed on an image chosen from the user’s files or an image acquired using the photo-taking function. The results of the classification are displayed for three neural networks, along with the respective classification times for each neural network and the associated image undergoing the classification task. The mobile application has been tested using various smartphones. The testing phase included an evaluation of image classification times and score accuracy, considering factors such as image acquisition method, i.e., camera or gallery. Full article
(This article belongs to the Special Issue Machine Learning for Object Detection and Scene Description in Images and Videos)
16 pages, 1884 KiB  
Article
Degradation and Ecotoxicity Mitigation of Perfluorooctane Sulfonate by Aeration-Assisted Cold Plasma
by Sengbin Oh, Joo-Youn Nam, Youngpyo Hong, Tae-Hun Lee, Jae-Cheol Lee and Hyun-Woo Kim
Appl. Sci. 2025, 15(6), 2936; https://doi.org/10.3390/app15062936 (registering DOI) - 8 Mar 2025
Abstract
Various advanced oxidation processes have been used to degrade perfluorooctane sulfonate (PFOS), one of the persistent organic pollutants that dissolves in aquatic ecosystems, but these processes suffer from inherent limitations. This study proposes aeration-assisted cold plasma (CP) technology as an alternative. PFOS removal [...] Read more.
Various advanced oxidation processes have been used to degrade perfluorooctane sulfonate (PFOS), one of the persistent organic pollutants that dissolves in aquatic ecosystems, but these processes suffer from inherent limitations. This study proposes aeration-assisted cold plasma (CP) technology as an alternative. PFOS removal via CP treatment reached 62.5% after 1 h of exposure, with a degradation rate constant of 3.1 h−1. The detection of sulfate (SO42−) in the solution provides evidence of effective PFOS degradation. The close agreement between the measured and estimated fluoride concentrations further confirms mass balance after degradation. Acute toxicity tests indicate that PFOS degradation may initially increase the acute toxicity, possibly due to the formation of degradation by-products. However, this increased toxicity can be mitigated through additional exposure to the reactive species generated by CP. Furthermore, investigations into the energy per order of CP and the quantification of hydroxyl radicals support its operational effectiveness. This study confirms that aeration-assisted CP has the potential to serve as a viable treatment option for mitigating the environmental threats posed by PFOS. Full article
(This article belongs to the Special Issue New Approaches to Water Treatment: Challenges and Trends)
►▼ Show Figures

Figure 1

Figure 1
<p>Schematic of the aeration-assisted CP system: (a) aeration pump; (b) CP generator; (c) air with reactive oxidizing species; (d) flowrate meter; (e) ceramic sparger; (f) magnetic bar; (g) magnetic stirrer; and (h) sampling port.</p> Full article ">Figure 2
<p>Overview of PFOS degradation and its associated dynamics in the aeration-assisted CP system. (<b>a</b>) PFOS concentration dynamics and regressed degradation line over time. (<b>b</b>) Variation in the TOC concentration in CP over time. (<b>c</b>) Detection of fluoride and sulfate produced during PFOS degradation in the system. (<b>d</b>) Mass balance of fluoride and sulfate, comparing experimental data with theoretical calculations, during PFOS degradation.</p> Full article ">Figure 3
<p>Variations in the acute toxicity of PFOS and its degradation intermediates in the aeration-assisted CP system.</p> Full article ">Figure 4
<p>Detection of formaldehyde and quantification of hydroxyl radicals. The left arrow denotes formaldehyde concentration, and the right arrow indicates hydroxyl radicals.</p> Full article ">Figure 5
<p>CP-enabled instantaneous formation rates of hydroxyl radical.</p> Full article ">
5 pages, 147 KiB  
Editorial
Internet of Things (IoT) Technologies in Cybersecurity: Challenges and Opportunities
by Grzegorz Kołaczek
Appl. Sci. 2025, 15(6), 2935; https://doi.org/10.3390/app15062935 (registering DOI) - 8 Mar 2025
Abstract
The continuous development and increasing availability of Internet of Things (IoT)[...] Full article
(This article belongs to the Special Issue Internet of Things (IoT) Technologies in Cybersecurity: Challenges and Opportunities)
23 pages, 2519 KiB  
Article
Application of Machine Learning for Bulbous Bow Optimization Design and Ship Resistance Prediction
by Yujie Shen, Shuxia Ye, Yongwei Zhang, Liang Qi, Qian Jiang, Liwen Cai and Bo Jiang
Appl. Sci. 2025, 15(6), 2934; https://doi.org/10.3390/app15062934 (registering DOI) - 8 Mar 2025
Abstract
Resistance is a key index of a ship’s hydrodynamic performance, and studying the design of the bulbous bow is an important method to reduce ship resistance. Based on the ship resistance sample data obtained from computational fluid dynamics (CFD) simulation, this study uses [...] Read more.
Resistance is a key index of a ship’s hydrodynamic performance, and studying the design of the bulbous bow is an important method to reduce ship resistance. Based on the ship resistance sample data obtained from computational fluid dynamics (CFD) simulation, this study uses a machine learning method to realize the fast prediction of ship resistance corresponding to different bulbous bows. To solve the problem of insufficient accuracy in the single surrogate model, this study proposes a CBR surrogate model that integrates convolutional neural networks with backpropagation and radial basis function models. The coordinates of the control points of the NURBS surface at the bulbous bow are taken as the design variables. Then, a convergence factor is introduced to balance the global and local search abilities of the whale algorithm to improve the convergence speed. The sample space is then iteratively searched using the improved whale algorithm. The results show that the mean absolute error and root mean square error of the CBR model are better than those of the BP and RBF models. The accuracy of the model prediction is significantly improved. The optimized bulbous bow design minimizes the ship resistance, which is reduced by 4.95% compared with the initial ship model. This study provides a reliable and efficient machine learning method for ship resistance prediction. Full article
(This article belongs to the Special Issue Computational Fluid Dynamics-Based for Ship Hydrodynamics Applications)
22 pages, 2439 KiB  
Article
Evaluation of UV-Curable Solid Rocket Propellants’ Properties for Advanced 3D Printing Technologies
by Filippo Masseni, Giacomo Tetti, Alessandra Zumbo, Camilla Noé, Giovanni Polizzi, Leonardo Stumpo, Andrea Ferrero and Dario Pastrone
Appl. Sci. 2025, 15(6), 2933; https://doi.org/10.3390/app15062933 (registering DOI) - 8 Mar 2025
Viewed by 46
Abstract
Challenges in the traditional cast-and-cure manufacturing of composite solid propellants, such as the use of mandrels and the toxicity of curing agents, are being addressed through new propellant formulations and additive manufacturing techniques. Within this framework, this study aimed to investigate the properties [...] Read more.
Challenges in the traditional cast-and-cure manufacturing of composite solid propellants, such as the use of mandrels and the toxicity of curing agents, are being addressed through new propellant formulations and additive manufacturing techniques. Within this framework, this study aimed to investigate the properties of UV-curable composite solid rocket propellants, focusing on their compatibility with advanced 3D printing technologies. Polybutadiene-based propellants incorporating a specific photoinitiator were examined. Key rheological properties, including the pseudoplasticity and pot-life, were assessed to evaluate the material’s behavior during the printing process. Furthermore, photopolymerization tests were performed using a customized delta illuminator to evaluate the conversion efficiency under UVA and UVC light sources. Concurrently, a modular Cartesian 3D printer was developed and preliminary tests were performed. Rheological tests also revealed a flow index n of 0.32 at 60 °C and 0.46 at 80 °C, indicating significant pseudoplastic behavior. The pot-life tests showed that the viscosity of the propellant reached the upper limit of 106 cP more quickly at higher temperatures, indicating a shorter time range of printability. UVA irradiation resulted in a polymerization conversion rate of about 90%, while UVC exposure did not significantly enhance the conversion rate beyond this value. Finally, the 3D printing tests confirmed the feasibility of producing solid propellant, though challenges related to material segregation and the extrusion consistency were observed. Material separation resulted in a significant impact on the printability, causing underextrusion and nozzle clogging, particularly with smaller nozzle diameters and higher extrusion pressures. Overall, this research represents a significant step forward in the development of UV-curable propellants for additive manufacturing, building on previous advancements by the research group. It demonstrates tangible progress in addressing key challenges such as the printability, material performance, and curing efficiency, while also highlighting areas requiring further refinement. These findings underscore the continuous evolution of this technology toward higher readiness levels, paving the way for its broader application in composite solid propellant manufacturing. Full article
(This article belongs to the Special Issue Recent Advances in 3D Printing and Additive Manufacturing Technology)
►▼ Show Figures

Figure 1

Figure 1
<p>Rotary viscometer (<b>a</b>) and isothermal test bench (<b>b</b>). Not to scale. (<b>a</b>) Rotary viscometer SAVISC 152-2 and standard LV spindles (purchased from SAMA tools, Viareggio, Italy). (<b>b</b>) (1) Submerged pump; (2) water supply line; (3) copper coil; (4) water return line; (5) support; (6) viscometer without spindle; (7) hotplate.</p> Full article ">Figure 2
<p>Electronic scheme of the PID circuit.</p> Full article ">Figure 3
<p>UVA-UVC illuminator assembly (<b>a</b>) and views of the LED platform (<b>b</b>,<b>c</b>). Not to scale. (<b>a</b>) (1) Mechanism moving LED platform; (2) print bed. (<b>b</b>) (1) UVA LEDs; (2) UVC LEDs; (3) metal plate (heat sink); (4) PLA plate (cinematism interface); (5) clamping system. (<b>c</b>) Layout of UVA and UVC LEDs.</p> Full article ">Figure 4
<p>On the left, the predicted irradiance; on the right, the measured irradiance of the UVA LEDs.</p> Full article ">Figure 5
<p>Extruder (<b>a</b>) and Cartesian 3D printer (<b>b</b>). (<b>a</b>) (1) Air piston seal; (2) air seal bushing; (3) piston body; (4) piston rod; (5) wear ring; (6) nozzle; (7) nozzle plate; (8) M5 screws; (9) bumper. Not to scale. (<b>b</b>) Prototype of the Cartesian printer developed in collaboration with the company Microdigit elettronica (Cazzago San Martino BS, Italy).</p> Full article ">Figure 6
<p>Specimens freshly extracted from toluene (<b>left</b>) and dried for three days, completely free of toluene (<b>right</b>).</p> Full article ">Figure 7
<p>Viscosity curves (blue lines) at two different temperatures as function of shear rate, and ratio of two curves (red line).</p> Full article ">Figure 8
<p>Pot-life tests at 60 °C and 80 °C with speed rotation of 0.6 RPM and maximum viscosity of <math display="inline"><semantics> <msup> <mn>10</mn> <mn>6</mn> </msup> </semantics></math> cP (mPa·s).</p> Full article ">Figure 9
<p>Slurry extracted from the extruder after a printing test.</p> Full article ">Figure 10
<p>Deterioration in print quality with increase in moisture absorbed by AS salts, resulting in underextrusion and severe nozzle clogging. (<b>a</b>) Monolayer printing with dry AS salts. Samples were 50 mm × 50 mm approx. (<b>b</b>) Multilayer printing with dry AS salts. (<b>c</b>) Printing with wet salts.</p> Full article ">
11 pages, 7935 KiB  
Article
Characterization of Invar Syntactic Foams Obtained by Spark Plasma Sintering
by Argentina Niculina Sechel, Călin-Virgiliu Prică, Traian Florin Marinca, Florin Popa, Loredana-Maria Baglaevschi, Gyorgy Thalmaier and Ioan Vida-Simiti
Appl. Sci. 2025, 15(6), 2932; https://doi.org/10.3390/app15062932 (registering DOI) - 8 Mar 2025
Viewed by 18
Abstract
This study presents the synthesis of sintered composite foams based on the Invar alloy (64Fe-36Ni), using hollow spherical particles from a nickel superalloy (NiCrSiB) in order to generate porosity. The Invar powder was obtained by mechanical alloying (MA), and the NiCrSiB hollow spherical [...] Read more.
This study presents the synthesis of sintered composite foams based on the Invar alloy (64Fe-36Ni), using hollow spherical particles from a nickel superalloy (NiCrSiB) in order to generate porosity. The Invar powder was obtained by mechanical alloying (MA), and the NiCrSiB hollow spherical particles were incorporated into the composite at 20 vol %. The sintering was realized using the spark plasma sintering (SPS) process in an argon atmosphere at 600 °C and 5 MPa, with 10 s holding time. The porous structures were structurally characterized by optical microscopy (OM), scanning electron microscopy (SEM) and X-ray diffraction (XRD). The coefficient of linear thermal expansion (CTE) of the Invar/NiCrSiB syntactic foams was found to be 2.52 × 10−6 °C−1 in the 25–150 °C temperature range and 19.68 × 10−6 °C−1 in the 150–400 °C range. Full article
►▼ Show Figures

Figure 1

Figure 1
<p>Optical images of Invar powder (<b>a</b>) and NiCrSiB superalloy (<b>b</b>).</p> Full article ">Figure 2
<p>Optical images of the wall thickness (g) of some NiCrSiB superalloy particles; (<b>a</b>) g ≅ 6 μm, (<b>b</b>) g ≅ 26 μm, (<b>c</b>) g ≅ 54 μm.</p> Full article ">Figure 3
<p>Photo images of spark plasma sintered Invar/20%NiCrSiB syntactic foam sample.</p> Full article ">Figure 4
<p>SEM images of Invar/20%NiCrSiB composite foam at different magnifications: 35× (<b>a</b>), 100× (<b>b</b>), 1000× (<b>c</b>) and 5000× (<b>d</b>).</p> Full article ">Figure 5
<p>EDX maps of element distributions of Invar/20%NiCrSiB syntactic foam – mixed elements map distribution (<b>a</b>), map of the Fe (<b>b</b>), map of the Ni (<b>c</b>), map of the Cr (<b>d</b>), map of the Si (<b>e</b>), map of the B (<b>f</b>) and map of the C (<b>g</b>).</p> Full article ">Figure 5 Cont.
<p>EDX maps of element distributions of Invar/20%NiCrSiB syntactic foam – mixed elements map distribution (<b>a</b>), map of the Fe (<b>b</b>), map of the Ni (<b>c</b>), map of the Cr (<b>d</b>), map of the Si (<b>e</b>), map of the B (<b>f</b>) and map of the C (<b>g</b>).</p> Full article ">Figure 6
<p>SEM image and EDX line scan of Invar/20%NiCrSiB sample.</p> Full article ">Figure 7
<p>XRD diffraction patterns of Invar 16 h milled powders (<b>a</b>), NiCrSiB hollow particles (<b>b</b>) and Invar/20%NiCrSBi spark plasma sintered composite foam (<b>c</b>).</p> Full article ">Figure 8
<p>Elongation variation (Δl) as a function of temperature for syntactic foam Invar/20% NiCrSiB and for Invar [<a href="#B17-applsci-15-02932" class="html-bibr">17</a>].</p> Full article ">
20 pages, 6467 KiB  
Article
A Lightweight TA-YOLOv8 Method for the Spot Weld Surface Anomaly Detection of Body in White
by Weijie Liu, Miao Jia, Shuo Zhang, Siyu Zhu, Jin Qi and Jie Hu
Appl. Sci. 2025, 15(6), 2931; https://doi.org/10.3390/app15062931 (registering DOI) - 8 Mar 2025
Viewed by 12
Abstract
The deep learning architecture YOLO (You Only Look Once) has demonstrated its superior visual detection performance in various computer vision tasks and has been widely applied in the field of automatic surface defect detection. In this paper, we propose a lightweight YOLOv8-based method [...] Read more.
The deep learning architecture YOLO (You Only Look Once) has demonstrated its superior visual detection performance in various computer vision tasks and has been widely applied in the field of automatic surface defect detection. In this paper, we propose a lightweight YOLOv8-based method for the quality inspection of car body welding spots. We developed a TA-YOLOv8 network structure which has an improved Task-Aligned (TA) head detection, designed to handle a small sample size, imbalanced positive and negative samples, and high-noise characteristics of Body-in-White welding spot data. By learning with fewer parameters, the model achieves more efficient and accurate classification. Additionally, our algorithm framework can perform anomaly segmentation and classification on our open-world raw datasets obtained from actual production environments. The experimental results show that the lightweight module improves the processing speed by an average of 2.8%, with increases in detection the mAP@50-95 and recall rate of 1.35% and 0.1226, respectively. Full article
(This article belongs to the Special Issue Motion Control for Robots and Automation)
►▼ Show Figures

Figure 1

Figure 1
<p>Architecture of YOLOv8 model. The different color parts of the input batches represent different image data. The different color parts of the architecture represent different function modules.</p> Full article ">Figure 2
<p>Improved backbone network of our architecture. On the left is a schematic diagram of the backbone network process for detecting spot weld images, while the right side shows the corresponding structure layer parameters and related information. The different color parts are the same as <a href="#applsci-15-02931-f001" class="html-fig">Figure 1</a>.</p> Full article ">Figure 3
<p>Proposed Multiple Cross-Layer FPN (MC-FPN) network. The different color dotted lines represent multiple cross layers, with P<sub>2</sub>-P<sub>5</sub> being simplified representations of the intermediate connection layers.</p> Full article ">Figure 4
<p>Task-Aligned head structure: to learn extensive task-interactive features from multiple convolutional layers.</p> Full article ">Figure 5
<p>Welding spots sample images and annotated data in Body-in-White production lines. (<b>a</b>) shows the samples we collected in the production lines, while (<b>b</b>) shows the pretraining dataset and the labels (yellow squares).</p> Full article ">Figure 5 Cont.
<p>Welding spots sample images and annotated data in Body-in-White production lines. (<b>a</b>) shows the samples we collected in the production lines, while (<b>b</b>) shows the pretraining dataset and the labels (yellow squares).</p> Full article ">Figure 6
<p>Performance comparison with typical object detection algorithms on test set.</p> Full article ">Figure 7
<p>Some results of WSDDM and comparison between small welding spot detection models. We use green to represent the detected weld spots are normal, and red to represent the detected weld spots have defects or abnormalities.</p> Full article ">Figure 8
<p>The weld spot dataset obtained from image segmentation using the WSDDM.</p> Full article ">Figure 9
<p>Data augmentation and labeling.</p> Full article ">Figure 10
<p>Visualization and validation sample results for model testing. The model effectively captures the location of welding defects through highlighted (green) regions.</p> Full article ">Figure 11
<p>Validation sample results for model generalization ability.</p> Full article ">Figure 12
<p>Experimental pipeline and integrated detection systems.</p> Full article ">
20 pages, 7214 KiB  
Article
Effect of Heterojunction Characteristics and Deep Electronic Levels on the Performance of (Cd,Zn)S/Sb2Se3 Solar Cells
by Alessio Bosio, Stefano Pasini, Donato Spoltore, Gianluca Foti, Antonella Parisini, Maura Pavesi, Samaneh Shapouri, Ildikó Cora, Zsolt Fogarassy and Roberto Fornari
Appl. Sci. 2025, 15(6), 2930; https://doi.org/10.3390/app15062930 (registering DOI) - 8 Mar 2025
Viewed by 6
Abstract
Antimony selenide (Sb2Se3) is an Earth-abundant and non-toxic material that stands out as a promising absorber for the fabrication of thin film solar cells. Despite significant advancements in recent years, all the devices reported in the literature exhibit open-circuit [...] Read more.
Antimony selenide (Sb2Se3) is an Earth-abundant and non-toxic material that stands out as a promising absorber for the fabrication of thin film solar cells. Despite significant advancements in recent years, all the devices reported in the literature exhibit open-circuit voltages well below the theoretical value. Identifying the factors contributing to this low voltage is an essential step for increasing the efficiency beyond the recently attained 10% milestone and moving closer to the theoretical limit. In this paper, we present the results of an in-depth analysis of a Sb2Se3 solar cell in the common superstrate configuration. By making use of current density–voltage characteristic as a function of both temperature and wavelength, capacitance–voltage measurements, and admittance spectroscopy, we ascribe the low open-circuit voltage to the presence of a potential barrier within the absorber material near the junction interface Furthermore, it was observed that the junction behavior in the dark and under illumination changes, which is compatible with the presence of deep electronic levels connected with intrinsic point defects. Full article
(This article belongs to the Special Issue Advanced Solar Energy Materials: Methods and Applications)
►▼ Show Figures

Figure 1

Figure 1
<p>(<b>a</b>) Non-scaled sketch of the Sb<sub>2</sub>Se<sub>3</sub>-based solar cell and cross-sectional bright field (BF) TEM image of the solar cell; (<b>b</b>) current density–voltage characteristics in dark (black) and under solar simulator illumination with 700 W/m<sup>2</sup> AM1.5G filtered light (red); the magnification highlights the crossover point V<sub>X</sub>.</p> Full article ">Figure 2
<p>(<b>a</b>) Linear fit (in red) of dV/dJ data for the determination of R<sub>S</sub>. (<b>b</b>) Linear fit (in red) of dJ/dV data for the determination of R<sub>sh</sub>. The curves were derived from J-V curves measured under solar simulator illumination of 700 W/m<sup>2</sup> AM1.5G at a temperature of 298 K.</p> Full article ">Figure 3
<p>Tauc plot of (<b>a</b>) (Cd,Zn)S and (<b>b</b>) CdS thin films. Red lines represent the linear fitting with the dotted part being the extrapolation to the energy axis.</p> Full article ">Figure 4
<p>(<b>a</b>) Cross-sectional HAADF TEM image with elemental EDX maps for (<b>c</b>) Zn, (<b>d</b>) O, (<b>e</b>) Se, (<b>f</b>) Sb, (<b>g</b>) Cd, (<b>h</b>) S. The white box on the HAADF image shows the area from where the (<b>b</b>) integrated intensity profile with the calculated at% was taken. The diffusion of Se and Sb atoms into the sulfide layer is observed.</p> Full article ">Figure 5
<p>HRTEM image from the ZnO/CdS/(Cd,Zn)S/Sb<sub>2</sub>Se<sub>3</sub> interfaces. The FFT from the signed area is indexed as orthorhombic (s.g. Pnma) Sb<sub>2</sub>Se<sub>3</sub> from the [010] projection. In the right-upper corner, the HAADF image with the projected atomic structure of the Sb<sub>2</sub>Se<sub>3</sub> is shown. CdS and the (Cd,Zn)S layer are polycrystalline.</p> Full article ">Figure 6
<p>(<b>a</b>) The solar cell model highlights the second barrier diode positioned in anti-series with respect to the primary barrier diode. R<sub>0</sub> and D<sub>B</sub> are the contact resistance and threshold barrier of the anti-series diode, respectively. (<b>b</b>) Non-scaled sketch for the proposed model of the (Cd,Zn)S/Sb<sub>2</sub>Se<sub>3</sub> junction highlighting the intermixing layer between (Cd,Zn)S and Sb2Se3, the penetration depth in which 63% of the incoming light is absorbed, and the depletion layer.</p> Full article ">Figure 7
<p>Schematic representation of the (Cd,Zn)S/Sb<sub>2</sub>Se<sub>3</sub> interface band profiles.</p> Full article ">Figure 8
<p>The data points on both graphs were obtained by illuminating the sample with a LQX 1000 lamp. (<b>a</b>) Arrhenius plot of ln(T·R<sub>S</sub>) (in K·Ωcm<sup>2</sup>) as a function of 1/T. Each experimental value was derived from the J-V measurements taken at temperatures ranging from 196 K to 353 K without optical filters. (<b>b</b>) The height of the additional barrier as a function of distance from the junction: each data point was obtained using bandpass filters centered at 650 nm, 750 nm, 850 nm, and 950 nm, and all were normalized to the same number of absorbed photons.</p> Full article ">Figure 9
<p>(<b>a</b>) Measurements of V<sub>x</sub> and V<sub>oc</sub> were conducted at different wavelengths, with all potential values standardized to an equal number of absorbed photons. The varied wavelengths were selected using passband filters centered at 650 nm, 750 nm, 850 nm, and 950 nm. (<b>b</b>) ΔV = V<sub>x</sub> − V<sub>oc</sub> as a function of the absorbed light wavelength.</p> Full article ">Figure 10
<p>Plot of ln(C) (C unit: F) against ln(V + V<sub>bi</sub>) (V + V<sub>bi</sub> unit: V) under light and dark conditions. The linear fit of the experimental data is made in the reverse bias region for V &gt; <math display="inline"><semantics> <mrow> <msub> <mi>V</mi> <mrow> <mi>b</mi> <mi>i</mi> </mrow> </msub> </mrow> </semantics></math>.</p> Full article ">Figure 11
<p>Plot of 1/C<sup>3</sup> as a function of the bias voltage: (<b>a</b>) under light, (<b>b</b>) in dark conditions.</p> Full article ">Figure 12
<p>Plotting log(J) (unit: mA/cm<sup>2</sup>) against log(V) (unit: V) under dark conditions reveals three distinct regions, delineated by dotted vertical lines. Linear fittings of the data in the respective region are shown in red, blue and green for the Ohmic, TFL and Child region respectively.</p> Full article ">Figure 13
<p>Admittance spectroscopy measurements (<b>a</b>) under light (<b>b</b>) in the dark.</p> Full article ">Figure 14
<p><math display="inline"><semantics> <mrow> <mi>l</mi> <mi>n</mi> <mrow> <mo>(</mo> <mrow> <msub> <mi>ω</mi> <mn>0</mn> </msub> <mo>/</mo> <msup> <mi>T</mi> <mn>2</mn> </msup> </mrow> <mo>)</mo> </mrow> </mrow> </semantics></math> (unit: <math display="inline"><semantics> <mrow> <msup> <mi mathvariant="normal">s</mi> <mrow> <mo>−</mo> <mn>1</mn> </mrow> </msup> <mi>·</mi> <msup> <mi mathvariant="normal">K</mi> <mrow> <mo>−</mo> <mn>2</mn> </mrow> </msup> </mrow> </semantics></math>) as a function of 1/KT, illustrating two distinct activation energies observed (<b>a</b>) under light and (<b>b</b>) in the dark.</p> Full article ">
12 pages, 2793 KiB  
Article
Numerical Modeling of Bowel Sound Propagation: Impact of Abdominal Tissue Properties
by Xingyu Deng, Yazhen Xu and Yuanwen Zou
Appl. Sci. 2025, 15(6), 2929; https://doi.org/10.3390/app15062929 (registering DOI) - 8 Mar 2025
Viewed by 5
Abstract
Bowel sounds, produced by intestinal peristalsis, are essential for diagnosing gastrointestinal disorders. However, acquiring and analyzing bowel sounds is challenging due to their unpredictable nature and individual variability. Biological tissues can affect bowel sounds during propagation, resulting in varying degrees of signal attenuation [...] Read more.
Bowel sounds, produced by intestinal peristalsis, are essential for diagnosing gastrointestinal disorders. However, acquiring and analyzing bowel sounds is challenging due to their unpredictable nature and individual variability. Biological tissues can affect bowel sounds during propagation, resulting in varying degrees of signal attenuation between the sound source and the transducer. This study aims to develop a numerical model of bowel sound propagation in the abdominal cavity, focusing on the impact of different biological layers on signal attenuation. Validation of the model demonstrated strong consistency between simulated and actual bowel sound signals, confirming the model’s accuracy and reliability. The model accounted for adipose tissue thickness, ranging from 5 to 20 mm across individuals, while muscle and skin thicknesses remained constant. Results indicated that signal attenuation increases with both the propagation distance and adipose tissue thickness. These findings provide insights into how tissue layers influence bowel sound propagation, offering a theoretical foundation for developing personalized and precise monitoring devices. Full article
(This article belongs to the Section Applied Physics General)
►▼ Show Figures

Figure 1

Figure 1
<p>An IWC with its bounding envelope.</p> Full article ">Figure 2
<p>Abdominal modeling process. (<b>a</b>) Abdominal anatomy in the standard anatomical position. (<b>b</b>) Tissue layers of the human abdomen. (<b>c</b>) Overall geometric model of the abdomen.</p> Full article ">Figure 3
<p>Analog IWC vs. real IWC.</p> Full article ">Figure 4
<p>IWC on tissues surfaces.</p> Full article ">Figure 5
<p>IWC on fat layer surface at varying thickness.</p> Full article ">Figure 6
<p>Absolute attenuation as a function of fat thickness.</p> Full article ">
27 pages, 899 KiB  
Article
Comparative Analysis of AlexNet, ResNet-50, and VGG-19 Performance for Automated Feature Recognition in Pedestrian Crash Diagrams
by Baraah Qawasmeh, Jun-Seok Oh and Valerian Kwigizile
Appl. Sci. 2025, 15(6), 2928; https://doi.org/10.3390/app15062928 (registering DOI) - 8 Mar 2025
Viewed by 5
Abstract
Pedestrians, as the most vulnerable road users in traffic crashes, prompt transportation researchers and urban planners to prioritize pedestrian safety due to the elevated risk and growing incidence of injuries and fatalities. Thorough pedestrian crash data are indispensable for safety research, as the [...] Read more.
Pedestrians, as the most vulnerable road users in traffic crashes, prompt transportation researchers and urban planners to prioritize pedestrian safety due to the elevated risk and growing incidence of injuries and fatalities. Thorough pedestrian crash data are indispensable for safety research, as the most detailed descriptions of crash scenes and pedestrian actions are typically found in crash narratives and diagrams. However, extracting and analyzing this information from police crash reports poses significant challenges. This study tackles these issues by introducing innovative image-processing techniques to analyze crash diagrams. By employing cutting-edge technological methods, the research aims to uncover and extract hidden features from pedestrian crash data in Michigan, thereby enhancing the understanding and prevention of such incidents. This study evaluates the effectiveness of three Convolutional Neural Network (CNN) architectures—VGG-19, AlexNet, and ResNet-50—in classifying multiple hidden features in pedestrian crash diagrams. These features include intersection type (three-leg or four-leg), road type (divided or undivided), the presence of marked crosswalk (yes or no), intersection angle (skewed or unskewed), the presence of Michigan left turn (yes or no), and the presence of nearby residentials (yes or no). The research utilizes the 2020–2023 Michigan UD-10 pedestrian crash reports, comprising 5437 pedestrian crash diagrams for large urbanized areas and 609 for rural areas. The CNNs underwent comprehensive evaluation using various metrics, including accuracy and F1-score, to assess their capacity for reliably classifying multiple pedestrian crash features. The results reveal that AlexNet consistently surpasses other models, attaining the highest accuracy and F1-score. This highlights the critical importance of choosing the appropriate architecture for crash diagram analysis, particularly in the context of pedestrian safety. These outcomes are critical for minimizing errors in image classification, especially in transportation safety studies. In addition to evaluating model performance, computational efficiency was also considered. In this regard, AlexNet emerged as the most efficient model. This understanding is precious in situations where there are limitations on computing resources. This study contributes novel insights to pedestrian safety research by leveraging image processing technology, and highlights CNNs’ potential use in detecting concealed pedestrian crash patterns. The results lay the groundwork for future research, and offer promise in supporting safety initiatives and facilitating countermeasures’ development for researchers, planners, engineers, and agencies. Full article
(This article belongs to the Special Issue Traffic Safety Measures and Assessment)
►▼ Show Figures

Figure 1

Figure 1
<p>Methodological framework for pedestrian crash diagram classification using CNNs.</p> Full article ">Figure 2
<p>(<b>a</b>) Mean training loss of all CNN models for all features’ classifications. (<b>b</b>) Mean validation loss of all CNN models for all features classifications.</p> Full article ">Figure 3
<p>Computational time of all CNN models for all features’ classifications over 50 epochs.</p> Full article ">
11 pages, 448 KiB  
Article
Accuracy and Reliability of Digital Dental Models Obtained by Intraoral Scans Compared with Plaster Models
by Vo Huyen Bao Tran, Tran Hung Lam, Truong Nhut Khue, Tran Ngoc Quang Phi and Hoang Viet
Appl. Sci. 2025, 15(6), 2927; https://doi.org/10.3390/app15062927 (registering DOI) - 8 Mar 2025
Viewed by 9
Abstract
Introduction: In orthodontics, determining the parameters of tooth size and dental arch and conducting Bolton analysis is crucial for diagnosis, treatment planning, and patient outcomes. This study evaluates the accuracy and reliability of measuring dental-arch dimensions on digital models created using the CEREC [...] Read more.
Introduction: In orthodontics, determining the parameters of tooth size and dental arch and conducting Bolton analysis is crucial for diagnosis, treatment planning, and patient outcomes. This study evaluates the accuracy and reliability of measuring dental-arch dimensions on digital models created using the CEREC Primescan intraoral scanner, compared to measurements taken from plaster models. Methods: The study included two types of dental models (plaster and intraoral scan) from sixty-three subjects. Impressions were taken to create plaster models, and the subjects’ mouths were scanned with the CEREC Primescan system (Dentsply Sirona, Charlotte, NC) to create digital models. Intra-arch measurements included tooth heights and widths, overjet, and overbite. The arch width and depth were examined at the first permanent upper or lower molar. The paired t-test and Bland–Altman plot were used to determine the accuracy, while intra-rater and inter-rater correlation coefficient values were calculated to assess the reliability of measurements from the intraoral scan compared to those from the plaster model. Results: For tooth heights, there was a statistically significant difference in only one measurement (tooth 34) between the plaster and digital models, with an average difference of 0.01 mm. For tooth widths, there was a statistically significant difference in only one measurement (tooth 15) with an average difference of 0.03 mm. The Bland–Altman plots of almost all of measurements of tooth heights and widths showed that differences between the two models were within the limits of agreement. The inter- and intra-rater correlation coefficient values for measurements on the digital model were found to be statistically insignificant. Conclusion: Measuring dental dimensions on digital models obtained through the Primescan intraoral digital system yielded similar results to those obtained from plaster models and showed excellent reliability, indicating its potential application in clinical practice. Full article
(This article belongs to the Special Issue State-of-the-Art Operative Dentistry)
►▼ Show Figures

Figure 1

Figure 1
<p>Bland–Altman analysis of the measurements of tooth heights and widths between the plaster and digital models.</p> Full article ">
24 pages, 2258 KiB  
Article
Automated Tunnel Point Cloud Segmentation and Extraction Method
by Zhe Wang, Zhenyi Zhu, Yong Wu, Qihao Hong, Donglai Jiang, Jinbo Fu and Sifa Xu
Appl. Sci. 2025, 15(6), 2926; https://doi.org/10.3390/app15062926 - 7 Mar 2025
Viewed by 213
Abstract
To address the issue of inaccurate tunnel segmentation caused by solely relying on point cloud coordinates, this paper proposes two algorithms, GuSAC and TMatch, along with a ring-based cross-section extraction method to achieve high-precision tunnel lining segmentation and cross-section extraction. GuSAC, based on [...] Read more.
To address the issue of inaccurate tunnel segmentation caused by solely relying on point cloud coordinates, this paper proposes two algorithms, GuSAC and TMatch, along with a ring-based cross-section extraction method to achieve high-precision tunnel lining segmentation and cross-section extraction. GuSAC, based on the RANSAC algorithm, introduces a minimum spanning tree to reconstruct the topological structure of the tunnel design axis. By using a sliding window, it effectively distinguishes between curved and straight sections of long tunnels while removing non-tunnel structural point clouds with normal vectors, thereby enhancing the lining boundary features and significantly improving the automation level of tunnel processing. At the same time, the TMatch algorithm, which combines cluster analysis and Gaussian Mixture Models (GMMs), achieves accurate segmentation of tunnel rings and inner ring areas and further determines the tunnel cross-section position based on this segmentation result to complete the cross-section extraction. Experimental results show that the proposed method achieves a segmentation accuracy of up to 95% on a standard tunnel point cloud dataset. Compared with traditional centerline extraction methods, the proposed cross-section extraction method does not require complex parameter settings, provides more stable positioning, and demonstrates high practicality and robustness. Full article
12 pages, 3588 KiB  
Article
Sensitivity Analysis of Numerical Coherency Model for Rock Sites
by Dongyeon Lee, Yonghee Lee, Hak-Sung Kim, Jeong-Seon Park and Duhee Park
Appl. Sci. 2025, 15(6), 2925; https://doi.org/10.3390/app15062925 - 7 Mar 2025
Viewed by 231
Abstract
Characterization of ground motion incoherency can significantly reduce the seismic load imposed on large scale infrastructures. Because of difficulties in developing an empirical coherency function from a site-specific dense array, it is seldom used in practice. A number of studies used numerical simulations [...] Read more.
Characterization of ground motion incoherency can significantly reduce the seismic load imposed on large scale infrastructures. Because of difficulties in developing an empirical coherency function from a site-specific dense array, it is seldom used in practice. A number of studies used numerical simulations to develop generic coherency models. However, they have only been developed for idealized profiles. A comprehensive parametric study evaluating the effect of various parameters influencing the calculated coherency function has not yet been performed. We utilized the measured shear wave velocity (Vs) profile at Pinyon Flat, located in California, to perform a suite of time history analyses. This site was selected because the empirical coherency function developed here has been used as a reference model for rock sites. We performed several sensitivity studies investigating the effect of both the site spatial variability and numerical analysis parameters in order to provide a guideline for developing a coherency model from numerical simulations. The outputs were compared against the empirical coherency model to better illustrate the importance of the parameters. The coefficient of variation (CV) of Vs was revealed to be the primary parameter influencing the calculated plane-wave coherency, whereas the correlation length (CL) has a secondary influence. Site-specific convergence analyses should be performed to determine the optimum numerical parameter, including the number of analyses and depth of numerical model. Considering the importance of CV and Vs, it is recommended to perform field tests to determine site-specific values to derive numerical coherency functions. Full article
(This article belongs to the Section Civil Engineering)
►▼ Show Figures

Figure 1

Figure 1
<p>Distribution of selected earthquake records in terms of epicentral distances, M, and PGV.</p> Full article ">Figure 2
<p>Reference <span class="html-italic">V<sub>s</sub></span> profile at the Pinyon Flat array.</p> Full article ">Figure 3
<p>Example realization of random field generated using the Gaussian model for the Pinyon Flat array.</p> Full article ">Figure 4
<p>Developed computational model.</p> Full article ">Figure 5
<p>Result of convergence with respect to number of numerical simulations.</p> Full article ">Figure 6
<p>Evaluation of sensitivity analysis regarding effect of CV on plane-wave coherency.</p> Full article ">Figure 7
<p>Evaluation of sensitivity analysis regarding effect of CLh on plane-wave coherency.</p> Full article ">Figure 8
<p>Evaluation of sensitivity analysis regarding effect of analysis depth on plane-wave coherency.</p> Full article ">Figure 9
<p>Comparison between numerical-based curve and empirical curve for Pinyon Flat site.</p> Full article ">
27 pages, 4262 KiB  
Article
A Self-Supervised Method for Speaker Recognition in Real Sound Fields with Low SNR and Strong Reverberation
by Xuan Zhang, Jun Tang, Huiliang Cao, Chenguang Wang, Chong Shen and Jun Liu
Appl. Sci. 2025, 15(6), 2924; https://doi.org/10.3390/app15062924 - 7 Mar 2025
Viewed by 244
Abstract
Speaker recognition is essential in smart voice applications for personal identification. Current state-of-the-art techniques primarily focus on ideal acoustic conditions. However, the traditional spectrogram struggles to differentiate between noise, reverberation, and speech. To overcome this challenge, MFCC can be replaced with the output [...] Read more.
Speaker recognition is essential in smart voice applications for personal identification. Current state-of-the-art techniques primarily focus on ideal acoustic conditions. However, the traditional spectrogram struggles to differentiate between noise, reverberation, and speech. To overcome this challenge, MFCC can be replaced with the output from a self-supervised learning model. This study introduces a TDNN enhanced with a pre-trained model for robust performance in noisy and reverberant environments, referred to as PNR-TDNN. The PNR-TDNN employs HuBERT as its backbone, while the TDNN is an improved ECAPA-TDNN. The pre-trained model employs the Canopy/Mini Batch k-means++ strategy. In the TDNN architecture, several enhancements are implemented, including a cross-channel fusion mechanism based on Res2Net. Additionally, a non-average attention mechanism is applied to the pooling operation, focusing on the weight information of each channel within the Squeeze-and-Excitation Net. Furthermore, the contribution of individual channels to the pooling of time-domain frames is enhanced by substituting attentive statistics with multi-head attention statistics. Validated by zhvoice in noisy conditions, the minimized PNR-TDNN demonstrates a 5.19% improvement in EER compared to CAM++. In more challenging environments with noise and reverberation, the minimized PNR-TDNN further improves EER by 3.71% and 9.6%, respectively, and MinDCF by 3.14% and 3.77%, respectively. The proposed method has also been validated on the VoxCeleb1 and cn-celeb_v2 datasets, representing a significant breakthrough in the field of speaker recognition under challenging conditions. This advancement is particularly crucial for enhancing safety and protecting personal identification in voice-enabled microphone applications. Full article
(This article belongs to the Section Computing and Artificial Intelligence)
19 pages, 4709 KiB  
Article
Study of Synergetic Optimization Operation for Distribution Network Considering Multiple Reactive Power Output Modes of Photovoltaics and Different Port Numbers of Flexible Interconnection Devices
by Yijin Li, Jibo Wang, Zihao Zhang, Wenhao Xu, Ming Wu and Geng Niu
Appl. Sci. 2025, 15(6), 2923; https://doi.org/10.3390/app15062923 - 7 Mar 2025
Viewed by 266
Abstract
Due to the integration of distributed photovoltaic (PV) into distribution networks, significant challenges have affected voltage regulation and power quality maintenance. To improve the flexibility and stability of system operation, a synergetic optimization operation method based on PV and a flexible interconnection device [...] Read more.
Due to the integration of distributed photovoltaic (PV) into distribution networks, significant challenges have affected voltage regulation and power quality maintenance. To improve the flexibility and stability of system operation, a synergetic optimization operation method based on PV and a flexible interconnection device (FID) is proposed. Both PV and FID hold the capability of controlling active power and reactive power. Besides the active power output of PV, three reactive power output schemes of power factor controlling, direct reactive power output, and night static var generator scheme are defined and analyzed. By adopting different schemes during the day or night, five reactive power output modes were built. FID with four-quadrant power control ability was used to coordinate with PV in system power balance. Different port numbers of FIDs are discussed. An optimization model with the aim of reducing voltage deviation, network loss, and the ratio of PV abandonment was constructed. Three algorithms were used for solving the multi-objective optimization model. Simulation results verify that the proposed synergetic optimization method can obviously improve power quality and decrease network loss. The optimal performance is obtained when PV operates in mode 5 and FID holds four ports. The proposed method shows potential in the coordinated operation of various resources and the flexible interconnection of the distribution network. Full article
►▼ Show Figures

Figure 1

Figure 1
<p>The reactive power output ranges of the PV inverter for five modes.</p> Full article ">Figure 2
<p>Flow chart of solving proposed model by NSGA-II.</p> Full article ">Figure 3
<p>Flow chart of solving proposed model by MOPSO.</p> Full article ">Figure 4
<p>The diagrams of the simulation cases.</p> Full article ">Figure 5
<p>Optimization results obtained by different algorithms for FID with different number of ports. “No” in horizontal axis means that reactive power of PVs is not optimized. “Yes” in horizontal axis means that reactive power of PVs is optimized. (<b>a</b>) FID of two ports. (<b>b</b>) FID of three ports. (<b>c</b>) FID of four ports.</p> Full article ">Figure 6
<p>Optimization result comparison by setting allowed ratio of PV reactive power output. Blue line—FID with two ports; red line—FID with three ports; green line—FID with four ports. (<b>a</b>) Active power loss of PV. (<b>b</b>) Voltage deviation. (<b>c</b>) Reduction in network loss.</p> Full article ">Figure 7
<p>Optimization results under five modes and with different FID port numbers.</p> Full article ">Figure 8
<p>Voltage distribution with (<b>a</b>) and without (<b>b</b>) synergetic optimization of PV and FID.</p> Full article ">Figure 9
<p>Active power (black) and reactive power (red) output curves of PVs.</p> Full article ">Figure 10
<p>Optimal scheme of FID. In (<b>a</b>), positive value means active power flow from port; negative value means active power flow into port. (<b>a</b>) Active power flows among ports. (<b>b</b>) Reactive power output of ports.</p> Full article ">
27 pages, 9713 KiB  
Article
HTSA-LSTM: Leveraging Driving Habits for Enhanced Long-Term Urban Traffic Trajectory Prediction
by Yiying Wei, Xiangyu Zeng, Xirui Chen, Hui Zhang, Zhengan Yang and Zhicheng Li
Appl. Sci. 2025, 15(6), 2922; https://doi.org/10.3390/app15062922 - 7 Mar 2025
Viewed by 277
Abstract
The rapid evolution of intelligent vehicle technology has significantly advanced autonomous decision-making and driving safety. However, the challenge of predicting long-term trajectories in complex urban traffic persists, as traditional methodologies usually handle spatiotemporal attention mechanisms in isolation and are typically limited to short-term [...] Read more.
The rapid evolution of intelligent vehicle technology has significantly advanced autonomous decision-making and driving safety. However, the challenge of predicting long-term trajectories in complex urban traffic persists, as traditional methodologies usually handle spatiotemporal attention mechanisms in isolation and are typically limited to short-term trajectory predictions. This paper proposes a Habit-based Temporal–Spatial Attention Long Short-Term Memory (HTSA-LSTM) network, a novel framework that integrates a dual spatiotemporal attention mechanism to capture dynamic dependencies across time and space, coupled with a driving style analysis module. The driving style analysis module employs Sparse Inverse Covariance Clustering and Spectral Clustering (SICC-SC) to extract driving primitives and cluster trajectory data, thereby revealing diverse driving behavior patterns without relying on predefined labels. By segmenting real-world driving data into fundamental behavioral units that reflect individual driving preferences, this approach enhances the model’s adaptability. These behavioral units, in conjunction with the spatiotemporal attention outputs, serve as inputs to the model, ultimately improving prediction accuracy and robustness in multi-vehicle scenarios. The model was evaluated by using the NGSIM dataset and real driving data from Wuhan, China. In comparison to benchmark models, HTSA-LSTM achieved a 20.72% reduction in the root mean square error (RMSE) and a 24.98% reduction in the negative log likelihood (NLL) for 5 s predictions of long-term trajectories. Furthermore, HTSA-LSTM achieved R2 values exceeding 97.9% for 5 s predictions on highways and expressways and over 92.7% for 3 s predictions on urban roads, highlighting its excellent performance in long-term trajectory prediction and adaptability across diverse driving conditions. Full article
►▼ Show Figures

Figure 1

Figure 1
<p>Discretization of the road network space into a 3 × 13 grid.</p> Full article ">Figure 2
<p>The traditional LSTM architecture.</p> Full article ">Figure 3
<p>The integration module of driving styles analysis.</p> Full article ">Figure 4
<p>HTSA-LSTM model diagram.</p> Full article ">Figure 5
<p>The data segmentation process.</p> Full article ">Figure 6
<p>The distribution of temporal attention.</p> Full article ">Figure 7
<p>The lane distribution at the Ventura Boulevard on-ramp.</p> Full article ">Figure 8
<p>Spatial attention allocation.</p> Full article ">Figure 9
<p>Spatial attention shift during lane change for vehicle 1585, where blue and purple means the car is changing lanes, and green means the car is unchanged in the current lane.</p> Full article ">Figure 10
<p>Driving behavior and kinematic features.</p> Full article ">Figure 11
<p>Standardized frequency distribution of driving primitives.</p> Full article ">Figure 12
<p>Examples of predictions for <span class="html-italic">v<sub>y</sub></span>, <span class="html-italic">a<sub>y</sub></span>, <span class="html-italic">v<sub>x</sub></span>, <span class="html-italic">a<sub>x</sub></span> for Vehicle 1585 (with H = 5 and H = 8).</p> Full article ">Figure 13
<p>The relationship between <math display="inline"><semantics> <mrow> <mi>e</mi> <mfenced open="[" close="]" separators="|"> <mrow> <msub> <mrow> <mi>v</mi> </mrow> <mrow> <mi>y</mi> </mrow> </msub> </mrow> </mfenced> </mrow> </semantics></math>, <math display="inline"><semantics> <mrow> <mi>e</mi> <mfenced open="[" close="]" separators="|"> <mrow> <msub> <mrow> <mi>a</mi> </mrow> <mrow> <mi>y</mi> </mrow> </msub> </mrow> </mfenced> </mrow> </semantics></math>, <math display="inline"><semantics> <mrow> <mi>e</mi> <mfenced open="[" close="]" separators="|"> <mrow> <msub> <mrow> <mi>v</mi> </mrow> <mrow> <mi>x</mi> </mrow> </msub> </mrow> </mfenced> </mrow> </semantics></math>, <math display="inline"><semantics> <mrow> <mi>e</mi> <mfenced open="[" close="]" separators="|"> <mrow> <msub> <mrow> <mi>a</mi> </mrow> <mrow> <mi>x</mi> </mrow> </msub> </mrow> </mfenced> </mrow> </semantics></math>, and the prediction horizon.</p> Full article ">Figure 14
<p>The trajectory predictions for vehicle 1585 and its neighboring vehicles.</p> Full article ">Figure 15
<p>Experimental routes.</p> Full article ">Figure 16
<p>Prediction results for speed and acceleration on urban city roads (CRs).</p> Full article ">Figure 17
<p>Prediction results for speed and acceleration on urban expressways (UEs).</p> Full article ">Figure 18
<p>Prediction results for speed and acceleration on highways (Hs).</p> Full article ">
16 pages, 6294 KiB  
Article
Polymer-Derived SiOC Ceramics by Digital Light Processing-Based Additive Manufacturing
by Xing Zhao, Jing Li, Ning Li, Lai Wei, Lin Zhang, Shuai Zhang and Haile Lei
Appl. Sci. 2025, 15(6), 2921; https://doi.org/10.3390/app15062921 - 7 Mar 2025
Viewed by 304
Abstract
Polymer-derived SiOC ceramics (PDCs-SiOC) possess advantages such as high temperature resistance, oxidation resistance, corrosion resistance, and customizable mechanical and dielectric properties. These attributes make them a promising material for high-temperature structural and functional applications. Based on polymer-derived ceramic conversion technology, this [...] Read more.
Polymer-derived SiOC ceramics (PDCs-SiOC) possess advantages such as high temperature resistance, oxidation resistance, corrosion resistance, and customizable mechanical and dielectric properties. These attributes make them a promising material for high-temperature structural and functional applications. Based on polymer-derived ceramic conversion technology, this study synthesized a photosensitive resin with high ceramic yield and low shrinkage from commercial MK resin, 3-(trimethoxysilyl) propyl methacrylate, and trimethylolpropane triacrylate monomer. Using digital light processing additive manufacturing technology, 3D diamond-structured SiOC ceramic and 3D octahedron-structured SiOC ceramic with high precision were fabricated. The pyrolysis of both structures at different temperatures (1000 °C to 1400 °C) yielded SiOC ceramics, which exhibited uniform shrinkage in all directions, with a linear shrinkage rate ranging from 31% to 36%. The microstructure was characterized by FTIR, XRD, and SEM, respectively. Additionally, the compressive strength and elastic modulus of the three-dimensional SiOC ceramics were studied. The SiOC ceramic diamond lattice structure, fabricated through pyrolysis at 1200 °C, demonstrated good mechanical properties with a geometric density of 0.76 g/cm³. Its compressive strength and elastic modulus were measured at 7.66 MPa and 1.47 GPa, respectively. This study offers valuable insights into the rapid and customized manufacturing of lightweight ceramic structures. Full article
(This article belongs to the Section Additive Manufacturing Technologies)
►▼ Show Figures

Figure 1

Figure 1
<p>Process diagram of the DLP-printed SiOC composite ceramics.</p> Full article ">Figure 2
<p>The hydrolysis condensation reaction process.</p> Full article ">Figure 3
<p>The FT-IR spectra of the polysiloxane resin: (a) MK and (b) MK-TMSPM.</p> Full article ">Figure 4
<p>The rheological properties of the MK-TMSPM ceramic slurry.</p> Full article ">Figure 5
<p>(<b>a</b>) The TG-DTG curves of the 3D-printed green parts. (<b>b</b>) The pyrolysis process temperature curve of the printed green parts.</p> Full article ">Figure 6
<p>(<b>a</b>) Diamond structure: the CAD model, green parts, and ceramic parts obtained by sintering at different temperatures. (<b>b</b>) Octahedron structures: the CAD model, green parts, and ceramic parts obtained by sintering at different temperatures.</p> Full article ">Figure 7
<p>The (<b>a</b>) XRD patterns, (<b>b</b>) Raman spectra, and (<b>c</b>) XPS spectra of the full spectrum of SiOC ceramics pyrolysis at 1000 °C, 1200 °C, and 1400 °C. The (<b>d</b>) Si 2p spectrum, (<b>e</b>) O 1s spectrum, and (<b>f</b>) C 1s spectrum of SiOC ceramics.</p> Full article ">Figure 8
<p>Surface microstructure of the SiOC ceramic after pyrolysis at different temperatures: (<b>a</b>,<b>b</b>) 1000 °C, (<b>c</b>,<b>d</b>) 1200 °C, and (<b>e</b>,<b>f</b>) 1400 °C.</p> Full article ">Figure 9
<p>Internal microstructure of SiOC ceramics after pyrolysis at different temperatures: (<b>a</b>) 1000 °C, (<b>b</b>) 1200 °C, (<b>c</b>) 1300 °C and (<b>d</b>) 1400 °C.</p> Full article ">Figure 10
<p>Mechanical compression performance testing: (<b>a</b>) three-dimensional ceramic structure compression testing process schematic diagram, and (<b>b</b>) compressive strength at different pyrolysis temperatures. Typical compression stress–strain curves at different pyrolysis temperatures: (<b>c</b>) 1000 °C, (<b>d</b>) 1200 °C, (<b>e</b>) 1300 °C, and (<b>f</b>) 1400 °C.</p> Full article ">
18 pages, 2974 KiB  
Article
Evolving Towards Artificial-Intelligence-Driven Sixth-Generation Mobile Networks: An End-to-End Framework, Key Technologies, and Opportunities
by Zexu Li, Jingyi Wang, Song Zhao, Qingtian Wang and Yue Wang
Appl. Sci. 2025, 15(6), 2920; https://doi.org/10.3390/app15062920 - 7 Mar 2025
Viewed by 192
Abstract
The incorporation of artificial intelligence (AI) into sixth-generation (6G) mobile networks is expected to revolutionize communication systems, transforming them into intelligent platforms that provide seamless connectivity and intelligent services. This paper explores the evolution of 6G architectures, as well as the enabling technologies [...] Read more.
The incorporation of artificial intelligence (AI) into sixth-generation (6G) mobile networks is expected to revolutionize communication systems, transforming them into intelligent platforms that provide seamless connectivity and intelligent services. This paper explores the evolution of 6G architectures, as well as the enabling technologies required to integrate AI across the cloud, core network (CN), radio access network (RAN), and terminals. It begins by examining the necessity of embedding AI into 6G networks, making it a native capability. The analysis then outlines potential evolutionary paths for the RAN architecture and proposes an end-to-end AI-driven framework. Additionally, key technologies such as cross-domain AI collaboration, native computing, and native security mechanisms are discussed. The study identifies potential use cases, including embodied intelligence, wearable devices, and generative AI, which offer valuable insights into fostering collaboration within the AI-driven ecosystem and highlight new revenue model opportunities and challenges. The paper concludes with a forward-looking perspective on the convergence of AI and 6G technology. Full article
(This article belongs to the Special Issue 5G/6G Mechanisms, Services, and Applications)
►▼ Show Figures

Figure 1

Figure 1
<p>Typical 6G use cases defined by the ITU-R.</p> Full article ">Figure 2
<p>Possible computing resource exposure options based on GPC and dedicated hardware.</p> Full article ">Figure 3
<p>AI-driven E2E next-generation framework.</p> Full article ">Figure 4
<p>AI collaboration across different domains.</p> Full article ">Figure 5
<p>RAN AI for wearable devices.</p> Full article ">Figure 6
<p>Cloud–edge–device collaboration for embodied intelligence.</p> Full article ">Figure 7
<p>Comparison: (a) Robotic guide dog without AI-native RAN. (b) Robotic guide dog with AI-native RAN.</p> Full article ">
19 pages, 9096 KiB  
Article
Speech Enhancement Based on Unidirectional Interactive Noise Modeling Assistance
by Yuewei Zhang, Huanbin Zou and Jie Zhu
Appl. Sci. 2025, 15(6), 2919; https://doi.org/10.3390/app15062919 - 7 Mar 2025
Viewed by 166
Abstract
It has been demonstrated that interactive speech and noise modeling outperforms traditional speech modeling-only methods for speech enhancement (SE). With a dual-branch topology that simultaneously predicts target speech and noise signals and employs bidirectional information communication between the two branches, the quality of [...] Read more.
It has been demonstrated that interactive speech and noise modeling outperforms traditional speech modeling-only methods for speech enhancement (SE). With a dual-branch topology that simultaneously predicts target speech and noise signals and employs bidirectional information communication between the two branches, the quality of the enhanced speech is significantly improved. However, the dual-branch topology greatly increases the model complexity and deployment cost, thus limiting its practicality. In this paper, we propose UniInterNet, a unidirectional information interaction-based dual-branch network to achieve noise modeling-assisted SE without any increase in complexity. Specifically, the noise branch still receives information from the speech branch to achieve more accurate noise modeling. Subsequently, the noise modeling results are utilized to assist the learning of the speech branch during backpropagation, while the speech branch no longer receives the auxiliary information from the noise branch, so only the speech branch is required during model deployment. Experimental results demonstrate that under the causal inference condition, the performance of UniInterNet only marginally decreases compared to the corresponding bidirectional information interaction scheme, while the model inference complexity is reduced by about 75%. With comparable overall performance, UniInterNet also outperforms previous interactive speech and noise modeling-based benchmarks in terms of causal inference and model complexity. Furthermore, UniInterNet surpasses other existing competitive methods. Full article
(This article belongs to the Section Computing and Artificial Intelligence)
►▼ Show Figures

Figure 1

Figure 1
<p>Overall architecture of unidirectional information interaction-based dual-branch network (UniInterNet).</p> Full article ">Figure 2
<p>(<b>a</b>) The detail of the two-dimensional convolutional (Conv2d) block. (<b>b</b>) The detail of the two-dimensional deconvolutional (DeConv2d) block.</p> Full article ">Figure 3
<p>The diagram of time–frequency sequence modeling (TFSM) block. During temporal sequence modeling, a causal gated recurrent unit (GRU) layer is employed in the speech branch, while a non-causal bidirectional GRU (BiGRU) layer is utilized in the noise branch.</p> Full article ">Figure 4
<p>Structure of the unidirectional interaction module.</p> Full article ">Figure 5
<p>Visualization of the spectrum of the following: (<b>a</b>) noisy speech; (<b>b</b>) clean speech; (<b>c</b>) enhanced speech by SiNet; (<b>d</b>) enhanced speech by BiInterNet; (<b>e</b>) enhanced speech by UniInterNet-CausalNoise; (<b>f</b>) enhanced speech by UniInterNet. The noise type is open area cafeteria noise.</p> Full article ">Figure 6
<p>Visualization of the spectrum of the following: (<b>a</b>) noisy speech; (<b>b</b>) clean speech; (<b>c</b>) enhanced speech by UniInterNet w/o EncInter; (<b>d</b>) enhanced speech by UniInterNet w/o RecInter; (<b>e</b>) enhanced speech by UniInterNet w/o DecInter; (<b>f</b>) enhanced speech by UniInterNet. The noise type is public square noise.</p> Full article ">
18 pages, 1176 KiB  
Article
Acute Effects of a High-Intensity Interval Training Protocol on Pain Sensitivity and Inflammatory Markers in Persons with Chronic Nonspecific Low Back Pain: A Controlled Clinical Trial
by Jonas Verbrugghe, Sim Klaps, Kenneth Verboven, Timo Meus, Kristof Kempeneers, Kristian Kjaer-Staal Petersen and Annick Timmermans
Appl. Sci. 2025, 15(6), 2918; https://doi.org/10.3390/app15062918 - 7 Mar 2025
Viewed by 176
Abstract
Chronic nonspecific low back pain (CNSLBP) might be associated with increased pain sensitivity and inflammation. High-intensity interval training (HIIT) has been suggested to reduce pain outcomes and inflammatory markers, but its effects compared to moderate-intensity continuous training (MICT) remain unclear. This study aimed [...] Read more.
Chronic nonspecific low back pain (CNSLBP) might be associated with increased pain sensitivity and inflammation. High-intensity interval training (HIIT) has been suggested to reduce pain outcomes and inflammatory markers, but its effects compared to moderate-intensity continuous training (MICT) remain unclear. This study aimed to evaluate the acute effects of HIIT on pain sensitivity and inflammatory markers in persons with CNSLBP compared to healthy controls (HCs) and to determine how these effects differ from MICT. Twenty persons with CNSLBP and twenty HCs were assessed before (PRE) and after (POST) a single HIIT and MICT protocol for pain sensitivity (cuff pressure pain threshold (cPPT), temporal summation of pain (TS), conditioned pain modulation (CPM)), and inflammatory markers (IL-6, TNF-α). Data were analyzed using one-way ANOVAs, paired t-tests, and correlation analyses. At PRE, persons with CNSLBP exhibited lower cPPT (28.2 ± 7.1, Δ = −5.5, p = 0.040), higher TS (1.11 ± 0.89, Δ = 0.79, p = 0.042), and lower CPM (36.2 ± 11.6, Δ = −10.0, p = 0.023) compared to HCs. HIIT resulted in PRE–POST improvements in cPPT (38.9 ± 12.6, Δ = 5.2, p = 0.019) in HCs. No PRE–POST differences were observed in pain processing in those with CLBP. No PRE or PRE–POST differences were observed in the inflammatory markers in either group. The current exploratory study suggests that a single HIIT session might have a beneficial effect on pain sensitivity in HCs but does not alter acute pain sensitivity or inflammatory markers in persons with CNSLBP. Further research is needed to clarify the involved mechanisms and explore the (relation with the) long-term effects of HIIT. Full article
(This article belongs to the Special Issue Exercise, Fitness, Human Performance and Health: 2nd Edition)
►▼ Show Figures

Figure 1

Figure 1
<p>Study design. Abbreviations: HCs: healthy controls; CNSLBP: chronic nonspecific low back pain; CPET: cardiorespiratory exercise test; HIIT: high-intensity interval training; MICT: moderate-intensity continuous training.</p> Full article ">Figure 2
<p>Content of a session. Abbreviations: IL-6: interleukin-6; TNF-α: cytokine tumor necrosis factor alpha; HIIT: high-intensity interval training; MICT: moderate-intensity continuous training.</p> Full article ">Figure 3
<p>Overview of the inflammation outcomes (panels <b>A</b>–<b>D</b>). Abbreviations: IL-6: interleukin-6; TNF-α: cytokine tumor necrosis factor alpha; HCs: healthy controls; CNSLBP: chronic nonspecific low back pain; HIIT: high-intensity interval training; MICT: moderate-intensity continuous training; ns: non-significant.</p> Full article ">
25 pages, 13369 KiB  
Article
Three-Dimensional Path-Following with Articulated 6DoF Robot and ToF Sensors
by Tymon F. Wawrzyniak, Ignacy D. Orłowski and Marek A. Galewski
Appl. Sci. 2025, 15(6), 2917; https://doi.org/10.3390/app15062917 - 7 Mar 2025
Viewed by 176
Abstract
This paper presents an algorithm for 3D path-following using an articulated 6-Degree-of-Freedom (DoF) robot as well as experimental verification of the proposed approach. This research extends the classic line-following concept, typically applied in 2D spaces, into a 3D space. This is achieved by [...] Read more.
This paper presents an algorithm for 3D path-following using an articulated 6-Degree-of-Freedom (DoF) robot as well as experimental verification of the proposed approach. This research extends the classic line-following concept, typically applied in 2D spaces, into a 3D space. This is achieved by equipping a standard industrial robot with a path detection tool featuring six low-cost Time-of-Flight (ToF) sensors. The primary objective is to enable the robot to follow a physically existing path defined in 3D space. The developed algorithm allows for step-by-step detection of the path’s orientation and calculation of consecutive positions and orientations of the detection tool that are necessary for the robot arm to follow the path. Experimental tests conducted using a Nachi MZ04D robot demonstrated the reliability and effectiveness of the proposed solution. Full article
(This article belongs to the Special Issue Motion Control for Robots and Automation)
►▼ Show Figures

Figure 1

Figure 1
<p>Control principles for typical line-following, 2-wheeled robot showing situations resulting in (<b>a</b>) turn left, (<b>b</b>) move straight, (<b>c</b>) turn right commands.</p> Full article ">Figure 2
<p>Robot holding path detection tool during the path-following.</p> Full article ">Figure 3
<p>Path detection tool schema showing ToF sensors (1–6) placement, sensor separation <span class="html-italic">a</span>, and locations of detection points (purple circles), e.g., point A is the detection point at the intersection of the lines of sight of sensors 2 and 4.</p> Full article ">Figure 4
<p>The <math display="inline"><semantics> <mrow> <mover accent="true"> <mrow> <mi mathvariant="bold-italic">w</mi> </mrow> <mo>→</mo> </mover> </mrow> </semantics></math> vector and its component vectors and location of <math display="inline"><semantics> <mrow> <msub> <mrow> <mi>P</mi> </mrow> <mrow> <mi>n</mi> </mrow> </msub> </mrow> </semantics></math> (next tool position) point in the global coordinate system; <math display="inline"><semantics> <mrow> <mover accent="true"> <mrow> <mi mathvariant="bold-italic">w</mi> </mrow> <mo>^</mo> </mover> </mrow> </semantics></math> is the unit vector parallel to <math display="inline"><semantics> <mrow> <mover accent="true"> <mrow> <mi mathvariant="bold-italic">w</mi> </mrow> <mo>→</mo> </mover> </mrow> </semantics></math>. Other symbols: <math display="inline"><semantics> <mrow> <mover accent="true"> <mrow> <mi mathvariant="bold-italic">t</mi> </mrow> <mo>→</mo> </mover> </mrow> </semantics></math>—tool displacement vector; <math display="inline"><semantics> <mrow> <mi>O</mi> </mrow> </semantics></math>—detection tool centre in the current position; <math display="inline"><semantics> <mrow> <mi>O</mi> <mi mathvariant="normal">’</mi> </mrow> </semantics></math> —detection tool centre in the previous position; and <span class="html-italic">S</span>—detected path position.</p> Full article ">Figure 5
<p>Three consecutive positions of the detection tool during path-following. Symbols: <math display="inline"><semantics> <mrow> <mi>O</mi> </mrow> </semantics></math>—detection tool centre in the current position; <math display="inline"><semantics> <mrow> <mi>O</mi> <mi mathvariant="normal">’</mi> </mrow> </semantics></math>—detection tool centre in the previous position; <span class="html-italic">S</span>—detected path position for the current detection tool position; and <span class="html-italic">Pn</span>—calculated next tool centre position, 1…6—ToF sensors.</p> Full article ">Figure 6
<p>The <math display="inline"><semantics> <mrow> <mover accent="true"> <mrow> <msub> <mrow> <mi mathvariant="bold-italic">w</mi> </mrow> <mrow> <mi mathvariant="bold-italic">a</mi> </mrow> </msub> </mrow> <mo>→</mo> </mover> </mrow> </semantics></math> vector with its components, <math display="inline"><semantics> <mrow> <msub> <mrow> <mi>w</mi> </mrow> <mrow> <mi>a</mi> <mo>,</mo> <mi>x</mi> </mrow> </msub> <mo>,</mo> <msub> <mrow> <mi>w</mi> </mrow> <mrow> <mi>a</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> <mo>,</mo> <msub> <mrow> <mi>w</mi> </mrow> <mrow> <mi>a</mi> <mo>,</mo> <mi>z</mi> </mrow> </msub> </mrow> </semantics></math>, and two rotation angles that can be determined on the basis of this vector: <math display="inline"><semantics> <mrow> <msub> <mrow> <mi>β</mi> </mrow> <mrow> <mi>n</mi> </mrow> </msub> </mrow> </semantics></math> and <math display="inline"><semantics> <mrow> <msub> <mrow> <mi>γ</mi> </mrow> <mrow> <mi>n</mi> </mrow> </msub> </mrow> </semantics></math>.</p> Full article ">Figure 7
<p>Example of combining the next position and tool rotation; detection tool seen from the top. Symbols: <math display="inline"><semantics> <mrow> <mi>O</mi> </mrow> </semantics></math>—detection tool centre in the current position; <math display="inline"><semantics> <mrow> <mi>O</mi> <mi mathvariant="normal">’</mi> </mrow> </semantics></math>—detection tool centre in the previous position; <span class="html-italic">S</span>—detected path position for the current detection tool position; <span class="html-italic">P<sub>n</sub></span>—calculated next tool centre position; vectors <math display="inline"><semantics> <mrow> <mover accent="true"> <mrow> <msub> <mrow> <mi mathvariant="bold-italic">w</mi> </mrow> <mrow> <mi mathvariant="bold-italic">a</mi> </mrow> </msub> </mrow> <mo>→</mo> </mover> </mrow> </semantics></math>, <math display="inline"><semantics> <mrow> <mover accent="true"> <mrow> <mi mathvariant="bold-italic">w</mi> </mrow> <mo>→</mo> </mover> </mrow> </semantics></math>, <math display="inline"><semantics> <mrow> <mover accent="true"> <mrow> <mi mathvariant="bold-italic">t</mi> </mrow> <mo>→</mo> </mover> </mrow> </semantics></math>, <math display="inline"><semantics> <mrow> <mover accent="true"> <mrow> <mi mathvariant="bold-italic">n</mi> </mrow> <mo>→</mo> </mover> </mrow> </semantics></math>, and <math display="inline"><semantics> <mrow> <mover accent="true"> <mrow> <mi mathvariant="bold-italic">m</mi> </mrow> <mo>→</mo> </mover> </mrow> </semantics></math> and <span class="html-italic">a</span> and <span class="html-italic">b</span> constants are described in the text.</p> Full article ">Figure 8
<p>Path detection tool used during experimental verification: (<b>a</b>) view of the real tool in the robot gripper placed over the path to follow; (<b>b</b>) main tool diameters.</p> Full article ">Figure 9
<p>General software architecture of the presented system.</p> Full article ">Figure 10
<p>Path-following algorithm schema.</p> Full article ">Figure 11
<p>Path 1, run 1—path (green line) detected by the detection tool (<b>a</b>) isometric view, (<b>b</b>) front view (YZ plane), (<b>c</b>) side view (XZ plane), and (<b>d</b>) top view (YX plane).</p> Full article ">Figure 12
<p>Path 1, run 1—(<b>a</b>–<b>c</b>) three selected examples of robot positions obtained during path-following; full video available as <a href="#app1-applsci-15-02917" class="html-app">Supplementary Material [S2]</a>.</p> Full article ">Figure 13
<p>Path 1, run 2—path (green line) detected by the detection tool (<b>a</b>) isometric view, (<b>b</b>) front view (YZ plane), (<b>c</b>) side view (XZ plane), and (<b>d</b>) top view (YX plane).</p> Full article ">Figure 14
<p>Path 1, run 2—(<b>a</b>–<b>c</b>) three selected examples of robot positions obtained during path-following; full video available as <a href="#app1-applsci-15-02917" class="html-app">Supplementary Material [S2]</a>.</p> Full article ">Figure 15
<p>Path 1, run 1 and 2 overlayed; the robot body is removed for a clearer view.</p> Full article ">Figure 16
<p>Path 2, run 1—path (green line) detected by the detection tool (<b>a</b>) isometric view, (<b>b</b>) front view (YZ plane), (<b>c</b>) side view (XZ plane), and (<b>d</b>) top view (YX plane).</p> Full article ">Figure 17
<p>Path 2, run 1—(<b>a</b>–<b>c</b>) three selected examples of robot positions obtained during path-following; full video available as <a href="#app1-applsci-15-02917" class="html-app">Supplementary Material [S2]</a>.</p> Full article ">Figure 18
<p>Path 3, run 1—path (green line) detected by the detection tool (<b>a</b>) isometric view, (<b>b</b>) front view (YZ plane), (<b>c</b>) side view (XZ plane), and (<b>d</b>) top view (YX plane).</p> Full article ">Figure 19
<p>Path 3, run 1—(<b>a</b>–<b>c</b>) three selected examples of robot positions obtained during path-following; full video available as <a href="#app1-applsci-15-02917" class="html-app">Supplementary Material [S2]</a>.</p> Full article ">Figure 20
<p>Path 4, run 1—path (green line) detected by the detection tool (<b>a</b>) isometric view, (<b>b</b>) front view (YZ plane), (<b>c</b>) side view (XZ plane), and (<b>d</b>) top view (YX plane).</p> Full article ">Figure 21
<p>Path 4, run 1—(<b>a</b>–<b>c</b>) three selected examples of robot positions obtained during path-following; full video available as <a href="#app1-applsci-15-02917" class="html-app">Supplementary Material [S2]</a>.</p> Full article ">
11 pages, 2089 KiB  
Article
An Experimental Study of Radiated Energy from an Optical Fiber and the Potential for an Optical MIMO System
by Hasan Farahneh, Jamal S. Rahhal, Dia I. Abualnadi, Ibrahim Mansour, Ahmad K. Atieh and Xavier Fernando
Appl. Sci. 2025, 15(6), 2916; https://doi.org/10.3390/app15062916 - 7 Mar 2025
Viewed by 185
Abstract
Leaky feeders provide seamless and uniform signal coverage in confined spaces like tunnels, mines, and buildings. Their easy scalability and integration with modern systems, like Multiple Input Multiple Output (MIMO), make them ideal for environments requiring reliable and consistent connectivity. However, using optical [...] Read more.
Leaky feeders provide seamless and uniform signal coverage in confined spaces like tunnels, mines, and buildings. Their easy scalability and integration with modern systems, like Multiple Input Multiple Output (MIMO), make them ideal for environments requiring reliable and consistent connectivity. However, using optical fiber as a radiating cable has never been investigated before. This may seem infeasible at first sight. However, our experimental study shows otherwise. We measured light leaking from a bent optical fiber transmitter. We also derived closed-form formulas to describe the amount of leakage energy and found that this energy exponentially varies with the square of the curvature radius. This allows us to design an Optical Leaky Feeder (OLF) transmission system for the first time. Then, we analytically show that a slotted optical fiber can be used as a MIMO receiver. The proposed system can ensure reliable, high-quality signal distribution even in challenging environments like tunnels, industrial settings, and dense urban areas. Full article
(This article belongs to the Special Issue Signal Processing and Communication for Wireless Sensor Network)
►▼ Show Figures

Figure 1

Figure 1
<p>Light intensity profile before and inside fiber bending.</p> Full article ">Figure 2
<p>Light intensity profile inside bent fiber for R = 100 µm to R = 1800 µm in 10 steps.</p> Full article ">Figure 3
<p>Efficiency (×100%) vs. Bend Radius (mm).</p> Full article ">Figure 4
<p>Experiment Setup.</p> Full article ">Figure 5
<p>Measured Output Power (dBm) vs. Bend Radius(mm), and a Plot of Equation (24) as Dotted Line.</p> Full article ">Figure 6
<p>OLF as MIMO Transmission System.</p> Full article ">Figure 7
<p>BER vs. SNR (dB) for OLF MIMO System.</p> Full article ">
26 pages, 397 KiB  
Systematic Review
Dynamic Low-Emission Zones for Urban Mobility: A Systematic Review
by Pablo Manglano-Redondo, Alvaro Paricio-Garcia and Miguel A. Lopez-Carmona
Appl. Sci. 2025, 15(6), 2915; https://doi.org/10.3390/app15062915 - 7 Mar 2025
Viewed by 178
Abstract
Urban air pollution, particularly from vehicular emissions, poses a significant challenge to public health and environmental sustainability. Low-Emission Zones (LEZs) have emerged as a solution, reducing pollution in high-traffic areas by restricting access to high-emission vehicles. However, most LEZ implementations are static, failing [...] Read more.
Urban air pollution, particularly from vehicular emissions, poses a significant challenge to public health and environmental sustainability. Low-Emission Zones (LEZs) have emerged as a solution, reducing pollution in high-traffic areas by restricting access to high-emission vehicles. However, most LEZ implementations are static, failing to account for real-time changes in traffic and emissions. This review focuses on dynamic LEZ systems, which are adjusted based on real-time data to optimize emission reduction without disrupting traffic flow. By categorizing LEZ strategies into static, hybrid, and dynamic systems, this study highlights key case studies and technologies, such as traffic simulation tools and sensor networks, that enable these adaptive systems. The review also discusses the challenges and future opportunities in LEZ implementation, emphasizing the need for data-driven approaches to achieve both environmental and mobility goals. This study aims to provide insights for policymakers and researchers seeking to enhance urban air quality management through more flexible, efficient LEZ strategies. Full article
(This article belongs to the Special Issue Feature Review Papers in ‘Transportation and Future Mobility’ Section)
►▼ Show Figures

Figure 1

Figure 1
<p>PRISMA flowchart.</p> Full article ">
Show export options expand_more Show export options expand_less
Select all
Export citation of selected articles as:
 
Displaying articles 1-72
Previous Issue
Back to TopTop