CN119359719B - A particle detection method and system based on precursor network - Google Patents
A particle detection method and system based on precursor network Download PDFInfo
- Publication number
- CN119359719B CN119359719B CN202411919681.1A CN202411919681A CN119359719B CN 119359719 B CN119359719 B CN 119359719B CN 202411919681 A CN202411919681 A CN 202411919681A CN 119359719 B CN119359719 B CN 119359719B
- Authority
- CN
- China
- Prior art keywords
- precursor
- branch
- features
- feature extraction
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/52—Scale-space analysis, e.g. wavelet analysis
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/7715—Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10056—Microscopic image
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
本发明涉及机器视觉技术领域，公开了一种基于前驱体网络的颗粒检测方法及系统。方法包括：获取待处理前驱体的显微镜图像，对所述显微镜图像进行预处理；将预处理后的显微镜图像输入至前驱体网络中，获取待处理前驱体的类别、前驱体目标分割区域以及用于识别该类别前驱体的多模态数组，前驱体网络为基于多分支特征提取块构建双分支输出的网络模型，多分支特征提取块按照预定比值对显微镜图像的特征进行分割，对其中一部分特征处理后再与其他部分特征重新融合获取强化特征。本发明解决了现有的人工检测前驱体颗粒多类别识别和质量检测的问题。
The present invention relates to the field of machine vision technology, and discloses a particle detection method and system based on a precursor network. The method comprises: obtaining a microscope image of a precursor to be processed, and preprocessing the microscope image; inputting the preprocessed microscope image into the precursor network, obtaining the category of the precursor to be processed, the precursor target segmentation area, and a multimodal array for identifying the precursor of this category, wherein the precursor network is a network model with dual-branch outputs constructed based on a multi-branch feature extraction block, and the multi-branch feature extraction block segments the features of the microscope image according to a predetermined ratio, processes a part of the features, and then re-integrates with other parts of the features to obtain enhanced features. The present invention solves the problems of existing manual detection of multi-category recognition and quality detection of precursor particles.
Description
Technical Field
The invention relates to the technical field of machine vision, in particular to a particle detection method and system based on a precursor network.
Background
In the production process of high-performance materials and chemicals, microscopic features such as the size, the morphology and the like of precursor particles have decisive influence on the quality and the characteristics of the final product, different types of precursor particles have different sizes and morphologies, and at present, the detection of different types of particles has various limitations although the detection is automated or intelligent.
Taking CN118552569A as an example, the method carries out multi-layer extraction and fusion on image characteristics through an image segmentation model of a neural network, thereby improving segmentation precision and efficiency. However, this technique focuses mainly on the accuracy of image segmentation, and is still significantly insufficient in terms of automatic recognition of multi-class particles and detection of particles of complex morphology. In addition, the model depends on a specific feature fusion strategy, and cannot be fully adapted to complex and diverse industrial application scenes, particularly the characteristic characterization of unknown particles. Similarly, the method in CN116840108a realizes synchronous detection of particle size and particle shape through optical image analysis, while improving instantaneity, its detection object is limited to precursor particles synthesized continuously, and has limited capability of deep analysis of multi-modal characteristics and complex morphology particles.
Further, existing mainstream technologies such as YOLO (especially YOLOv and YOLOv 8) and UNet also face certain limitations. YOLOv5 and YOLOv are excellent in real-time performance in target detection, but the adopted meshing prediction mechanism is poor in coping with particles with blurred boundaries or complex morphology, and a detection frame is easy to deviate or undercover. In addition, YOLO models are weak in generalization ability when dealing with particles with large differences in size and morphology, and are not stable enough especially when detecting unknown classes of particles. While UNet realizes feature fusion through encoding-decoding structure and jump connection, although suitable for high-resolution target segmentation task, the capability of feature extraction and fusion is limited when processing particle morphology complexity under microscopic images, and meanwhile, the calculation cost is larger and the real-time performance is insufficient under high-resolution scenes. Accordingly, YOLO and UNet have difficulty fully meeting the needs of complex particle detection tasks.
Disclosure of Invention
The invention provides a particle detection method and system based on a precursor network, which are used for solving the problems of multi-category identification and quality detection of existing manual detection precursor particles.
In order to achieve the above object, the present invention is realized by the following technical scheme:
In a first aspect, the present invention provides a particle detection method based on a precursor network, where the precursor network includes an extraction module for extracting features, a multi-branch module for enhancing features, a fusion module for fusing features, and a dual-branch output module for outputting detection results, where the multi-branch module and the fusion module are constructed based on the multi-branch feature extraction module, and the dual-branch output module includes a class-division branch for outputting a class of a precursor and a target-division region, and a multi-mode branch for outputting a multi-mode array of the precursor;
The method comprises the following steps:
step 1, acquiring a microscope image of a precursor to be processed, and preprocessing the microscope image;
step 2, inputting the preprocessed microscope image into a precursor network, and obtaining the category of the precursor to be processed, a precursor target dividing area and a multi-mode array for identifying the category of the precursor;
The precursor network is a network model based on a multi-branch feature extraction block for constructing dual-branch output, the multi-branch feature extraction block is used for dividing the features of the microscope image according to a preset ratio, and the processed part of the divided features are recombined with other part of the features to obtain reinforced features;
the multi-branch feature extraction block comprises an input layer, an SE block, a fusion layer, an output layer, a segmentation layer for segmenting features according to a preset ratio and a depth separable convolution layer for extracting the features;
The segmentation layer acquires the features from the input layer, segments the features according to a preset ratio, then transmits partial features to the depth separable convolution layer, the depth separable convolution layer extracts the features of the segmented partial features, enhances the expression through the SE block, and finally fuses the features with another partial feature through the fusion layer and outputs the fused features through the output layer.
And determining whether the precursor meets the requirement or not by combining the last identified category with the dividing region, wherein the multi-mode array of the category precursor is a corresponding characteristic array used for identifying the unknown precursor, and the multi-mode array can be stored for identifying the unknown precursor next time.
Further, the multi-branch module comprises a first trunk for increasing receptive fields, a second trunk for enhancing feature extraction and a low-level feature extraction node for supplementing feature information;
The first trunk is connected with the extraction module and the second trunk respectively, the second trunk is connected with the low-level feature extraction node, and the first trunk and the low-level feature extraction node are connected with the fusion module;
the first trunk and the second trunk both comprise a multi-branch feature extraction block.
Further, the first trunk comprises a convolution layer with a convolution kernel of 3 and a step length of 2 and a multi-branch feature extraction block;
The second trunk comprises a convolution layer with a convolution kernel of 3 and a step length of 2 and a multi-branch feature extraction block;
The low-level feature extraction node comprises a convolution layer with a convolution kernel of 1.
Further, the fusion module comprises a multi-scale fusion unit and a dimension reduction unit, wherein the multi-scale fusion unit carries out fusion enhancement on the characteristics and then transmits the characteristics to the dimension reduction unit to carry out fusion dimension reduction treatment on the characteristics;
The first trunk and the low-level feature extraction nodes are connected with the multi-scale fusion unit, the multi-scale fusion unit is connected with the dimension reduction unit, and the low-level feature extraction nodes and the dimension reduction unit are connected with the double-branch output module.
Further, the multi-scale fusion unit comprises a convolution layer with a convolution kernel of 3, a multi-branch feature extraction block and a convolution layer with a convolution kernel of 1;
The dimension reduction unit sequentially comprises a splicing layer and a convolution layer with a convolution kernel of 3.
Further, the class division branch and the multi-mode branch both comprise convolution layers with convolution kernels of 1.
Further, a hierarchical self-adaptive initialization method is adopted in the construction process of the precursor network, and the precursor network is optimized through at least one of previous transmission, loss calculation, back transmission, verification set evaluation, early-stop strategy and super-parameter adjustment in the training process;
The loss calculation comprises the steps of obtaining the category of the precursor, a target segmentation area and a multi-mode array through precursor loss function constraint;
the precursor loss function expression is:
;
Wherein, Representing a precursor loss function; representing branch loss; representing a classification loss; Representing a mask penalty; representing a consistency loss; representing a weighted loss of attention; 、、、、 representing the weight coefficient.
In a second aspect, the invention also provides a particle detection system based on a precursor network, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of any of the methods described above when executing the computer program.
The beneficial effects are that:
According to the particle detection method and system based on the precursor network, the precursor network with double branch output respectively acquires the types, the segmentation areas and the multi-mode arrays of the type of precursors to be detected, the multi-mode arrays can be stored for the next identification of the unknown precursors, and the multi-branch feature extraction block segments and fuses the features, so that the calculated amount is reduced, the high-efficiency features are reserved, the feature output is enhanced, the accuracy of particle detection classification results is ensured, and the problems of multi-type identification and quality detection of the existing manual detection precursor particles are effectively solved;
The accuracy and the diversity recognition capability of particle detection are remarkably improved. The precursor network model with double branch outputs is introduced and is respectively used for extracting particle types, dividing regions and generating a multi-mode array, so that the high-efficiency identification and subsequent storage support of unknown particles are realized. Secondly, the multi-branch feature extraction block segments and fuses the features of the microscope image according to a preset ratio, so that the calculated amount is reduced, the expressive capacity of feature output is enhanced, and efficient feature extraction and reservation are ensured;
The method not only can automatically process the morphology, distribution and types of the particles, but also supports the fine detection of the morphology of the complex particles, and is particularly suitable for the multi-type detection requirement in the industrial production. By introducing the multi-mode array, the method can construct a characteristic representation mechanism with strong adaptability, and lays a solid foundation for the identification of unknown particles. Meanwhile, the robustness and generalization capability of the network model are further improved by applying the hierarchical self-adaptive initialization and optimization strategy, so that higher detection precision is maintained in a dynamic environment;
The multi-branch feature extraction block is used for dividing the features according to a preset ratio, and then one part of the features are processed and recombined with other parts of the features to obtain the reinforced features, so that the features are selectively enhanced, and the features are efficiently extracted and reserved.
Drawings
FIG. 1 is a flow chart of a particle detection method based on a precursor network according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a precursor network according to an embodiment of the present invention;
Fig. 3 is a schematic structural diagram of a multi-branch feature extraction block according to an embodiment of the invention.
Detailed Description
The following description of the present invention will be made clearly and fully, and it is apparent that the embodiments described are only some, but not all, of the embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Unless defined otherwise, technical or scientific terms used herein should be given the ordinary meaning as understood by one of ordinary skill in the art to which this invention belongs. The terms "first," "second," and the like, as used herein, do not denote any order, quantity, or importance, but rather are used to distinguish one element from another. Likewise, the terms "a" or "an" and the like do not denote a limitation of quantity, but rather denote the presence of at least one. The terms "connected" or "connected," and the like, are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "upper", "lower", "left", "right", etc. are used merely to indicate a relative positional relationship, which changes accordingly when the absolute position of the object to be described changes.
Referring to fig. 1, an embodiment of the present invention provides a particle detection method based on a precursor network, the method including the following steps:
step 1, acquiring a microscope image of a precursor to be processed, and preprocessing the microscope image;
wherein the preprocessing includes normalized size scaling and pixel normalization processing (e.g., normalizing pixel values to a [0,1] range) on each image and dividing the data into a training set, a validation set, and a test set. In addition, the data set is further expanded by using data enhancement methods such as rotation, random clipping and brightness adjustment, so that the robustness of the model under different input conditions is improved.
Step 2, inputting the preprocessed microscope image into a precursor network, and obtaining the category of the precursor to be processed, a precursor target dividing area and a multi-mode array for identifying the category of the precursor;
The precursor network is a network model based on a multi-branch feature extraction block for constructing dual-branch output, the multi-branch feature extraction block divides the features of the precursor according to a preset ratio, and the segmented part of features are recombined with other part of features to obtain reinforced features after being processed;
referring to fig. 2, the precursor network includes an extracting module for extracting features, a multi-branch module for enhancing features, a fusion module for fusing features, and a dual-branch output module for outputting detection results;
the multi-branch module and the fusion module are constructed based on the multi-branch feature extraction block.
In fig. 2, 15 nodes are included, wherein node 1 is an extraction module, nodes 2 to 6 are multi-branch modules, nodes 7 to 11 are fusion modules, and nodes 12 to 15 are dual-branch output modules.
The multi-branch module comprises a first trunk for increasing the receptive field, a second trunk for enhancing feature extraction and a low-level feature extraction node for supplementing feature information, in fig. 2, the first trunk is a node 2 and a node 3, the second trunk is a node 4 and a node 5, and the low-level feature extraction node is a node 6;
The output end of the extraction module is connected with the input end of the first trunk, the output end of the first trunk is connected with the input end of the second trunk, the output end of the second trunk is connected with the input end of the low-level characteristic extraction node, and the output ends of the first trunk and the low-level characteristic extraction node are connected with the input end of the fusion module.
Residual connection and splicing operations are introduced between different feature layers, so that the fusion capability of the feature images among multiple scales is enhanced, and the problem of gradient disappearance is avoided. The method has stronger sensitivity and robustness in the expression of multi-scale characteristics, and is more suitable for detecting tiny particles with complex morphology.
The first trunk comprises a convolution layer with a convolution kernel of 3 and a step length of 2 and a multi-branch feature extraction block, so that information of a larger receptive field can be captured;
the second trunk comprises a convolution layer with a convolution kernel of 3 and a step length of 2 and a multi-branch feature extraction block, so that the diversity of feature extraction is enhanced;
The low-level feature extraction node comprises a convolution layer with a convolution kernel of 1, and detail information of the features is added.
The fusion module comprises a multi-scale fusion unit and a dimension reduction unit, wherein the multi-scale fusion unit carries out fusion enhancement on the characteristics and then transmits the characteristics to the dimension reduction unit to carry out fusion dimension reduction processing on the characteristics, the multi-scale fusion unit is from node 7 to node 9 in the figure 2, and the dimension reduction unit is from node 10 to node 11;
The output ends of the first trunk and the low-level feature extraction nodes are connected with the input end of the multi-scale fusion unit, the output end of the multi-scale fusion unit is connected with the input end of the dimension reduction unit, and the output ends of the low-level feature extraction nodes and the dimension reduction unit are connected with the input end of the double-branch output module.
The multi-scale fusion unit comprises a convolution layer with a convolution kernel of 3, a multi-branch feature extraction block and a convolution layer with a convolution kernel of 1;
the enhancement unit sequentially comprises a splicing layer and a convolution layer with a convolution kernel of 3.
The double-branch output module comprises a category division branch and a multi-mode branch;
the multi-mode branch is used for outputting a multi-mode array of the precursor;
Both the class-split branches and the multi-modal branches include a convolution layer with a convolution kernel of 1.
For the multi-branch feature extraction block in the multi-branch module and the fusion module, referring to fig. 3 specifically, the multi-branch feature extraction block includes an input layer, a segmentation layer for segmenting features according to a predetermined ratio, a depth separable convolution layer for extracting features, a SE block, a fusion layer, and an output layer;
The segmentation layer acquires the features from the input layer, segments the features according to a preset ratio, then transmits partial features to the depth separable convolution layer, the depth separable convolution layer extracts the features of the segmented partial features, enhances the expression through the SE block, and finally fuses the features with another partial feature through the fusion layer and outputs the fused features through the output layer;
in this embodiment, the segmentation layer segments the feature into two parts, one part being One part isFor the split-outThe feature is processed by a depth separable convolution layer, the parameter and the calculated amount are reduced, the sensitivity to the spatial feature is kept, the extraction of the fine granularity feature is realized, the adaptive weighting is quickly carried out by SE after the processing, the expression of important features is enhanced, and finally the processing is carried out by a fusion layer and untreated featuresAnd outputting the fused features through an output layer.
The multi-branch feature extraction block is used for comprehensively capturing the particle information of different scales, so that the resolving power of complex backgrounds and tiny particles is effectively improved, a large amount of original information can be reserved, key features are highlighted, and the detection precision and the detail recognition capability are greatly improved.
In the invention, a hierarchical self-adaptive initialization method is adopted in the construction process of the precursor network, and the precursor network is optimized through previous propagation, loss calculation, back propagation, verification set evaluation, early-stop strategy and super-parameter adjustment in the training process.
The method comprises the steps of calculating the loss, wherein the loss calculation comprises the category of a precursor, a target segmentation area and a multi-mode array, which are obtained through precursor loss function constraint;
The precursor loss function expression is:
;
Wherein, Representing a precursor loss function; representing branch loss; representing a classification loss; Representing a mask penalty; representing a consistency loss; representing a weighted loss of attention; 、、、、 representing the weight coefficient.
Branch loss is used to supervise and encourage efficient expression of features such as particle morphology, size, and defects, while classification loss ensures proper classification of particles by cross entropy, mask loss is used for contour segmentation, and can be usedLoss or loss ofLoss to optimize edge detection of particles, enhance details of defect identification, consistency loss to constrain feature expression consistency between multiple branches, where one can useThe norm constraint distance ensures the consistency of different branch feature expressions, the attention weighting loss and the SE block integrating the attention mechanism in the multi-branch feature extraction block are further combined with the attention score to give different loss weights to different feature areas, so that the network pays more attention to a key area in counter propagation, the precursor loss function constructed by the loss is utilized fully, the multi-task and multi-branch structure of the precursor network are utilized fully, the innovation of the loss function is enhanced by the feature consistency and the attention mechanism, and the network performance in complex particle detection tasks is promoted.
Finally, the multi-modal array is used as a corresponding feature array when identifying the unknown precursor, and the multi-modal array can be stored for the next identification of the unknown precursor.
After the precursor network provided by the embodiment executes a particle detection method based on the precursor network, yolov is adopted to process the same microscope image, and after the detection result is obtained, the result is compared transversely, see table 1;
TABLE 1 Yolov8 lateral comparison with precursor networks
According to Table 1, the precursor network provided by the invention is better than Yolov in both the particle type and the partition area, so that the precursor particles can be detected and identified more accurately and effectively, and the requirement of precursor particle size and quality control can be met.
The foregoing describes in detail preferred embodiments of the present invention. It should be understood that numerous modifications and variations can be made in accordance with the concepts of the invention by one of ordinary skill in the art without undue burden. Therefore, all technical solutions which can be obtained by logic analysis, reasoning or limited experiments based on the prior art by the person skilled in the art according to the inventive concept shall be within the scope of protection defined by the claims.
Claims (8)
1. The particle detection method based on the precursor network is characterized in that the precursor network comprises an extraction module for extracting characteristics, a multi-branch module for strengthening the characteristics, a fusion module for fusing the characteristics and a dual-branch output module for outputting detection results, wherein the multi-branch module and the fusion module are constructed based on the multi-branch characteristic extraction module, and the dual-branch output module comprises a category segmentation branch for outputting categories of precursors and target segmentation areas and a multi-mode branch for outputting multi-mode arrays of the precursors;
The method comprises the following steps:
step 1, acquiring a microscope image of a precursor to be processed, and preprocessing the microscope image;
step 2, inputting the preprocessed microscope image into a precursor network, and obtaining the category of the precursor to be processed, a precursor target dividing area and a multi-mode array for identifying the category of the precursor;
The precursor network is a network model based on a multi-branch feature extraction block for constructing dual-branch output, the multi-branch feature extraction block is used for dividing the features of the microscope image according to a preset ratio, and the processed part of the divided features are recombined with other part of the features to obtain reinforced features;
the multi-branch feature extraction block comprises an input layer, an SE block, a fusion layer, an output layer, a segmentation layer for segmenting features according to a preset ratio and a depth separable convolution layer for extracting the features;
The segmentation layer acquires the features from the input layer, segments the features according to a preset ratio, then transmits partial features to the depth separable convolution layer, the depth separable convolution layer extracts the features of the segmented partial features, enhances the expression through the SE block, and finally fuses the features with another partial feature through the fusion layer and outputs the fused features through the output layer.
2. The precursor network-based particle detection method of claim 1, wherein the multi-branch module comprises a first stem that increases receptive field, a second stem that enhances feature extraction, and a low-level feature extraction node that supplements feature information;
The first trunk is connected with the second trunk, the second trunk is connected with the low-level feature extraction node, and the second trunk and the low-level feature extraction node are connected with the fusion module;
the first trunk and the second trunk both comprise a multi-branch feature extraction block.
3. The precursor network-based particle detection method of claim 2, wherein the first backbone comprises a convolution layer with a convolution kernel of 3 and a step size of 2 and a multi-branch feature extraction block;
The second trunk comprises a convolution layer with a convolution kernel of 3 and a step length of 2 and a multi-branch feature extraction block;
The low-level feature extraction node comprises a convolution layer with a convolution kernel of 1.
4. The precursor network-based particle detection method according to claim 2, wherein the fusion module comprises a multi-scale fusion unit and a dimension reduction unit, and the multi-scale fusion unit carries out fusion enhancement on the characteristics and then transmits the characteristics to the dimension reduction unit for fusion dimension reduction treatment;
the second trunk and the low-level feature extraction nodes are connected with the multi-scale fusion unit, the multi-scale fusion unit is connected with the dimension reduction unit, and the low-level feature extraction nodes and the dimension reduction unit are connected with the double-branch output module.
5. The precursor network-based particle detection method of claim 4, wherein the multi-scale fusion unit comprises a convolution layer with a convolution kernel of 3, a multi-branch feature extraction block, and a convolution layer with a convolution kernel of 1;
The dimension reduction unit sequentially comprises a splicing layer and a convolution layer with a convolution kernel of 3.
6. The precursor network-based particle detection method of claim 2, wherein the class-split branches and the multi-modal branches each comprise a convolution layer with a convolution kernel of 1.
7. The precursor network-based particle detection method of any one of claims 1-6, wherein a hierarchical adaptive initialization method is employed in the precursor network construction process, and the precursor network is optimized during training by at least one of previous propagation, loss calculation, back propagation, validation set evaluation, early-stop strategy, and super-parameter adjustment;
The loss calculation comprises a precursor category, a target segmentation area and a multi-mode array, wherein the precursor category, the target segmentation area and the multi-mode array are obtained through precursor loss function constraint;
the precursor loss function expression is:
;
Wherein, Representing a precursor loss function; representing branch loss; representing a classification loss; Representing a mask penalty; representing a consistency loss; representing a weighted loss of attention; 、、、、 representing the weight coefficient.
8. A precursor network based particle detection system comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method of any of the preceding claims 1-7 when the computer program is executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411919681.1A CN119359719B (en) | 2024-12-25 | 2024-12-25 | A particle detection method and system based on precursor network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411919681.1A CN119359719B (en) | 2024-12-25 | 2024-12-25 | A particle detection method and system based on precursor network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN119359719A CN119359719A (en) | 2025-01-24 |
| CN119359719B true CN119359719B (en) | 2025-04-18 |
Family
ID=94309140
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202411919681.1A Active CN119359719B (en) | 2024-12-25 | 2024-12-25 | A particle detection method and system based on precursor network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN119359719B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110866526A (en) * | 2018-08-28 | 2020-03-06 | 北京三星通信技术研究有限公司 | Image segmentation method, electronic device and computer-readable storage medium |
| CN112528904A (en) * | 2020-12-18 | 2021-03-19 | 中山艾尚智同信息科技有限公司 | Image segmentation method for sand particle size detection system |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPWO2005066657A1 (en) * | 2004-01-09 | 2007-12-20 | 浩之 高橋 | Particle detector and particle detection method |
| WO2017053592A1 (en) * | 2015-09-23 | 2017-03-30 | The Regents Of The University Of California | Deep learning in label-free cell classification and machine vision extraction of particles |
| CN114973073B (en) * | 2022-05-13 | 2024-10-18 | 中南大学 | Measuring method and experimental device for numerical information of dynamic granular materials |
| CN115937099A (en) * | 2022-11-08 | 2023-04-07 | 温州大学智能锁具研究院 | A BGA Solder Joint Segmentation Method Based on UNet Depth Separable Residual Network |
-
2024
- 2024-12-25 CN CN202411919681.1A patent/CN119359719B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110866526A (en) * | 2018-08-28 | 2020-03-06 | 北京三星通信技术研究有限公司 | Image segmentation method, electronic device and computer-readable storage medium |
| CN112528904A (en) * | 2020-12-18 | 2021-03-19 | 中山艾尚智同信息科技有限公司 | Image segmentation method for sand particle size detection system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN119359719A (en) | 2025-01-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113239782B (en) | Pedestrian re-recognition system and method integrating multi-scale GAN and tag learning | |
| CN116958825B (en) | Mobile remote sensing image acquisition method and highway maintenance monitoring method | |
| CN113762138B (en) | Identification method, device, computer equipment and storage medium for fake face pictures | |
| CN111768388B (en) | Product surface defect detection method and system based on positive sample reference | |
| CN114782311B (en) | CENTERNET improvement-based multi-scale defect target detection method and system | |
| CN114581434A (en) | Pathological image processing method and electronic equipment based on deep learning segmentation model | |
| CN115205616B (en) | Cloth flaw data enhancement method based on generation countermeasure network | |
| CN114463759A (en) | Lightweight character detection method and device based on anchor-frame-free algorithm | |
| CN118229548A (en) | Infrared and visible image fusion method based on progressive multi-branch and improved UNet3+ deep supervision | |
| CN112669343A (en) | Zhuang minority nationality clothing segmentation method based on deep learning | |
| CN117067112B (en) | Water cutting machine and control method thereof | |
| CN114863199A (en) | Target detection method based on optimized anchor frame mechanism | |
| Liu et al. | A generative adversarial network for infrared and visible image fusion using adaptive dense generator and Markovian discriminator | |
| CN120766047B (en) | Industrial product surface defect detection method based on characteristic coupling | |
| CN119359719B (en) | A particle detection method and system based on precursor network | |
| CN120598933A (en) | A surface defect detection method for aluminum profiles based on YOLOv8 | |
| CN110889858A (en) | Automobile part segmentation method and device based on point regression | |
| CN120235882A (en) | Magnetic mineral particle detection method and system under microscope integrating attention mechanism | |
| CN113723453A (en) | Pollen image classification method and device | |
| CN119295823A (en) | PointNet++ 3D point cloud object classification method integrating contextual anchor attention mechanism | |
| CN119495004A (en) | Water extraction method from remote sensing images based on multi-scale feature fusion of compressed pyramid gating units | |
| CN119048802A (en) | Fabric defect image classification method based on deep learning | |
| CN121033552B (en) | A Magnetic Resonance Imaging Classification Method and System Based on Dual-Branch Networks | |
| Abdou et al. | Deep learning and computer vision techniques for microcirculation analysis: A review | |
| CN119919474B (en) | Method, device and medium for detecting forging size in thermoforming process |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |