CN118116554B - Medical image caching processing method based on big data processing - Google Patents
Medical image caching processing method based on big data processing Download PDFInfo
- Publication number
- CN118116554B CN118116554B CN202410232171.0A CN202410232171A CN118116554B CN 118116554 B CN118116554 B CN 118116554B CN 202410232171 A CN202410232171 A CN 202410232171A CN 118116554 B CN118116554 B CN 118116554B
- Authority
- CN
- China
- Prior art keywords
- data
- medical image
- image data
- cache
- feature
- 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
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
-
- 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
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
- G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
- G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
-
- 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
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Multimedia (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Biodiversity & Conservation Biology (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The invention discloses a medical image caching method based on big data processing, which relates to the technical field of big data processing, and comprises the steps of firstly acquiring and preprocessing image data, then carrying out classification marking on image types, case types, parts and disease severity on the medical image data, then carrying out compression processing on the medical image data subjected to classification marking processing through an improved deep learning algorithm model, then carrying out image distributed caching processing, and carrying out cache data management through a cache management model; the invention solves the problems of limited capacity, low cache efficiency and difficult timely update of cache data of the traditional medical image cache processing method, can realize quick and accurate cache processing of medical image data, improves the diagnosis and treatment efficiency of doctors on patients, and greatly improves the data information processing capability.
Description
Technical Field
The invention relates to the technical field of big data processing, in particular to a medical image caching processing method based on big data processing.
Background
With the continuous development and popularization of medical imaging technology, the amount of medical image data has been increasing explosively. Medical image data is one of important basic data in medical diagnosis and treatment, the amount of medical image data generated by medical institutions is increasingly large, and because medical image files are usually very large, a large amount of storage space is required, direct access and processing of the data are often inefficient, a large amount of time and bandwidth are required for processing the data, and efficient storage and management of the medical image data files are required. Therefore, how to quickly and accurately access, share and manage such data becomes a great problem. The traditional medical image data management method, such as storing the image data in a local storage device, cannot meet the diagnosis and treatment needs of doctors.
However, the traditional medical image caching method lacks data processing capability, can only store and access medical image data in a simple caching manner, cannot perform deep feature extraction and management on the medical image data, cannot quickly and accurately perform classification identification and feature extraction on the medical image data, lacks a targeted management strategy, has certain limitation on data precision, adopts a single caching mechanism, cannot quickly and accurately cache the medical image data, and has low caching efficiency, the traditional medical image caching method lacks an optimized cache data management model, cannot perform efficient and stable management on the cache data, and has low data management efficiency.
Therefore, the invention discloses a medical image caching method based on big data processing, which optimizes the traditional medical image caching method in such modes as classifying and marking medical images by adopting a self-supervision image characterization model, compressing medical images by adopting an improved deep learning algorithm model, carrying out distributed caching by a cloud cache server, carrying out data management by a cache management model and the like, and has higher efficiency, more accuracy and safer.
Disclosure of Invention
Aiming at the defects of the prior art, the invention discloses a medical image caching method based on big data processing, which optimizes the traditional medical image caching method in such modes as classifying and marking medical images by adopting a self-supervision image characterization model, compressing medical images by adopting an improved deep learning algorithm model, carrying out distributed caching on a cloud cache server, carrying out data management on a cache management model and the like, and has higher efficiency, more accuracy and safer; the self-supervision image characterization model is adopted to carry out classification and marking processing on the medical image data, the improved deep learning algorithm model is adopted to carry out compression processing on the medical image data, the data processing capability is high, and the characteristics of the medical image data can be extracted rapidly and accurately; the self-supervision image characterization model and the improved deep learning algorithm model are adopted to process the medical image data, so that the accuracy and the accuracy of data processing can be effectively improved, and the diagnosis and the treatment of a doctor on a patient are facilitated; safety encryption algorithm and authority control mechanism are adopted to protect safety in the transmission and storage processes of medical image data, so that potential safety hazards such as data leakage and tampering are avoided; the cloud cache server is used for carrying out distributed cache processing, so that medical image data can be cached rapidly and accurately, and the cache efficiency is improved; the cache management model is used for managing cache data, including index establishment, cache updating, cleaning and the like, so that the cache data can be efficiently and stably managed, and the data management efficiency is improved. Meanwhile, the image classification optimizing model is used for carrying out iterative optimization on the classification recognition result, so that the accuracy and stability of the classification recognition result are improved; and the automation degree and the intelligent degree are high.
The invention adopts the following technical scheme:
a medical image caching processing method based on big data processing comprises the following steps:
firstly, acquiring and preprocessing image data, namely reading and acquiring medical image data through medical image equipment, and performing format conversion, image enhancement, noise removal and standardization operation on the input medical image data through an image processing server GPU, wherein the medical image data comprises X-ray films, CT scanning images, MRI magnetic resonance imaging and ultrasonic imaging;
Step two, image classification marking and segmentation processing, namely performing classification marking on image types, case types, parts and disease severity of medical image data through a self-supervision image characterization model, and performing image classification segmentation through a multi-attention mechanism, wherein the self-supervision image characterization model comprises a feature extractor, a decoder, a loss function unit, a contrast learning unit and a classifier, the contrast learning unit performs iterative optimization on classification recognition results through an image classification optimization model, and the multi-attention mechanism captures attention weighting of different areas in an image through a local search and global calculation mode;
Performing image compression processing, namely performing distortion-free compression processing on the medical image data of small blocks subjected to classification marking and segmentation processing through an improved deep learning algorithm model, wherein the improved deep learning algorithm model comprises an adaptive quantization unit, a sparse entropy coder, an estimator, a wavelet transformation unit and a compression decoder, the adaptive quantization unit dynamically quantizes and adjusts the pixel points of the medical image data of the small blocks according to the characteristics and the pixel point value range of the medical image data of the small blocks, the sparse entropy coder performs predictive coding and differential coding on the pixel points of the medical image data of the small blocks through entropy coding, the estimator estimates errors between input data and quantized data through calculating mean square errors of signals before and after quantization, the wavelet transformation unit optimizes the compression efficiency of the medical image data through high-efficiency wavelet transformation, the compression decoder decodes the compressed data into bit streams through decompression and restores original data, and the output end of the adaptive quantization unit is connected with the sparse entropy coder and the sparse entropy coder, and the output end of the adaptive quantization unit is connected with the input end of the sparse entropy coder and the wavelet transformation unit;
Step four, accelerating data processing in real time, namely accelerating uploading, processing and retrieving of medical images through a full-connection multi-source accelerating network, wherein the full-connection multi-source accelerating network adopts a forwarding mode defined by a four-layer TCP/UDP transmission protocol and an eight-layer HTTP/HTTPS cache protocol monitoring route type, and distributes image traffic to a rear-end server cluster for processing and retrieving based on the forwarding mode;
Step five, performing image distributed caching, namely performing image distributed caching through a cloud cache server, wherein the cloud cache server automatically adjusts the caching distribution of the medical image data by adopting a self-adaptive dynamic balance model, and ensures the data consistency of distributed computer nodes through a data synchronization and backup mechanism, and the cloud cache server performs secure encryption in the medical image data transmission and storage process through a secure encryption algorithm and a permission control mechanism;
And step six, cache data management, namely performing cache data management through a cache management model, wherein the cache management model comprises an index building unit, a cache updating unit and a cache cleaning unit, and the index building unit, the cache updating unit and the cache cleaning unit work in parallel.
According to the invention, the characteristic extractor extracts the characteristic of the medical image data through a convolutional neural network, the decoder performs characteristic mapping and restoration to an original image through a deconvolution neural network, the parameters of the convolutional neural network are updated through back propagation, the loss function unit adopts contrast loss or triple loss to minimize the distance between the medical image data of the same type and maximize the distance between the medical image data of different types, the contrast learning unit realizes data classification and identification through encoding the medical image data of the same type into similar characteristic vectors and encoding the characteristic vectors of the medical image data of different types into dissimilar characteristic vectors, iterative optimization is performed on classification identification results through an image classification optimization model, the classifier adopts an activation function to obtain class probability distribution of the medical image data, the cross entropy loss is minimized to optimize the parameters of the convolutional neural network, the classifier realizes classification identification of the medical image data through a multi-classification network structure, the output end of the characteristic extractor is connected with the input end of the decoder, the output end of the decoder is connected with the input end of the input function unit of the decoder, and the input end of the input function unit is connected with the input end of the input function learning unit.
As a further technical scheme of the invention, the image classification optimization model obtains a feature vector by calculating a second-order information matrix and feature value decomposition of the medical image data, and the calculation formula of the second-order information matrix is as follows:
In formula (1), G represents a second-order information matrix, N is the number of the medical image data, x i is the ith medical image data, u is the classification feature mean value of the medical image data, (x i+u)T represents the transposition of the ith medical image data and the classification feature mean value of the medical image data, T represents the transposition operation, and the second-order information matrix is subjected to feature value decomposition to obtain feature values and feature vectors, and the calculation formula is as follows:
V∧VT＝diag(λ1,λ2,...,λp) (2)
In formula (2), diag represents a diagonal matrix, p is a dimension of a feature vector, λ 1,λ2,...,λp is a corresponding feature value in the feature vector, V represents the feature vector, V T represents a transpose of the feature vector, the feature vector V is used as an orthogonal transformation matrix, and the medical image data is mapped into a new low-dimensional space, where the formula is:
In formula (3), y i represents a low-dimensional spatial feature quantity mapped by the ith medical image data; the image classification optimization model obtains classification results based on dimensional space reconstruction and feature mapping, and a calculation formula is as follows:
In the formula (4), X i represents the dimension space reconstruction and feature mapping classification result of the ith medical image data, and τ represents the time-frequency feature of the ith medical image data; comparing and analyzing the classification result with the real classification result, wherein the calculation formula is as follows:
In the formula (5), F i represents the i-th real classification result of the medical image data, δ represents the degree of difference between the calculated classification result and the real classification result, and when the degree of difference exceeds the error threshold, iterative optimization is continued.
As a further technical scheme of the invention, the self-adaptive dynamic balancing model is based on the automatic allocation of a dynamic load balancing strategy and a dynamic routing mechanism and the automatic expansion of data nodes, and the work of the self-adaptive dynamic balancing model comprises the following steps:
S1, monitoring the load state of a data node, monitoring the running state of the data node through a task manager and a performance monitor, and inputting monitoring data into the self-adaptive dynamic balance model, wherein the running state comprises memory occupancy rate, disk occupancy rate, CPU utilization rate, node load and network bandwidth;
S2, dynamically adjusting the data nodes, wherein the self-adaptive dynamic balancing model dynamically adjusts the data nodes by adopting a dynamic load balancing strategy, the dynamic load balancing strategy balances the data cache loads among the data nodes by a polling algorithm, and the self-adaptive adjustment of node faults is carried out by increasing the task quantity of normal nodes;
S3, dynamically expanding the data nodes, and automatically expanding the data nodes according to actual load conditions and an automatic expansion mechanism, wherein the automatic expansion mechanism realizes high concurrency or large-scale data processing by adding the data nodes;
And S4, a dynamic routing mechanism, wherein the self-adaptive dynamic balancing model realizes automatic allocation of data cache based on the dynamic routing mechanism, and the dynamic routing mechanism selects a shortest path or an optimal path through an intelligent routing algorithm to perform data transmission of a complex scene of a network topology structure.
As a further technical scheme of the invention, the security encryption algorithm adopts a symmetric encryption algorithm to ensure the security and confidentiality of data, and encrypts a data transmission channel through an SSL/TLS protocol, and the authority control mechanism records the access relation between a user and resources through an access control matrix and realizes authority control by dividing the user into doctor, nurse, patient and manager roles.
As a further technical scheme of the invention, the index establishing unit establishes a quick retrieval index through a hash table and a multi-path search data structure, dynamically adjusts the index structure based on the access frequency and access evaluation of data, the cache updating unit manages the updating process of cache data through an increment updating mode and a concurrency control strategy, realizes data updating synchronization among distributed cache nodes through a multi-copy consistency protocol, and the cache cleaning unit records the called times of cache data items through an access counter and triggers a clearing operation through a clock circuit.
Has the positive beneficial effects that:
The invention discloses a medical image caching method based on big data processing, which optimizes the traditional medical image caching method in such modes as classifying and marking medical images by adopting a self-supervision image characterization model, compressing medical images by adopting an improved deep learning algorithm model, carrying out distributed caching by a cloud cache server, carrying out data management by a cache management model and the like, and has higher efficiency, more accuracy and safer; the self-supervision image characterization model is adopted to carry out classification and marking processing on the medical image data, the improved deep learning algorithm model is adopted to carry out compression processing on the medical image data, the data processing capability is high, and the characteristics of the medical image data can be extracted rapidly and accurately; the self-supervision image characterization model and the improved deep learning algorithm model are adopted to process the medical image data, so that the accuracy and the accuracy of data processing can be effectively improved, and the diagnosis and the treatment of a doctor on a patient are facilitated; safety encryption algorithm and authority control mechanism are adopted to protect safety in the transmission and storage processes of medical image data, so that potential safety hazards such as data leakage and tampering are avoided; the cloud cache server is used for carrying out distributed cache processing, so that medical image data can be cached rapidly and accurately, and the cache efficiency is improved; the cache management model is used for managing cache data, including index establishment, cache updating, cleaning and the like, so that the cache data can be efficiently and stably managed, and the data management efficiency is improved. Meanwhile, the image classification optimizing model is used for carrying out iterative optimization on the classification recognition result, so that the accuracy and stability of the classification recognition result are improved; and the automation degree and the intelligent degree are high.
Drawings
FIG. 1 is a flow chart of a medical image caching method based on big data processing according to the present invention;
FIG. 2 is a block diagram of a self-monitoring image characterization model in a medical image caching method based on big data processing;
FIG. 3 is a diagram showing the model architecture of an improved deep learning algorithm model in a medical image caching method based on big data processing
FIG. 4 is a diagram of a cache management model in a medical image cache processing method based on big data processing according to the present invention;
fig. 5 is a schematic flow chart of a self-adaptive dynamic balance model in a medical image caching method based on big data processing.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. 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.
1-4, A medical image caching method based on big data processing comprises the following steps:
firstly, acquiring and preprocessing image data, namely reading and acquiring medical image data through medical image equipment, and performing format conversion, image enhancement, noise removal and standardization operation on the input medical image data through an image processing server GPU, wherein the medical image data comprises X-ray films, CT scanning images, MRI magnetic resonance imaging and ultrasonic imaging;
Step two, image classification marking and segmentation processing, namely performing classification marking on image types, case types, parts and disease severity of medical image data through a self-supervision image characterization model, and performing image classification segmentation through a multi-attention mechanism, wherein the self-supervision image characterization model comprises a feature extractor, a decoder, a loss function unit, a contrast learning unit and a classifier, the contrast learning unit performs iterative optimization on classification recognition results through an image classification optimization model, and the multi-attention mechanism captures attention weighting of different areas in an image through a local search and global calculation mode;
Performing image compression processing, namely performing distortion-free compression processing on the medical image data of small blocks subjected to classification marking and segmentation processing through an improved deep learning algorithm model, wherein the improved deep learning algorithm model comprises an adaptive quantization unit, a sparse entropy coder, an estimator, a wavelet transformation unit and a compression decoder, the adaptive quantization unit dynamically quantizes and adjusts the pixel points of the medical image data of the small blocks according to the characteristics and the pixel point value range of the medical image data of the small blocks, the sparse entropy coder performs predictive coding and differential coding on the pixel points of the medical image data of the small blocks through entropy coding, the estimator estimates errors between input data and quantized data through calculating mean square errors of signals before and after quantization, the wavelet transformation unit optimizes the compression efficiency of the medical image data through high-efficiency wavelet transformation, the compression decoder decodes the compressed data into bit streams through decompression and restores original data, and the output end of the adaptive quantization unit is connected with the sparse entropy coder and the sparse entropy coder, and the output end of the adaptive quantization unit is connected with the input end of the sparse entropy coder and the wavelet transformation unit;
Step four, accelerating data processing in real time, namely accelerating uploading, processing and retrieving of medical images through a full-connection multi-source accelerating network, wherein the full-connection multi-source accelerating network adopts a forwarding mode defined by a four-layer TCP/UDP transmission protocol and an eight-layer HTTP/HTTPS cache protocol monitoring route type, and distributes image traffic to a rear-end server cluster for processing and retrieving based on the forwarding mode;
Step five, performing image distributed caching, namely performing image distributed caching through a cloud cache server, wherein the cloud cache server automatically adjusts the caching distribution of the medical image data by adopting a self-adaptive dynamic balance model, and ensures the data consistency of distributed computer nodes through a data synchronization and backup mechanism, and the cloud cache server performs secure encryption in the medical image data transmission and storage process through a secure encryption algorithm and a permission control mechanism;
And step six, cache data management, namely performing cache data management through a cache management model, wherein the cache management model comprises an index building unit, a cache updating unit and a cache cleaning unit, and the index building unit, the cache updating unit and the cache cleaning unit work in parallel.
In the above embodiment, the feature extractor extracts features of the medical image data through a convolutional neural network, the decoder performs feature mapping and restoration to an original image through a deconvolution neural network, updates parameters of the convolutional neural network through back propagation, the loss function unit minimizes a distance between the medical image data of the same type and maximizes a distance between the medical image data of different types through a multi-classification network structure, the contrast learning unit realizes data classification and identification through encoding the medical image data of the same type into similar feature vectors and encoding the feature vectors of the medical image data of different types into dissimilar feature vectors, performs iterative optimization on classification identification results through an image classification optimization model, the classifier obtains class probability distribution of the medical image data through an activation function, optimizes parameters of the convolutional neural network through minimizing cross entropy loss, the classifier realizes classification identification of the medical image data through a multi-classification network structure, an output end of the feature extractor is connected with an input end of the decoder, an output end of the feature extractor is connected with an input end of the decoder, and an output end of the classifier is connected with an input end of the learning unit of the contrast learning unit.
In a specific embodiment, the feature extractor performs feature extraction through a convolutional neural network, and generally adopts a combination of components such as a convolutional layer, a pooling layer, an activation function and the like. The convolution layer can learn information such as characteristics, textures and the like of different positions in the image, the pooling layer can reduce the dimension of the characteristic diagram, strengthen the perception of local characteristics, the activation function can add nonlinearity to the output, and the fitting capacity of the model is improved. The medical image data can be converted into the high-dimensional vector with abstract characteristics through the convolutional neural network, so that the subsequent processing is convenient. The decoder restores the feature map by deconvoluting the neural network, thereby reconstructing the original image. Deconvolution neural networks are similar to convolutional neural networks, but have opposite layers and structures to convolutional neural networks, so that feature vectors can be restored to original medical image data. Through deconvolution neural network, can realize functions such as denoising, compression and feature extraction of image. The loss function unit adopts contrast loss or triplet loss, and aims to make the distance between medical image data of the same category as small as possible and the distance between medical image data of different categories as large as possible, thereby realizing data classification and identification. The contrast loss is to compare the data of the same category with the data of different categories, and train the model by minimizing the distance between the data of the same category and maximizing the distance between the data of different categories. The triple loss is to compare the data of the same category with the other two data selected randomly, and train the model by minimizing the distance from the data of the same category to the data of the other same category and maximizing the distance from the data of the same category to the data of the different categories. The contrast learning unit encodes the medical image data of the same category into similar feature vectors, and encodes the data of different categories into dissimilar feature vectors, so that data classification and recognition are realized. Similar feature vectors can be clustered together, and dissimilar feature vectors can be dispersed together, thereby achieving clustering and classification of data. The contrast learning unit may be implemented by adding special layers or modules to the convolutional neural network. And carrying out iterative optimization on the medical image data through an image classification optimization model, and carrying out training data, verification data and test data.
In the above embodiment, the image classification optimization model obtains the feature vector by calculating a second-order information matrix and feature value decomposition of the medical image data, and a calculation formula of the second-order information matrix is:
in formula (1), G represents a second-order information matrix, N is the number of the medical image data, x i is the ith medical image data, u is the classification feature mean value of the medical image data, (x i+u)T represents the transposition of the ith medical image data and the classification feature mean value of the medical image data, T represents the transposition operation, and the feature value and the feature vector are obtained by performing feature value decomposition on the second-order information matrix, wherein the calculation formula is as follows:
V∧VT＝diag(λ1,λ2,...,λp)(2)
In formula (2), aiag denotes a diagonal matrix, p is a dimension of a feature vector, λ 1,λ2,...,λp is a corresponding feature value in the feature vector, V denotes a feature vector, V T denotes a transpose of the feature vector, the feature vector V is used as an orthogonal transformation matrix, and the medical image data is mapped into a new low-dimensional space, where the formula is:
In formula (3), y i represents a low-dimensional spatial feature quantity mapped by the ith medical image data; the image classification optimization model obtains classification results based on dimensional space reconstruction and feature mapping, and a calculation formula is as follows:
In the formula (4), X i represents the dimension space reconstruction and feature mapping classification result of the ith medical image data, and τ represents the time-frequency feature of the ith medical image data; comparing and analyzing the classification result with the real classification result, wherein the calculation formula is as follows:
In the formula (5), F i represents the i-th real classification result of the medical image data, δ represents the degree of difference between the calculated classification result and the real classification result, and when the degree of difference exceeds the error threshold, iterative optimization is continued.
In a specific embodiment, the image classification optimization model is a technical scheme for mapping medical image data into a new low-dimensional space and classifying by calculating time-frequency characteristics. Second-order information matrix calculation: and calculating a second-order information matrix according to the characteristics of the medical image data. Characteristic value decomposition: and decomposing the characteristic value of the obtained second-order information matrix to obtain a characteristic vector and a corresponding characteristic value. Mapping to a new low-dimensional space: the obtained eigenvectors are used as orthogonal transformation matrixes, and the medical image data are mapped into a new low-dimensional space. Acquiring time-frequency characteristics: and acquiring time-frequency characteristics based on dimensional space reconstruction and characteristic mapping. And (3) calculating a classification result: and calculating a classification result by using the obtained time-frequency characteristics. Model optimization: and comparing the difference degree between the actual classification result and the simulation classification result, and performing self-adaptive optimization according to an error threshold value to improve the classification effect. In summary, the image classification optimization model is a technical scheme for classifying by mapping medical image data into a new low-dimensional space and calculating time-frequency characteristics. The model not only can improve the classification accuracy and stability, but also can be optimized in a self-adaptive way, so that the classification effect of the model is further improved. The hardware working environment of the image classification optimization model comprises the following:
1. high performance computer: for processing a number of complex computing tasks including data preprocessing, feature extraction, feature mapping, etc.
2. Graphics processor: the method is used for accelerating graphic operation and complex parallel calculation, and improves the calculation efficiency of the model.
3. A data storage device: for storing large-scale medical image data and the large amount of data required for model training.
4. Monitor and display device: the method is used for displaying medical images and model calculation results.
5. Network equipment: for the transmission of data and networking of models.
6. Other auxiliary devices: such as a keyboard, mouse, printer, etc.
A certain amount of medical image data is collected, including CT images and MRI images, and the collected medical image data is randomly divided into two groups, one group for training a model and one group for testing a model. The training data is input into an image classification optimization model (A group) and a traditional neural network classification method (B group), feature vectors are calculated, feature mapping is carried out, classification results are obtained, and comparison analysis is carried out with actual classification results. And respectively, analyzing according to experimental results, comparing the classification accuracy and stability of various methods, evaluating the effectiveness and superiority of the image classification optimization model, repeating five groups of experiments, and recording in table 1.
Table 1 results statistics table
According to experimental result analysis, the image classification optimization model has higher classification accuracy and higher classification stability compared with the traditional neural network classification method. The five groups of experiments have the average classification accuracy of 90 percent and the average classification stability of high level, prove the effectiveness and the superiority of the image classification optimization model, and can have wide application prospects in the fields of classification, diagnosis, treatment and the like of medical image data.
In the above embodiment, as shown in fig. 5, the adaptive dynamic balancing model is based on a dynamic load balancing policy and a dynamic routing mechanism to automatically allocate and automatically expand data nodes, and the operation of the adaptive dynamic balancing model includes the following steps:
S1, monitoring the load state of a data node, monitoring the running state of the data node through a task manager and a performance monitor, and inputting monitoring data into the self-adaptive dynamic balance model, wherein the running state comprises memory occupancy rate, disk occupancy rate, CPU utilization rate, node load and network bandwidth;
S2, dynamically adjusting the data nodes, wherein the self-adaptive dynamic balancing model dynamically adjusts the data nodes by adopting a dynamic load balancing strategy, the dynamic load balancing strategy balances the data cache loads among the data nodes by a polling algorithm, and the self-adaptive adjustment of node faults is carried out by increasing the task quantity of normal nodes;
S3, dynamically expanding the data nodes, and automatically expanding the data nodes according to actual load conditions and an automatic expansion mechanism, wherein the automatic expansion mechanism realizes high concurrency or large-scale data processing by adding the data nodes;
And S4, a dynamic routing mechanism, wherein the self-adaptive dynamic balancing model realizes automatic allocation of data cache based on the dynamic routing mechanism, and the dynamic routing mechanism selects a shortest path or an optimal path through an intelligent routing algorithm to perform data transmission of a complex scene of a network topology structure.
In a specific embodiment, the adaptive dynamic equalization model: in the architecture of the cloud cache server, an adaptive dynamic balance model is required to be adopted to realize the cache distribution and dynamic adjustment of medical image data. Wherein, a load balancing algorithm and a hash algorithm can be used to realize the distributed storage and access of data, thereby improving the performance and the expandability of the system.
In the above embodiment, the security encryption algorithm adopts a symmetric encryption algorithm to ensure the security and confidentiality of data, encrypts the data transmission channel through SSL/TLS protocol, and the authority control mechanism records the access relationship between the user and the resource through the access control matrix, and realizes authority control by dividing the user into doctor, nurse, patient and manager roles.
In a specific embodiment, the security encryption algorithm adopts a symmetric encryption algorithm to ensure the security and confidentiality of data, and the specific steps are as follows: symmetric encryption algorithms are a technique for converting plaintext into ciphertext, using the same key for encryption and decryption. Thus, the sender and the receiver need to share the same key at the time of data transmission. Common symmetric encryption algorithms are AES, DES, 3DES, etc. The algorithm has the characteristics of high calculation speed, high encryption and decryption efficiency and the like, and can effectively protect the confidentiality of data. In practical applications, to further enhance data security, SSL/TLS protocols may be used to encrypt the data transmission channel. The protocol can establish a secure channel on the network and ensure the identity authentication and information integrity between the two communication parties through a public key encryption technology. The authority control mechanism records the access relation between the user and the resource through the access control matrix, and realizes the authority control by dividing the user into doctor, nurse, patient and manager roles, and the specific steps are as follows: an access control matrix is a data structure that records the access relationship between users and resources. The matrix contains two dimensions: a user ID and a resource ID, and marks whether the user has access to the resource. The users are assigned corresponding rights according to their roles (e.g., doctor, nurse, patient, and administrator). For example, a doctor can view and modify patient medical record information, while a nurse can only view medical record information. When a user accesses a resource, the system determines whether the user has access to the resource according to the access control matrix. If yes, allowing the user to operate; otherwise, the access request is denied. To further enhance system security, other auxiliary measures may be employed, such as identity authentication, session management, log auditing, etc.
In the above embodiment, the index establishing unit establishes the fast search index through the hash table and the multi-path search data structure, dynamically adjusts the index structure based on the access frequency and the access evaluation of the data, the cache updating unit manages the updating process of the cache data through an incremental updating mode and a concurrency control strategy, realizes the data updating synchronization between distributed cache nodes through a multi-copy consistency protocol, and the cache cleaning unit records the number of times the cache data item is called through the access counter, and triggers the cleaning operation through a clock circuit.
In a specific embodiment, the index establishing unit establishes a quick search index through a hash table and a multi-path search data structure, and the specific steps are as follows: and selecting a proper hash function according to the data characteristics and the requirements, and mapping the data into a hash table. In order to improve the retrieval efficiency, multiple search trees (such as b+ tree, B tree, etc.) may be used to optimize the hash table. This reduces the occurrence of hash collisions and enables fast lookups between different nodes. The index structure is dynamically adjusted based on the access frequency and access rating of the data. For example, when a certain data item is frequently accessed, it can be moved to a higher level search tree to improve retrieval efficiency; when a certain data item is not accessed for a longer time, it can be moved into a low-level search tree to reduce memory usage. The cache updating unit manages the updating process of the cache data through an incremental updating mode and a concurrency control strategy, and realizes the data updating synchronization among distributed cache nodes through a multi-copy consistency protocol, and the method comprises the following specific steps: with incremental updating, only the part that needs to be modified is updated instead of the entire cache block. This reduces network transmission overhead and CPU resource consumption. In order to avoid the problem of inconsistent data caused by that a plurality of clients modify the data of the same cache block at the same time, a concurrency control strategy (such as a read-write lock, a mutual exclusion lock and the like) is adopted to protect the cache block. Data update synchronization between distributed cache nodes is achieved through multi-copy consistency protocols (such as Paxos algorithm, raft algorithm, etc.). Thus, even if one node is down, the data can be restored through other nodes. The cache cleaning unit records the called times of the cache data item through the access counter, and adopts a clock circuit to trigger the cleaning operation, and the specific steps are as follows: an access counter is set for each cache block and incremented by 1 at each access. When the memory space is insufficient, judging which cache blocks are hot spots and cold doors according to the access counter, the latest use time and other factors, and removing the cold door cache blocks from the memory to release the space. A clock circuit is used to trigger the purge operation. Specifically, all cache blocks are ordered by most recently used time and one pointer is pointed to the oldest cache block. When a certain "cold gate" cache block needs to be moved out, the cache block is scanned from the pointer position, and the first unused cache block is found and moved out. At the same time, the pointer is pointed to the next cache block for the next cleaning operation.
While specific embodiments of the present invention have been described above, it will be understood by those skilled in the art that these specific embodiments are by way of example only, and that various omissions, substitutions, and changes in the form and details of the methods and systems described above may be made by those skilled in the art without departing from the spirit and scope of the invention. For example, it is within the scope of the present invention to combine the above-described method steps to perform substantially the same function in substantially the same way to achieve substantially the same result. Accordingly, the scope of the invention is limited only by the following claims.
Claims (5)
1. A medical image caching method based on big data processing is characterized in that: the method comprises the following steps:
firstly, acquiring and preprocessing image data, namely reading and acquiring medical image data through medical image equipment, and performing format conversion, image enhancement, noise removal and standardization operation on the input medical image data through an image processing server GPU, wherein the medical image data comprises X-ray films, CT scanning images, MRI magnetic resonance imaging and ultrasonic imaging;
Step two, image classification marking and segmentation processing, namely performing classification marking on image types, case types, parts and disease severity of medical image data through a self-supervision image characterization model, and performing image classification segmentation through a multi-attention mechanism, wherein the self-supervision image characterization model comprises a feature extractor, a decoder, a loss function unit, a contrast learning unit and a classifier, the contrast learning unit performs iterative optimization on classification recognition results through an image classification optimization model, and the multi-attention mechanism captures attention weighting of different areas in an image through a local search and global calculation mode;
Performing image compression processing, namely performing distortion-free compression processing on the medical image data of small blocks subjected to classification marking and segmentation processing through an improved deep learning algorithm model, wherein the improved deep learning algorithm model comprises an adaptive quantization unit, a sparse entropy coder, an estimator, a wavelet transformation unit and a compression decoder, the adaptive quantization unit dynamically quantizes and adjusts the pixel points of the medical image data of the small blocks according to the characteristics and the pixel point value range of the medical image data of the small blocks, the sparse entropy coder performs predictive coding and differential coding on the pixel points of the medical image data of the small blocks through entropy coding, the estimator estimates errors between input data and quantized data through calculating mean square errors of signals before and after quantization, the wavelet transformation unit optimizes the compression efficiency of the medical image data through high-efficiency wavelet transformation, the compression decoder decodes the compressed data into bit streams through decompression and restores original data, and the output end of the adaptive quantization unit is connected with the sparse entropy coder and the sparse entropy coder, and the output end of the adaptive quantization unit is connected with the input end of the sparse entropy coder and the wavelet transformation unit;
Step four, accelerating data processing in real time, namely accelerating uploading, processing and retrieving of medical images through a full-connection multi-source accelerating network, wherein the full-connection multi-source accelerating network adopts a forwarding mode defined by a four-layer TCP/UDP transmission protocol and an eight-layer HTTP/HTTPS cache protocol monitoring route type, and distributes image traffic to a rear-end server cluster for processing and retrieving based on the forwarding mode;
Step five, performing image distributed caching, namely performing image distributed caching through a cloud cache server, wherein the cloud cache server automatically adjusts the caching distribution of the medical image data by adopting a self-adaptive dynamic balance model, and ensures the data consistency of distributed computer nodes through a data synchronization and backup mechanism, and the cloud cache server performs secure encryption in the medical image data transmission and storage process through a secure encryption algorithm and a permission control mechanism;
Step six, cache data management, namely cache data management is carried out through a cache management model, wherein the cache management model comprises an index building unit, a cache updating unit and a cache cleaning unit, and the index building unit, the cache updating unit and the cache cleaning unit work in parallel;
The feature extractor extracts the features of the medical image data through a convolutional neural network, the decoder performs feature mapping and restoration to an original image through a deconvolution neural network, and updates parameters of the convolutional neural network through back propagation, the loss function unit adopts contrast loss or triple loss to minimize the distance between the medical image data of the same type and maximize the distance between the medical image data of different types, the contrast learning unit realizes data classification and identification by encoding the medical image data of the same type into similar feature vectors and encoding the feature vectors of the medical image data of different types into dissimilar feature vectors, the classifier adopts an activation function to obtain category probability distribution of the medical image data, optimizes the parameters of the convolutional neural network through minimized cross entropy loss, the classifier realizes classification and identification of the medical image data through a multi-classification network structure, the output end of the feature extractor is connected with the input end of the decoder, the output end of the decoder is connected with the input end of the contrast learning unit, and the output end of the contrast learning unit is connected with the input end of the contrast learning unit;
The feature extractor performs feature extraction through a convolutional neural network, adopts a convolutional layer, a pooling layer and an activation function component combination, the convolutional layer learns the feature and texture information of different positions in an image, the pooling layer reduces the dimension of a feature map, strengthens the perception of local features, adds nonlinearity to output by an activation function, improves the fitting capacity of a model, converts medical image data into a high-dimensional vector with abstract features through the convolutional neural network, restores the feature mapping through the deconvolution neural network so as to reconstruct an original image, the deconvolution neural network is similar to the convolutional neural network, but the layer number and the structure are opposite to the convolutional neural network, restores the feature vector into original medical image data, realizes denoising, compression and feature extraction of the image through the deconvolution neural network, the loss function unit adopts contrast loss or triplet loss, the distance between medical image data of different categories is as large as possible, the contrast loss is that the data of the same category is compared with the data of different categories, training vectors are trained by minimizing the distance between the data of the same category data and the data of different categories, the feature loss is similar to the medical image data of different categories by selecting the random data and the feature loss is similar to the feature vector of different categories through the feature information, the feature information is similar to the feature vector of the data by comparing the different categories of the data of different categories through the feature models, the feature loss is similar to the feature vector of the feature information by the feature information, therefore, the clustering and classification of the data are realized, the comparison learning unit is realized by adding a plurality of special layers or modules in the convolutional neural network, the iterative optimization is carried out on the medical image data through the image classification optimization model, and the training data, the verification data and the test data are used for carrying out the iterative optimization.
2. The medical image caching method based on big data processing according to claim 1, wherein the method comprises the following steps: the image classification optimization model obtains a feature vector by calculating a second-order information matrix and feature value decomposition of the medical image data, and the calculation formula of the second-order information matrix is as follows:
In the case of the formula (1), Representing a matrix of information of the second order,The number of the medical image data,For the ith of the medical image data,The mean value of the classification characteristics of the medical image data,And expressing the transposition of the ith medical image data and the classification characteristic mean value of the medical image data, wherein T expresses transposition operation, and carrying out eigenvalue decomposition on the second-order information matrix to obtain eigenvalues and eigenvectors, wherein the calculation formula is as follows:
In the formula (2) of the present invention, A diagonal matrix is represented and,As the dimension of the feature vector,For the corresponding feature value in the feature vector,The feature vector is represented by a vector of features,Representing the transpose of the feature vector, taking the feature vectorAs an orthogonal transformation matrix, and map the medical image data into a new low-dimensional space, the formula is:
In the formula (3) of the present invention, Represent the firstLow-dimensional space feature values mapped by the medical image data; the image classification optimization model obtains classification results based on dimensional space reconstruction and feature mapping, and a calculation formula is as follows:
In the formula (4) of the present invention, Represent the firstThe dimension space reconstruction and feature mapping classification result of the medical image data,Represent the firstTime-frequency characteristics of the medical image data; comparing and analyzing the classification result with the real classification result, wherein the calculation formula is as follows:
in the formula (5) of the present invention, Represent the firstThe true classification result of the medical image data,And (3) representing the degree of difference between the calculated classification result and the real classification result, and continuing to perform iterative optimization when the degree of difference exceeds an error threshold.
3. The medical image caching method based on big data processing according to claim 1, wherein the method comprises the following steps: the self-adaptive dynamic balancing model is based on the automatic allocation of a dynamic load balancing strategy and a dynamic routing mechanism and the automatic expansion of data nodes, and the work of the self-adaptive dynamic balancing model comprises the following steps:
S1, monitoring the load state of a data node, monitoring the running state of the data node through a task manager and a performance monitor, and inputting monitoring data into the self-adaptive dynamic balance model, wherein the running state comprises memory occupancy rate, disk occupancy rate, CPU utilization rate, node load and network bandwidth;
S2, dynamically adjusting the data nodes, wherein the self-adaptive dynamic balancing model dynamically adjusts the data nodes by adopting a dynamic load balancing strategy, the dynamic load balancing strategy balances the data cache loads among the data nodes by a polling algorithm, and the self-adaptive adjustment of node faults is carried out by increasing the task quantity of normal nodes;
S3, dynamically expanding the data nodes, and automatically expanding the data nodes according to actual load conditions and an automatic expansion mechanism, wherein the automatic expansion mechanism realizes high concurrency or large-scale data processing by adding the data nodes;
And S4, a dynamic routing mechanism, wherein the self-adaptive dynamic balancing model realizes automatic allocation of data cache based on the dynamic routing mechanism, and the dynamic routing mechanism selects a shortest path or an optimal path through an intelligent routing algorithm to perform data transmission of a complex scene of a network topology structure.
4. The medical image caching method based on big data processing according to claim 1, wherein the method comprises the following steps: the security encryption algorithm ensures the security and confidentiality of data by adopting a symmetric encryption algorithm, the data transmission channel is encrypted by SSL/TLS protocol, the authority control mechanism records the access relation between the user and the resource by an access control matrix, and the authority control is realized by dividing the user into doctor, nurse, patient and manager roles.
5. The medical image caching method based on big data processing according to claim 1, wherein the method comprises the following steps: the index establishing unit establishes a quick search index through a hash table and a multipath search data structure, dynamically adjusts the index structure based on the access frequency and access evaluation of data, the cache updating unit manages the updating process of cache data through an increment updating mode and a concurrency control strategy, realizes data updating synchronization among distributed cache nodes through a multi-copy consistency protocol, and the cache cleaning unit records the called times of cache data items through an access counter and triggers a clearing operation through a clock circuit.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410232171.0A CN118116554B (en) | 2024-03-01 | 2024-03-01 | Medical image caching processing method based on big data processing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410232171.0A CN118116554B (en) | 2024-03-01 | 2024-03-01 | Medical image caching processing method based on big data processing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118116554A CN118116554A (en) | 2024-05-31 |
| CN118116554B true CN118116554B (en) | 2024-08-27 |
Family
ID=91211672
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410232171.0A Active CN118116554B (en) | 2024-03-01 | 2024-03-01 | Medical image caching processing method based on big data processing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118116554B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118381859B (en) * | 2024-06-25 | 2024-08-23 | 成都科玛奇信息科技有限责任公司 | Medical image data transmission method |
| CN120639597B (en) * | 2025-08-15 | 2025-12-05 | 国网智能科技股份有限公司 | Transformer substation secondary system loop anomaly analysis and positioning method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110543364A (en) * | 2019-07-21 | 2019-12-06 | 聊城市光明医院 | Medical image rapid loading method and system |
| CN117611926A (en) * | 2024-01-22 | 2024-02-27 | 重庆医科大学绍兴柯桥医学检验技术研究中心 | A medical image recognition method and system based on AI model |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH08214308A (en) * | 1995-02-06 | 1996-08-20 | Matsushita Graphic Commun Syst Inc | Image compression encoder and image expansion decoder |
| US8760453B2 (en) * | 2010-09-01 | 2014-06-24 | Microsoft Corporation | Adaptive grid generation for improved caching and image classification |
| US9060032B2 (en) * | 2010-11-01 | 2015-06-16 | Seven Networks, Inc. | Selective data compression by a distributed traffic management system to reduce mobile data traffic and signaling traffic |
| CN113723603B (en) * | 2020-05-26 | 2025-01-21 | 华为技术有限公司 | A method, device and storage medium for updating parameters |
| US12211202B2 (en) * | 2021-10-13 | 2025-01-28 | GE Precision Healthcare LLC | Self-supervised representation learning paradigm for medical images |
| CN116935239A (en) * | 2023-07-18 | 2023-10-24 | 赛思倍斯(绍兴)智能科技有限公司 | On-orbit satellite intelligent detection method based on self-supervision remote sensing image classification |
-
2024
- 2024-03-01 CN CN202410232171.0A patent/CN118116554B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110543364A (en) * | 2019-07-21 | 2019-12-06 | 聊城市光明医院 | Medical image rapid loading method and system |
| CN117611926A (en) * | 2024-01-22 | 2024-02-27 | 重庆医科大学绍兴柯桥医学检验技术研究中心 | A medical image recognition method and system based on AI model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118116554A (en) | 2024-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN118116554B (en) | Medical image caching processing method based on big data processing | |
| Liu et al. | Enhancing the privacy of federated learning with sketching | |
| CN112367167B (en) | Quantum secret sharing method and system based on tensor network state dynamic compression | |
| CN117835246B (en) | Task-oriented privacy semantic communication method | |
| Chehimi et al. | Quantum semantic communications for resource-efficient quantum networking | |
| Akter et al. | Edge intelligence-based privacy protection framework for iot-based smart healthcare systems | |
| Toktas et al. | Parameter optimization of chaotic system using Pareto-based triple objective artificial bee colony algorithm | |
| CN118503685B (en) | Device fingerprint extraction method and system based on device attributes and passive traffic characteristics | |
| Wang et al. | Multi-key spatio-temporal chaotic image encryption and retrieval scheme | |
| Yao et al. | A color image compression and encryption algorithm combining compressed sensing, Sudoku matrix, and hyperchaotic map | |
| Liu et al. | A new image encryption scheme based on block compressive sensing and chaotic laser system for IoT | |
| Liu et al. | To deliver more information in coverless information hiding | |
| Chaker et al. | Color image encryption system based fractional hyperchaotic, fibonacci matrix and quaternion algebra | |
| Tang et al. | Secure and Efficient Image Compression‐Encryption Scheme Using New Chaotic Structure and Compressive Sensing | |
| Yan et al. | Privacy-preserving content-based image retrieval in edge environment | |
| CN120223360A (en) | A network intrusion detection method based on federated graph neural network | |
| CN120301646A (en) | A method for abnormal traffic detection based on sparse matrix | |
| Duan et al. | Efficient federated learning method for cloud-edge network communication | |
| CN119719911A (en) | Encrypted traffic classification method, device, computer equipment and medium | |
| CN117349685A (en) | A communication data clustering method, system, terminal and medium | |
| Optimization-Based | Compressive Sensing for Image Compression and Recovery | |
| CN115329032B (en) | Learning data transmission method, device, equipment and storage medium based on federated dictionary | |
| Kyrychenko et al. | Research on Hybrid Image Storage Models to Ensure Data Security and Privacy | |
| CN120372685B (en) | Privacy protection method and system for segmentation learning based on selective decryption | |
| CN118886065B (en) | Forgetting learning method and device based on distributed conversion and storage medium |
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 |