Multi-client collaborative SAR image target recognition method based on federated learning

Abstract

The invention relates to a multi-client collaborative SAR image target identification method based on federated learning, and the method comprises the steps: collecting SAR image data, carrying out the preprocessing, generating a corresponding target identification tag, and obtaining a data set with a tag; building a federal learning framework, initializing and distributing global model parameters to each client; distributing local private data to each client according to the data set with the label, performing local model training by each client by using the local private data, and uploading updated model parameters to a server; the server aggregates the model parameters uploaded by all the clients, updates global model parameters, and distributes the updated global model parameters to all the clients; repeating the steps until the global model converges; and testing the trained global model, and selecting the tested global model to carry out SAR image target identification. Compared with the prior art, the method has the advantages that the accuracy of the model is effectively maintained while privacy is protected.

Description

Technical Field

[0001] The present invention relates to the application field of remote sensing synthetic aperture radar, and in particular to a multi-client collaborative SAR image target recognition method based on federated learning. Background Art

[0002] Synthetic Aperture Radar (SAR) is capable of performing radar imaging in all weather conditions (such as cloud cover, rainfall, etc.) and during both day and night. With the advent of the big data era, deep learning has become the main method for data analysis, showing good application results in the fields of earth observation and remote sensing.

[0003] The SAR target recognition method based on deep learning uses deep learning to extract features and classify SAR images to achieve high-precision recognition of targets. Compared with traditional methods, the deep learning method has the following advantages: 1. Automatic feature extraction: The deep learning model can automatically extract effective features from the data, greatly reducing dependence on manual work. 2. Strong generalization ability: Through large-scale data training, the deep learning model can learn features with good generalization ability and maintain high recognition performance in different scenarios and environments. 3. Ability to handle complex scenes: The deep learning model can handle complex backgrounds and multi-target scenes, and has stronger robustness and adaptability. Therefore, the SAR target recognition and classification method based on deep learning has important application value in many fields such as military reconnaissance, disaster monitoring, and environmental monitoring.

[0004] However, although the application of deep learning models in SAR target recognition has shown strong recognition capabilities, deep learning models usually rely on a large amount of data training. In the intelligent interpretation of SAR images, due to the high cost and difficulty of data acquisition, a single satellite agency often faces the problem of insufficient data samples, which seriously limits the improvement of model performance and the expansion of application scope.

[0005] In order to improve model performance, collaborative data sharing between multiple satellite agencies has become an effective way. By increasing the number and diversity of samples, the generalization ability of the model can be significantly improved. However, SAR image data contains sensitive geographic location and target information, which involves national security and commercial secrets. Improper handling may lead to serious consequences, making data privacy protection crucial in the data sharing process. When sharing data, each client (such as different satellite agencies) must ensure the privacy and security of the data to prevent the leakage of sensitive information. Therefore, in multi-client collaborative SAR target recognition systems, data privacy protection becomes a key challenge. Summary of the invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a multi-client collaborative SAR image target recognition method based on federated learning. By transferring model parameters rather than original data between the participants, multiple satellite mechanisms can collaboratively train the model without sharing data, effectively ensuring model performance while protecting data privacy, and can gradually approach the performance of centralized training through iterative optimization.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] A multi-client collaborative SAR image target recognition method based on federated learning includes the following steps:

[0009] S1: Collect SAR image data and preprocess it to generate corresponding target recognition labels to obtain labeled data sets;

[0010] S2: Build a federated learning framework, which includes a server and multiple clients participating in collaborative training. Each client shares model parameters with the server; build a neural network model for target recognition, initialize and distribute global model parameters to each client;

[0011] S3: allocating local private data to each client according to the labeled data set, each client using the local private data to perform local model training, and uploading updated model parameters to the server;

[0012] S4: The server aggregates the model parameters uploaded by all clients, updates the global model parameters, and distributes the updated global model parameters to each client;

[0013] S5: Repeat steps S3 and S4 until the global model parameters reach a convergence state, and obtain a trained global model;

[0014] S6: Test the trained global model and select the global model that has passed the test for SAR image target recognition.

[0015] Furthermore, step S2 specifically includes the following steps:

[0016] S201: Establish a semi-decentralized federated learning framework, determine the participating clients and servers, and the number of clients, each client shares the model parameters with the server, and the server aggregates the model parameters of each client to update the global model;

[0017] S202: pre-build a neural network model for target recognition and initialize global model parameters;

[0018] S203: The server distributes the initialized global model parameters to each client.

[0019] Furthermore, step S3 specifically includes the following steps:

[0020] S301: Each client loads the global model parameters transmitted by the server;

[0021] S302: Each client holds a local data set allocated based on the labeled data set. Each client performs local training based on the allocated global model parameters, calculates the loss function and performs gradient update. The corresponding update expression of the local model parameters is:

[0022]

[0023] In the formula, is the local model parameter at the t+1th iteration, is the local model parameter at the tth iteration, η is the learning rate, Based on the model parameters The loss function calculation results under , l is the cross entropy loss function, D i is the i-th local data set, (x i,j ,y i,j ) is the local data and the corresponding target identification label;

[0024] S303: Each client sends the updated local model parameters to the server respectively.

[0025] Furthermore, step S4 specifically includes the following steps:

[0026] S401: The server receives local model parameters transmitted from each client;

[0027] S402: The server aggregates the local model parameters of each client according to a predetermined aggregation algorithm to generate new global model parameters;

[0028] S403: The server updates the global model parameters based on the new global model parameters;

[0029] S404: The server distributes the updated global model parameters to each client.

[0030] Furthermore, the aggregation algorithm is a FedAvg algorithm, which is used to perform a weighted average operation on the model parameters of all clients. The calculation expression of the FedAvg algorithm is:

[0031]

[0032] In the formula, is the global model parameter of the t+1th iteration, N is the number of clients, |Di | is the data volume of the i-th client, is the total amount of data, is the local model parameter of the i-th client at the t-th iteration.

[0033] Furthermore, step S5 specifically includes the following steps:

[0034] S501: The server distributes the updated global model parameters to each client;

[0035] S502: Each client executes step S3 to perform local model training;

[0036] S503: The server executes step S4 to aggregate and update the global model;

[0037] S504: Iteratively execute steps S501-S503 until the global model parameters converge.

[0038] Furthermore, step S1 specifically includes the following steps:

[0039] S101: Collect SAR image data through satellite mechanism;

[0040] S102: preprocessing the collected SAR image data, the preprocessing including denoising, correction and enhancement;

[0041] S103: Generate corresponding target recognition labels according to the preprocessed SAR image data.

[0042] Furthermore, the neural network model for target recognition includes a backbone network for feature extraction and a classifier composed of fully connected layers for classification;

[0043] The backbone network gradually extracts high-level features from the input SAR image through a series of convolutional layers;

[0044] The classifier maps the high-level features extracted by the backbone network to the final output category.

[0045] Furthermore, step S6 specifically includes the following steps:

[0046] S601: Select and prepare a test data set;

[0047] S602: Load the trained global model parameters into the test environment;

[0048] S603: Infer the global model in the test data set to obtain a prediction result, the calculation expression of which is:

[0049]

[0050] In the formula, To predict the results, is the global model parameter, x test is the test data, f(·) is the model function;

[0051] S604: Compare the prediction result with the true label y of the test data set test For comparison, the verification accuracy Acc of the model is calculated. The calculation expression of the accuracy Acc is:

[0052]

[0053] Where D test is the test data set, 1() is the indicator function, when the predicted result is the same as the true label, the value is 1, otherwise it is 0. is the i-th prediction result, y test,i is the i-th true label y test ;

[0054] S605: Evaluate model performance according to the verification accuracy Acc of the model;

[0055] S607: Optimize and improve the global model according to the model performance evaluation results.

[0056] Furthermore, step S605 further includes:

[0057] Record the training loss L under different number of clients train , used to evaluate the learning process of the global model and the convergence speed under different client configurations;

[0058] Record the verification accuracy Acc of the global model in each communication round round , used to evaluate the change and stability of model accuracy during training;

[0059] Record the global model verification accuracy Acc after completing the maximum communication round final , used to measure the overall effectiveness and performance of the training process for different numbers of clients.

[0060] This paper proposes a multi-client collaborative synthetic aperture radar (SAR) image target recognition method based on federated learning, aiming to solve the dual challenges of insufficient number of SAR data samples and data privacy protection. In traditional SAR image intelligent interpretation, due to the high cost and difficulty of data acquisition, a single satellite organization often faces the problem of insufficient number of data samples, which seriously limits the improvement of model performance and the expansion of application scope. At the same time, the sensitive geographic location and target information contained in SAR data make data privacy protection crucial in the data sharing process.

[0061] Compared with the prior art, the present invention has the following advantages:

[0062] (1) The present invention uses federated learning technology to allow multiple clients (such as different satellite organizations) to collaboratively train models without sharing original data. Specifically, each client uses private data to train the model locally, and uploads the updated model parameters to the central server for aggregation to form new global model parameters. In this way, the data is fully utilized to improve the generalization ability and recognition performance of the model, while ensuring data privacy and security.

[0063] The innovation of this invention lies in the fact that it applies the federated learning method to the SAR image target recognition task for the first time, effectively solving the problems of insufficient SAR data samples and data privacy protection. By transferring model parameters rather than raw data between participants, multiple satellite agencies can jointly train models without sharing data.

[0064] (2) Experimental results show that the multi-client collaborative SAR image target recognition method based on federated learning proposed in this invention effectively guarantees the model performance under the premise of data privacy protection, and can gradually approach the performance of centralized training through iterative optimization. The specific performance is as follows:

[0065] 1. Ensure data privacy and security: Without sharing the original data, each client only passes the model parameters, which effectively protects data privacy and security.

[0066] 2. Guarantee recognition accuracy: Through multi-client collaborative training, the model can make full use of the data of each client, while ensuring data privacy, and can gradually approach the performance of centralized training through iterative optimization. In the task of 10-client collaborative training, after 500 rounds of communication, the verification accuracy can be as high as 99.25%.

[0067] Therefore, the present invention is of great significance in solving the problems of insufficient data samples and data privacy protection in SAR image target recognition. Under the premise of data privacy protection, the model performance is effectively guaranteed. The method has broad application prospects in military, civilian and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 A flowchart of a multi-client collaborative synthetic aperture radar (SAR) image target recognition method based on federated learning provided in an embodiment of the present invention;

[0069] Figure 2Schematic diagram of vehicle optical and SAR data in the MSTAR dataset provided in an embodiment of the present invention, wherein (a) ZSU234; (b) ZIL131; (c) T72; (d) T62; (e) D7; (f) BTR70; (g) BTR60; (h) BRDM2; (i) BMP2; (j) 2S1;

[0070] Figure 3 A schematic diagram of a multi-client collaborative synthetic aperture radar (SAR) image target recognition task framework provided in an embodiment of the present invention;

[0071] Figure 4 A schematic diagram of a network structure of a client and a server provided in an embodiment of the present invention;

[0072] Figure 5 A training loss result diagram obtained by using a federated learning algorithm under collaborative tasks with different numbers of clients provided in an embodiment of the present invention;

[0073] Figure 6 This is a graph of verification accuracy results obtained by using a federated learning algorithm under collaborative tasks with different numbers of clients provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0075] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0077] Example 1

[0078] This embodiment provides a multi-client collaborative SAR image target recognition method based on federated learning, which is used to enable multiple satellite organizations to jointly train models without sharing SAR data, thereby achieving data privacy protection and improving the performance of the target recognition model, including the following steps:

[0079] S1: Collect SAR image data and perform preprocessing to generate corresponding target recognition labels to obtain a labeled data set. This embodiment also divides the labeled data set into a training set and a test set. The training set is used in the training process of the federated learning framework, and the test set is used in the testing process of the federated learning framework.

[0080] S2: Build a federated learning framework, which includes a server and multiple clients participating in collaborative training. Each client shares model parameters with the server; build a neural network model for target recognition, initialize and distribute global model parameters to each client;

[0081] S3: allocating local private data to each client according to the labeled data set, each client using the local private data to perform local model training, and uploading updated model parameters to the server;

[0082] S4: The server aggregates the model parameters uploaded by all clients, updates the global model parameters, and distributes the updated global model parameters to each client;

[0083] S5: Repeat steps S3 and S4 until the global model parameters reach a convergence state, and obtain a trained global model;

[0084] S6: Test the trained global model, obtain the verification accuracy, evaluate the model performance, and select the tested global model for SAR image target recognition.

[0085] The following combination Figure 1 The process of the multi-client collaborative synthetic aperture radar (SAR) image target recognition method based on federated learning in this embodiment is described.

[0086] Step S1, SAR image data acquisition and preprocessing: acquire SAR image data, preprocess the data, generate labels and divide the data set into a training set and a test set.

[0087] Step S1 includes the following sub-steps:

[0088] Step S101: Data collection: The satellite mechanism collects SAR image data.

[0089] Step S102: Data preprocessing: Perform preprocessing operations such as denoising, correction, and enhancement on the collected SAR image data to improve data quality.

[0090] Step S103: Generate a label. Generate a corresponding target recognition label y according to the SAR image data.

[0091] Step S104: Divide the dataset. In the example of collaborative multi-client SAR target recognition, we used the Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset. The MSTAR dataset is a widely used SAR target recognition dataset that contains SAR images of 10 types of military vehicles. Figure 2 The schematic diagram of optical and SAR images corresponding to various types of vehicles in the MSTAR dataset is shown. Table 1 shows the number of samples corresponding to the 10 types of targets in the MSTAR dataset.

[0092] Each satellite agency will process its own labeled dataset D i Divide into training set D tr and the test set D te In this example, the training set and the test set consist of target data with pitch angles of 17° and 15°, respectively.

[0093] Specifically, in the example of collaborative multi-client SAR target recognition, we assume that the samples of each category are randomly and evenly distributed among all clients, ensuring that there is no overlap or intersection between the data of each client. The example simulates independent and identically distributed (IID) and divides the dataset evenly and randomly to the clients participating in the collaborative training. To verify the effectiveness and generalization of the proposed method, the example simulates the configuration scenarios of 5 and 10 clients, and compares the distributed training results with the centralized training (all data is held by a single client).

[0094] Table 1 The number of samples corresponding to the 10 types of targets in the MSTAR dataset

[0095]

[0096] Step S2, build a federated learning framework and initialize global model parameters: establish a federated learning framework and initialize the parameters of the global model for collaborative training.

[0097] Step S2 includes the following sub-steps:

[0098] Step S201: Establish a federated learning framework and determine the participating clients and servers. Specifically, it includes: determining the client U participating in the collaborative training ll , the number of clients N and the central server S fl The example of multi-client collaborative SAR target recognition simulates configuration scenarios with 5 and 10 clients, (i.e., N = 5, 10), and compares the distributed training results with centralized training (all data is held by a single client, N = 1).

[0099] Figure 3A framework for multi-client collaborative synthetic aperture radar (SAR) image target recognition. The collaborative federated learning framework for multi-client SAR target recognition involves multiple SAR data clients that collaboratively train a global model without sharing the original data to protect data privacy. Each client trains the model locally using its own data and only shares the model parameters with the central server, which aggregates these parameters to update the global model. The whole process is as follows Figure 3 As shown in Figure 1, there are three main steps: training task initialization, local model training and updating, and global model aggregation and updating. In this framework, the blue arrows represent uploading local model parameters from the client to the server, and the orange arrows represent distributing aggregated global model parameters from the server to the client. Specifically:

[0100] 1. Training task initialization: The server initializes the global model parameters and distributes them to each client.

[0101] 2. Local model training and updating: Each client trains the model locally using its private data and updates the local model parameters.

[0102] 3. Global model aggregation and update: The server receives the local model parameters uploaded by all clients, aggregates and updates the global model parameters, and then distributes the updated global model parameters to each client.

[0103] Step S202: Initialize global model parameters, set model parameters such as initial weights and learning rates. Specifically, it includes: defining the architecture of the model; initializing global model parameters w according to the preset neural network structure; g , the global model parameters after initialization are Set the learning rate η and other related hyperparameters, such as batch size B, number of training rounds E, etc.

[0104] In order to effectively perform SAR target recognition in a federated learning environment, in this example, we designed a neural network model called ClientNet, as shown in Figure 4 As shown in Figure 2, the model mainly consists of two parts: a backbone network for feature extraction and a classifier composed of fully connected layers for classification.

[0105] Figure 4The overall architecture of ClientNet is shown. The backbone network progressively extracts high-level features from the input SAR image through a series of convolutional layers, each followed by batch normalization, ReLU activation, and pooling layers. This layered approach ensures that the model can capture the spatial and hierarchical features of the input data. Specifically, the backbone network consists of four convolutional modules, as detailed in Table 2. After feature extraction, the fully connected layer acts as a classifier. The fully connected layer maps the extracted features to the final output categories. Specifically, the fully connected layer consists of two fully connected layers. The first fully connected layer has 256 units, followed by a ReLU activation function and a Dropout layer to prevent overfitting. The second fully connected layer maps the 256 units to the number of output categories, and finally uses a Softmax activation function to convert logits into probabilities for multi-class classification.

[0106] In addition, in the example, the specific hyperparameters are set as follows: Batch Size is 100, Maximum Communication Rounds is 500, Learning Rate is 0.001, and AdamW is used as the optimizer. All experiments are conducted on a single RTX 3090 GPU. The global model parameters are initialized through the above steps to ensure the basic performance of the model, which provides a basis for subsequent local training and global aggregation.

[0107] Table 2 Configuration of convolutional modules in the ClientNet backbone network

[0108]

[0109] Step S203: The server distributes initial global model parameters. fl The global model parameters that will be initialized Distribute to each client U fl .

[0110] Step S3: Local model training and updating: Each client uses private data to perform local model training and uploads the updated model parameters to the server.

[0111] Step S3 includes the following sub-steps:

[0112] Step S301: Each client loads the global model parameters. fl Load global model parameters at current iteration t

[0113]

[0114] Step S302: Use local private data to perform model training.fl Use its private data D i Perform several rounds of iterative training with the device, calculate the loss function and perform gradient updates until the local model parameters converge. fl By minimizing the local loss function To update the local model parameters w i :

[0115]

[0116] Among them, η is the learning rate, l is the cross entropy loss function, (x i,j ,y i,j ) are local data and corresponding labels.

[0117] Step S303: Upload the updated local model parameters to the server. fl The updated local model parameters Send to server S fl .

[0118] Step S4: Global model aggregation and update: The server aggregates the model parameters uploaded by all clients and updates the global model parameters.

[0119] Step S4 includes the following sub-steps:

[0120] Step S401: Receive local model parameters uploaded by the client. fl From each client fl Receive transmitted local model parameters

[0121] Step S402: Aggregate local model parameters. fl According to the predetermined aggregation algorithm, such as the FedAvg algorithm, the model parameters of all clients are weighted averaged or other aggregation operations are performed to generate new global model parameters.

[0122]

[0123] Where N is the number of clients, |D i | is the data volume of the i-th client, is the total amount of data.

[0124] Step S403: Update global model parameters. fl The aggregated global model parameters Replace the old global model parameters:

[0125]

[0126] Step S404: prepare to transmit the global model parameters. The server prepares to transmit the new global model parameters Distribute to each client U fl To start a new round of training.

[0127] Step S5: Iterate training until the global model converges.

[0128] Step S5 includes the following sub-steps:

[0129] Step S501: Distribute the updated global model parameters to each client. fl The new global model parameters Distribute to each client U fl .

[0130] Step S502: Each client repeats steps S301 and S302 to perform local model training and updating. fl Use the updated global model parameters Perform local training and upload the updated local model parameters to the server. Specifically include: Client U fl Receive new global model parameters Then replace the local model parameters:

[0131]

[0132] Use local data D i Training is performed by minimizing the local loss function Update local model parameters w i :

[0133]

[0134] After training, the updated local model parameters Send to server S fl .

[0135] Step S503: The server repeats steps S401 to S403 to aggregate and update the global model. fl Receive local model parameters uploaded by the client And use the FedAvg algorithm for weighted averaging to generate new global model parameters

[0136]

[0137] Step S504: Iterate steps S5-1 to S5-3 until the global model parameter w gConvergence means that the model performance reaches the predetermined convergence standard or the number of training rounds reaches the preset maximum value.

[0138] Step S6: Model testing and performance evaluation: Test the trained global model on the test data set, obtain the verification accuracy, and evaluate the model performance.

[0139] Step S6 includes the following sub-steps:

[0140] Step S601: Prepare a test data set. Select and prepare an independent test data set D separated from the training data set. te to ensure the objectivity of the test results.

[0141] Step S602: Load the global model parameters. Load into the test environment.

[0142] Step S603: Perform model inference on the test data set. te Infer the global model and obtain the model's prediction results

[0143]

[0144] Among them, x test is the test data, and f(·) is the model function.

[0145] Step S604: Calculate the verification accuracy. and the true label y of the test dataset test Compare and calculate the verification accuracy Acc of the model val , to evaluate the recognition performance of the model:

[0146]

[0147] Among them, 1() is an indicator function, which takes the value of 1 when the predicted result is the same as the true label, otherwise it is 0.

[0148] Step S605: Evaluate model performance. In addition to verifying the accuracy, the model can also be comprehensively evaluated by calculating other performance indicators to determine the effectiveness and reliability of the model in practical applications. The details are as follows:

[0149] 1. Record the training loss L under different numbers of clients train , which provides the model's learning process and convergence speed under different client configurations:

[0150]

[0151] Figure 5The figure shows the changes in training loss for different numbers of clients in 500 rounds of communication. As can be seen from the figure, as the number of communication rounds increases, the training loss gradually decreases and tends to stabilize, indicating that the model gradually converges. Specifically, the training loss decreases fastest when the number of clients is 1 (centralized training). When the number of clients is 5 (federated learning), the curve change is similar to the result of centralized training. When the number of clients is 10 (federated learning), the training loss decreases slightly slower but eventually converges. This shows that although the training loss of federated learning fluctuates slightly in the early stage, as the number of communication rounds increases, the training loss gradually decreases and eventually approaches the effect of centralized training.

[0152] 2. Record the verification accuracy of the global model in each communication round Helps understand the changes and stability of model accuracy during training.

[0153] Figure 6 The figure shows the changes in verification accuracy for different numbers of clients in 500 rounds of communication. As can be seen from the figure, with the increase in the number of communication rounds, the verification accuracy of the model gradually increases and tends to stabilize, indicating that the model gradually converges. The verification accuracy increases fastest when the number of clients is 1 (centralized training), while when the number of clients is 5 and 10 (federated learning), the verification accuracy fluctuates greatly in the early stage, but eventually stabilizes and approaches the effect of centralized training.

[0154] 3. Record the global model verification accuracy after completing the maximum number of communication rounds (e.g., 500 rounds) Measure the overall effectiveness and performance of the training process for different numbers of clients (e.g., 1, 5, and 10).

[0155] Table 3 shows the verification accuracy of the model after 500 rounds of communication under different client number configurations. Experimental results show that centralized training (N=1) achieves a verification accuracy of 99.26%, while federated learning achieves 99.17% and 99.25% under 5 and 10 client configurations, respectively. These results show that the proposed scheme achieves multi-client collaborative training while maintaining high model performance.

[0156] Table 3 Verification accuracy of the model under different client number configurations (500 rounds of communication)

[0157]

[0158] Step S606: Optimize the model. According to the performance evaluation results, optimize and improve the model to improve its performance and generalization ability. The specific optimization measures that can be taken are as follows:

[0159] 1. Adjust hyperparameters:

[0160] Learning rate: Adjust the learning rate appropriately according to the convergence of the model during training. If the model converges too slowly, increase the learning rate appropriately; if the model oscillates or overfits during training, reduce the learning rate appropriately.

[0161] Batch size: Adjust the batch size based on the model's training efficiency and memory usage. A larger batch size can speed up training but may cause insufficient memory; a smaller batch size can improve the model's generalization ability but slow down training.

[0162] Communication rounds: Adjust the maximum communication rounds according to the performance of the model in different communication rounds. Experimental results show that the model gradually converges after 500 rounds of communication, and the communication rounds can be appropriately increased or decreased according to specific needs.

[0163] 2. Improve model architecture:

[0164] Increase network depth or width: Improve the model’s feature extraction capabilities by increasing the number of convolutional layers or the number of filters.

[0165] Introduce other neural network modules: Use more advanced neural network modules (such as residual networks, attention mechanisms, etc.) to improve the performance of the model.

[0166] 3. Optimize training strategies:

[0167] Use more advanced optimization algorithms: such as AdamW, RMSprop, etc. to accelerate the convergence process of the model.

[0168] Early stopping strategy: During the training process, when the verification accuracy no longer improves significantly, stop training early to prevent overfitting.

[0169] Regularization method: Introduce L2 regularization or Dropout technology to prevent the model from overfitting and improve its generalization ability.

[0170] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A multi-client collaborative SAR image target recognition method based on federated learning, characterized in that: The following steps are involved: S1: Collect SAR image data and preprocess it to generate corresponding target recognition labels to obtain labeled data sets; S2: Build a federated learning framework, which includes a server and multiple clients participating in collaborative training. Each client shares model parameters with the server; build a neural network model for target recognition, initialize and distribute global model parameters to each client; S3: allocating local private data to each client according to the labeled data set, each client using the local private data to perform local model training, and uploading updated model parameters to the server; S4: The server aggregates the model parameters uploaded by all clients, updates the global model parameters, and distributes the updated global model parameters to each client; S5: Repeat steps S3 and S4 until the global model parameters reach a convergence state, and obtain a trained global model; S6: Test the trained global model and select the global model that has passed the test for SAR image target recognition.

2. According to the method of multi-client collaborative SAR image target recognition based on federated learning in claim 1, it is characterized in that: Step S2 specifically includes the following steps: S201: Establish a semi-decentralized federated learning framework, determine the participating clients and servers, and the number of clients, each client shares the model parameters with the server, and the server aggregates the model parameters of each client to update the global model; S202: pre-build a neural network model for target recognition and initialize global model parameters; S203: The server distributes the initialized global model parameters to each client.

3. According to the method of multi-client collaborative SAR image target recognition based on federated learning in claim 1, it is characterized in that: Step S3 specifically includes the following steps: S301: Each client loads the global model parameters transmitted by the server; S302: Each client holds a local data set allocated based on the labeled data set. Each client performs local training based on the allocated global model parameters, calculates the loss function and performs gradient update. The corresponding update expression of the local model parameters is: In the formula, is the local model parameter at the t+1th iteration, is the local model parameter at the tth iteration, η is the learning rate, Based on the model parameters The loss function calculation results under , l is the cross entropy loss function, D i is the i-th local data set, (x i,j ,y i,j ) is the local data and the corresponding target identification label; S303: Each client sends the updated local model parameters to the server respectively.

4. According to the method of multi-client collaborative SAR image target recognition based on federated learning in claim 1, it is characterized in that: Step S4 specifically includes the following steps: S401: The server receives local model parameters transmitted from each client; S402: The server aggregates the local model parameters of each client according to a predetermined aggregation algorithm to generate new global model parameters; S403: The server updates the global model parameters based on the new global model parameters; S404: The server distributes the updated global model parameters to each client.

5. The method for multi-client collaborative SAR image target recognition based on federated learning according to claim 4 is characterized in that: The aggregation algorithm is the FedAvg algorithm, which is used to perform a weighted average operation on the model parameters of all clients. The calculation expression of the FedAvg algorithm is: In the formula, is the global model parameter of the t+1th iteration, N is the number of clients, |D i | is the data volume of the i-th client, is the total amount of data, is the local model parameter of the i-th client at the t-th iteration.

6. The method for multi-client collaborative SAR image target recognition based on federated learning according to claim 1, characterized in that: Step S5 specifically includes the following steps: S501: The server distributes the updated global model parameters to each client; S502: Each client executes step S3 to perform local model training; S503: The server executes step S4 to aggregate and update the global model; S504: Iteratively execute steps S501-S503 until the global model parameters converge.

7. The method for multi-client collaborative SAR image target recognition based on federated learning according to claim 1 is characterized in that: Step S1 specifically includes the following steps: S101: Collect SAR image data through satellite mechanism; S102: preprocessing the collected SAR image data, the preprocessing including denoising, correction and enhancement; S103: Generate corresponding target recognition labels according to the preprocessed SAR image data.

8. The method for multi-client collaborative SAR image target recognition based on federated learning according to claim 1 is characterized in that: The neural network model for target recognition includes a backbone network for feature extraction and a classifier composed of fully connected layers for classification; The backbone network gradually extracts high-level features from the input SAR image through a series of convolutional layers; The classifier maps the high-level features extracted by the backbone network to the final output category.

9. The method for multi-client collaborative SAR image target recognition based on federated learning according to claim 1, characterized in that: Step S6 specifically includes the following steps: S601: Select and prepare a test data set; S602: Load the trained global model parameters into the test environment; S603: Infer the global model in the test data set to obtain a prediction result, the calculation expression of which is: In the formula, To predict the results, is the global model parameter, x test is the test data, f(·) is the model function; S604: Compare the prediction result with the true label y of the test data set test For comparison, the verification accuracy Acc of the model is calculated. The calculation expression of the accuracy Acc is: Where D test is the test data set, 1() is the indicator function, which takes the value of 1 when the predicted result is the same as the true label, otherwise it takes the value of 0. is the i-th prediction result, y test，i is the i-th true label y test ; S605: Evaluate model performance according to the verification accuracy Acc of the model; S607: Optimize and improve the global model according to the model performance evaluation results.

10. The method for multi-client collaborative SAR image target recognition based on federated learning according to claim 9, characterized in that: Step S605 also includes: Record the training loss L under different number of clients train , used to evaluate the learning process of the global model and the convergence speed under different client configurations; Record the verification accuracy Acc of the global model in each communication round round , used to evaluate the change and stability of model accuracy during training; Record the global model verification accuracy Acc after completing the maximum communication round final , used to measure the overall effectiveness and performance of the training process for different numbers of clients.

Images