A Deep Learning-Based Radar Individual Open Set Recognition Method and Device
By using deep learning network models to preprocess, extract features, and classify individual radar signals, the problem of being unable to identify unknown radiation sources in existing technologies is solved, and accurate identification and classification of known and unknown signals are achieved.
Patent Information
- Application Number
- CN202411719143.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-28
- Publication Date
- 2025-10-31
- Estimated Expiration
- 2044-11-28
AI Technical Summary
Existing radar radiation source individual identification methods are mainly for closed set identification and cannot effectively identify unknown radiation sources, resulting in low reliability of identification results, especially in complex electromagnetic environments where it is difficult to distinguish between known and unknown signals.
A deep learning-based open set radar individual identification method is adopted. Convolutional neural networks, long short-term memory networks, and multilayer perceptron modules are used in combination with unsupervised clustering algorithms to preprocess, extract features, and classify radar individual signals. A loss function is designed to balance the class weights to achieve the identification of known and unknown signals.
It enables the effective identification and classification of known radar individuals in complex electromagnetic environments, and can also identify unknown types of radar individuals, improving the accuracy and reliability of identification.
Smart Images
Figure CN119716772B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of radar signal processing technology, and in particular to a radar individual open set identification method and apparatus based on deep learning. Background Technology
[0002] However, with the increasing complexity of the electromagnetic environment and the growing diversity of radar radiation sources, the task of identifying individual radiation sources to meet the needs of modern electronic warfare presents a severe challenge. In particular, electronic reconnaissance receivers often face open electromagnetic environments where the types and numbers of radiation sources are constantly changing, but the signal database has limited capacity. Therefore, completely unknown new radiation sources may appear during identification. Most existing methods are essentially designed for closed-set identification, only able to identify radar individual types already present in the database, lacking the ability to identify open-set sources. This can lead to the misidentification of new radiation sources as known ones. Especially with the widely used deep learning-based methods, if radiation source samples not included in the training phase are encountered during identification, identification errors are most likely to occur, severely impacting the reliability of the identification results.
[0003] Because different radiation sources have different total categories and varying numbers of individuals, the methods for identification and classification also differ. In reality, signals emitted by any individual inherently possess certain differences, i.e., fingerprint characteristics. Therefore, theoretically, a deep learning network model can be designed to learn the deep features of signals. By training on a large number of samples of known types, the deep fingerprint characteristics of these known signals can be learned. For signals from individuals of known types, a well-trained model can reconstruct the signal features and fully extract the characteristics of signals of known types. The fingerprint characteristics of signals emitted by the same category should be highly similar, and their distance in the feature space should be closer than that of other individual categories. Based on this, individuals of the same category can be identified and classified. For signals from individuals of unknown types, models trained using known signals cannot extract the feature fingerprints of these signals, and their performance in the feature space differs significantly from that of known types. In summary, the model can narrow the feature distance of signals of known types, but it has the opposite effect on signals of unknown categories. Summary of the Invention
[0004] The technical problem to be solved by the present invention is to provide a radar individual open set identification method and device based on deep learning, which can not only identify and classify signals of known types, but also identify unknown electromagnetic signals.
[0005] The technical solution adopted by this invention to solve its technical problem is: to provide a radar individual open set recognition method based on deep learning, comprising the following steps:
[0006] Acquire the radar individual signal to be identified;
[0007] The radar individual signal to be identified is preprocessed to obtain several segments of sequence signal;
[0008] The aforementioned signal sequences are input into a deep learning network model to obtain the category of the radar individual signal to be identified; wherein, the deep learning network model includes:
[0009] A convolutional neural network module is used to extract feature information from the several segments of sequential signals;
[0010] The Long Short-Term Memory (LSTM) network module is used to learn the characteristic relationships between the feature information of several sequence signals.
[0011] A multilayer perceptron module is used to reduce the dimensionality of the features output by the long short-term memory network module;
[0012] The classification module is used to identify individual radar signals based on the reduced-dimensional features, and to obtain the category of the individual radar signal to be identified.
[0013] The preprocessing of the radar individual signal to be identified to obtain several sequence signals specifically includes:
[0014] The radar individual signal to be identified is converted into an IQ data signal;
[0015] The IQ data signal is divided into several segments of sequential signal.
[0016] The feature extraction module includes a first convolutional layer, a first batch of normalization layers, a ReLU activation layer, a second convolutional layer, a third convolutional layer, a Dropout layer, a fourth convolutional layer, a fifth convolutional layer, and a second batch of normalization layers connected in sequence; the second, third, fourth, and fifth convolutional layers all have a kernel size of 3, and the kernel size of the first convolutional layer is larger than that of the second convolutional layer.
[0017] The classification module employs a softmax classifier during the training of the deep learning network model and an unsupervised clustering algorithm model during the use of the deep learning network model.
[0018] The deep learning network model is trained in the following way:
[0019] Collect individual radar signals of different types;
[0020] Preprocessing is performed on the individual radar signals of the different types to obtain multi-segment sequence signals for each type of radar individual;
[0021] The multiple sequence signals are divided into training dataset, test dataset, and validation dataset;
[0022] The deep learning network model is trained using a training dataset, and after training, the deep learning network model is tested using a test dataset.
[0023] If the test results meet the preset requirements, the classification module in the deep learning network model will be set as an unsupervised clustering algorithm model, and the trained parameters will be retained.
[0024] The modified deep learning network model is tested using the test dataset. If the test results meet the preset requirements, the training of the deep learning network model is completed. If the test results do not meet the preset requirements, the deep learning network model is trained again until the test results meet the preset requirements.
[0025] The loss function of the deep learning network model is: loss = lam * CNLoss(pred, label) + (1 - lam) * CNLoss(pred, label) randon In this context, `loss` represents the loss function value of the deep learning network model, `lam` represents the samples drawn from the β distribution, `CNLoss()` is the cross-entropy loss function with predefined weights, `pred` is the predicted signal category, and `label` is the true signal category. randon This refers to the category of signals after their order has been randomly shuffled.
[0026] The technical solution adopted by this invention to solve its technical problem is: to provide a radar individual open set identification device based on deep learning, comprising:
[0027] The acquisition module is used to acquire the radar individual signals to be identified;
[0028] The preprocessing module is used to preprocess the radar individual signal to be identified to obtain several segments of sequence signal;
[0029] An identification module is used to input the several sequence signals into a deep learning network model to obtain the category of the radar individual signal to be identified; wherein, the deep learning network model includes:
[0030] The feature extraction module is used to extract feature information from the plurality of sequence signals;
[0031] The Long Short-Term Memory (LSTM) network module is used to learn the characteristic relationships between the feature information of several sequence signals.
[0032] A multilayer perceptron module is used to reduce the dimensionality of the features output by the long short-term memory network module;
[0033] The classification module is used to identify individual radar signals based on the reduced-dimensional features, and to obtain the category of the individual radar signal to be identified.
[0034] The technical solution adopted by the present invention to solve its technical problem is: to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-mentioned deep learning-based radar individual open set recognition method.
[0035] The technical solution adopted by the present invention to solve its technical problem is: to provide a computer-readable storage medium on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the above-mentioned radar individual open set recognition method based on deep learning.
[0036] Beneficial effects
[0037] Due to the adoption of the above technical solution, the present invention has the following advantages and positive effects compared with the prior art: The present invention targets the signals of various radar radiation sources existing in complex electromagnetic environments, and uses a designed deep learning network model to learn and classify the signals in the environment, thereby realizing the effective identification and classification of known individual types, and can also identify unknown types of radar individuals that have never been seen before. Attached Figure Description
[0038] Figure 1 This is a flowchart of the radar individual open set recognition method based on deep learning according to the first embodiment of the present invention;
[0039] Figure 2 This is a schematic diagram of the preprocessed IQ signal in the first embodiment of the present invention;
[0040] Figure 3 This is a schematic diagram of the preprocessed sequence signal in the first embodiment of the present invention;
[0041] Figure 4 This is a schematic diagram of the structure of a deep learning network model (closed set) in the first embodiment of the present invention;
[0042] Figure 5 This is a schematic diagram of the structure of the deep learning network model (open set) in the first embodiment of the present invention;
[0043] Figure 6 This is a flowchart of the training process of the deep learning network model in the first embodiment of the present invention;
[0044] Figure 7 This is a schematic diagram of the loss function weights in the first embodiment of the present invention;
[0045] Figure 8 This is a flowchart of the accuracy calculation in the first embodiment of the present invention;
[0046] Figure 9 This is a schematic diagram illustrating the changes in training set loss in the first embodiment of the present invention;
[0047] Figure 10 This is a schematic diagram showing the change in the accuracy of closed set identification verification set in the first embodiment of the present invention;
[0048] Figure 11 This is a partial open set recognition result diagram in the first embodiment of the present invention. Detailed Implementation
[0049] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.
[0050] The first embodiment of the present invention relates to a radar individual open set identification method based on deep learning, such as... Figure 1 As shown, it includes the following steps:
[0051] Step 1: Acquire the radar individual signal to be identified.
[0052] Step 2 involves preprocessing the radar individual signal to be identified to obtain several sequence signals. Specifically, the preprocessing in this step involves using a down-conversion DDC to convert the radar individual signal to be identified into an IQ data signal (see...). Figure 2 The IQ data signal is divided into several segments of sequential signal (see...). Figure 3 During segmentation, each segment of the sequence signal has an equal length. The preprocessed IQ data signal contains more signal information, including amplitude and phase.
[0053] Step 3: Input the several segments of the sequence signal into the deep learning network model to obtain the category of the radar individual signal to be identified.
[0054] like Figure 4 and Figure 5As shown, the deep learning network model in this step includes: a convolutional neural network module for extracting feature information of the several segments of sequence signals; a long short-term memory network (LSTM) module for learning the feature relationships between the feature information of the several segments of sequence signals; a multilayer perceptron (MLP) module for dimensionality reduction of the features output by the long short-term memory network module; and a classification module for identifying individual radar signals based on the dimensionality-reduced features, thereby obtaining the category of the individual radar signal to be identified.
[0055] In this embodiment, the feature extraction module employs a convolutional neural network (CNN) for feature extraction. It includes a first convolutional layer, a first batch normalization layer, a ReLU activation layer, a second convolutional layer, a third convolutional layer, a Dropout layer, a fourth convolutional layer, a fifth convolutional layer, and a second batch normalization layer, connected sequentially. To extract information from the entire signal sequence, obtain greater contextual information, and have a larger receptive field, the first convolutional layer uses a convolutional network with relatively large kernels, allowing for the processing of feature information from a longer sequence at once. However, using large kernels can easily lead to the loss of detailed features, as convolutional networks with large kernels focus more on global information. Therefore, to allow the model to learn more refined local information, in this embodiment, the second, third, fourth, and fifth convolutional layers use a convolutional network with small kernels of 3 and a stride of 2. Simultaneously, this embodiment uses a ReLU activation layer after the large kernels to increase the non-linear relationship between feature dimensions through the ReLU activation function. This embodiment also uses a batch normalization layer to increase the model's convergence speed, resulting in higher model fitting efficiency. However, this also increases the risk of model overfitting. Therefore, a Dropout layer is used after the third convolutional layer to increase the model's generalization ability. By randomly dropping data, the model's fault tolerance is increased, and the overfitting phenomenon is mitigated.
[0056] For closed-set models (see...) Figure 4 For example, the classification module uses a softmax classifier to classify the output of the MLP module. Finally, the error between the classification result and the actual true value is calculated, and the error value is used to optimize the model.
[0057] Open set model (see) Figure 5The result is obtained by further optimizing the closed-set model. Therefore, after training the closed-set model, the classification module needs to be modified, changing the softmax classifier to an unsupervised clustering algorithm to achieve the result of the open-set model. To obtain the result of the unknown type, the network of the trained closed-set model, including its parameters, is used to perform unsupervised clustering on the output of the MLP module. The softmax classifier is no longer used. The unsupervised clustering has one more category than the closed-set model; this extra category is the unknown category that the open-set model needs to obtain.
[0058] The deep learning network model in this embodiment is trained in the following way:
[0059] To achieve open set identification and classification of radar individual signals, this invention first collects different types of radar individual signals and preprocesses them by converting AD signals into IQ data signals via down-conversion DDC, and then slicing the converted IQ data signals to divide long sequence signals into multiple small sequence signals.
[0060] After completing the above preprocessing of the data, the dataset is divided into three parts, serving as the training dataset, test dataset, and validation dataset. Before training the deep learning network model using these datasets, it is necessary to design the loss function of the deep learning network model to determine the error in object recognition and classification. This implementation uses the cross-entropy loss function. It should be noted that the classes in the dataset used in this implementation are extremely imbalanced, which is a problem that needs to be addressed in practice. To address this issue, this implementation sets corresponding weights for each class to balance the model's learning ability for each class. The weight settings are as follows: Figure 7 As shown. In calculating the loss function, this implementation extracts samples from the β distribution as additional calculation weights:
[0061] lam=β(15,15)
[0062] Where lam represents a sample drawn from the β distribution. Since the number of positive and negative iterations is the same in this implementation, the probability is approximately 0.5. The loss function in this implementation is expressed as:
[0063] loss=lam*CNLoss(pred,label)+(1-lam)*CNLoss(pred,label) randon )
[0064] CNLoss() adds Figure 7The cross-entropy loss function with medium weights, where pred is the predicted signal class and label is the true signal class. randon This refers to the category of signals after their order has been randomly shuffled.
[0065] After setting the loss function, the deep learning network model is trained on the data. The Adam optimizer is used, the learning rate is set to 0.0001, and the batch size is 8. The model training process in this implementation is as follows: Figure 6 As shown, specifically: the deep learning network model is trained using a training dataset; after training, the deep learning network model is tested using a test dataset; if the test results meet the preset requirements, the classification module in the deep learning network model is set to an unsupervised clustering algorithm model, and the trained parameters are retained; the modified deep learning network model is tested using the test dataset; if the test results meet the preset requirements, the training of the deep learning network model is completed; if the test results do not meet the preset requirements, the deep learning network model is trained again until the test results meet the preset requirements.
[0066] After labeling and classifying the samples, they are fed into a deep learning network model for training and recognition, which yields results such as... Figure 9 and Figure 10 The results were obtained. After training the network, online identification and classification of new samples (including samples of individuals with unknown categories) were performed to verify the model's performance. New signal samples were fed into the deep learning model for detection, yielding results such as... Figure 11 The results are shown.
[0067] Therefore, by learning from known samples in the signal sequence library, the deep learning network model of this embodiment will automatically find the fingerprint features of signal data of known categories, narrowing the feature distance between fingerprint features of the same category. For fingerprint features of unknown categories, the model cannot learn effectively, and therefore the feature distance in the feature space is greater. Through identification and classification using these features, a model network capable of identifying and classifying individual radar signals is formed. This network can then be used to identify and classify individual radar signals.
[0068] The performance parameters of the deep learning network model in this embodiment are evaluated below. The accuracy of the model is calculated as follows:
[0069] TP+TN
[0070] accuracy = TP + TN + FP + FN
[0071] Where TP represents the retrieved positive examples, TN represents the unretrieved negative examples, FN represents the unretrieved positive examples, and FP represents the retrieved negative examples.
[0072] Because this implementation method divides a signal into multiple small samples during data preprocessing, but the accuracy calculation is based on the entire large signal segment, a voting method is used. The category to which the long signal sequence belongs is determined by the number of small samples of that category. This process is as follows: Figure 8 As shown. Then, the accuracy of the validation set and test set is calculated according to the above formula to test the recognition performance of the model.
[0073] It is easy to see that this invention targets the signals of various radar radiation sources in complex electromagnetic environments. It uses a designed deep learning network model to learn and classify the signals in the environment, thereby achieving effective identification and classification of known individual types, and can also identify unknown types of radar individuals that have never been seen before.
[0074] A second embodiment of the present invention relates to a radar individual open set identification device based on deep learning, comprising:
[0075] The acquisition module is used to acquire the radar individual signals to be identified;
[0076] The preprocessing module is used to preprocess the radar individual signal to be identified to obtain several segments of sequence signal;
[0077] An identification module is used to input the several sequence signals into a deep learning network model to obtain the category of the radar individual signal to be identified; wherein, the deep learning network model includes:
[0078] The feature extraction module is used to extract feature information from the plurality of sequence signals;
[0079] The Long Short-Term Memory (LSTM) network module is used to learn the characteristic relationships between the feature information of several sequence signals.
[0080] A multilayer perceptron module is used to reduce the dimensionality of the features output by the long short-term memory network module;
[0081] The classification module is used to identify individual radar signals based on the reduced-dimensional features, and to obtain the category of the individual radar signal to be identified.
[0082] The preprocessing module includes:
[0083] A conversion unit is used to convert the radar individual signal to be identified into an IQ data signal;
[0084] The slicing unit is used to divide the IQ data signal into several segments of sequential signal.
[0085] The feature extraction module includes a first convolutional layer, a first normalization layer, a ReLU activation layer, a second convolutional layer, a third convolutional layer, a Dropout layer, a fourth convolutional layer, a fifth convolutional layer, and a second normalization layer connected in sequence; the second, third, fourth, and fifth convolutional layers all have a kernel size of 3, and the kernel size of the first convolutional layer is larger than that of the second convolutional layer.
[0086] The classification module employs a softmax classifier during the training of the deep learning network model and an unsupervised clustering algorithm model during the use of the deep learning network model.
[0087] The deep learning network model is trained in the following way:
[0088] Collect individual radar signals of different types;
[0089] Preprocessing is performed on the individual radar signals of the different types to obtain multi-segment sequence signals for each type of radar individual;
[0090] The multiple sequence signals are divided into training dataset, test dataset, and validation dataset;
[0091] The deep learning network model is trained using a training dataset, and after training, the deep learning network model is tested using a test dataset.
[0092] If the test results meet the preset requirements, the classification module in the deep learning network model will be set as an unsupervised clustering algorithm model, and the trained parameters will be retained.
[0093] The modified deep learning network model is tested using the test dataset. If the test results meet the preset requirements, the training of the deep learning network model is completed. If the test results do not meet the preset requirements, the deep learning network model is trained again until the test results meet the preset requirements.
[0094] The loss function of the deep learning network model is: loss = lam * CNLoss(pred, label) + (1 - lam) * CNLoss(pred, label) randon In this context, `loss` represents the loss function value of the deep learning network model, `lam` represents the samples drawn from the β distribution, `CNLoss()` is the cross-entropy loss function with predefined weights, `pred` is the predicted signal category, and `label` is the true signal category. randon This refers to the category of signals after their order has been randomly shuffled.
[0095] The third embodiment of the present invention relates to an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the deep learning-based radar individual open set identification method of the first embodiment.
[0096] The fourth embodiment of the present invention relates to a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the deep learning-based radar individual open set identification method of the first embodiment.
[0097] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.
[0098] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0099] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction methods implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0100] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0101] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A radar individual open set recognition method based on deep learning, characterized in that, Includes the following steps: Acquire the radar individual signal to be identified; The radar individual signal to be identified is preprocessed to obtain several segments of sequence signal; The aforementioned signal sequences are input into a deep learning network model to obtain the category of the radar individual signal to be identified; wherein, the deep learning network model includes: The feature extraction module is used to extract feature information from the plurality of sequence signals; The Long Short-Term Memory (LSTM) network module is used to learn the characteristic relationships between the feature information of several sequence signals. A multilayer perceptron module is used to reduce the dimensionality of the features output by the long short-term memory network module; The classification module is used to identify individual radar signals based on the reduced-dimensional features, and to obtain the category of the individual radar signal to be identified. The loss function of the deep learning network model is: loss = lam * CNLoss(pred, label) + (1 - lam) * CNLoss(pred, label) randon In this context, `loss` represents the loss function value of the deep learning network model, `lam` represents the samples drawn from the β distribution, `CNLoss()` is the cross-entropy loss function with predefined weights, `pred` is the predicted signal category, and `label` is the true signal category. randon This refers to the category of signals after their order has been randomly shuffled.
2. The radar individual open set recognition method based on deep learning according to claim 1, characterized in that, The preprocessing of the radar individual signal to be identified to obtain several sequence signals specifically includes: The radar individual signal to be identified is converted into an IQ data signal; The IQ data signal is divided into several segments of sequential signal.
3. The radar individual open set recognition method based on deep learning according to claim 1, characterized in that, The feature extraction module includes a first convolutional layer, a first batch of normalization layers, a ReLU activation layer, a second convolutional layer, a third convolutional layer, a Dropout layer, a fourth convolutional layer, a fifth convolutional layer, and a second batch of normalization layers connected in sequence; the second, third, fourth, and fifth convolutional layers all have a kernel size of 3, and the kernel size of the first convolutional layer is larger than that of the second convolutional layer.
4. The radar individual open set recognition method based on deep learning according to claim 1, characterized in that, The classification module employs a softmax classifier during the training of the deep learning network model and an unsupervised clustering algorithm model during the use of the deep learning network model.
5. The radar individual open set recognition method based on deep learning according to claim 4, characterized in that, The deep learning network model is trained in the following way: Collect individual radar signals of different types; Preprocessing is performed on the individual radar signals of the different types to obtain multi-segment sequence signals for each type of radar individual; The multiple sequence signals are divided into training dataset, test dataset, and validation dataset; The deep learning network model is trained using a training dataset, and after training, the deep learning network model is tested using a test dataset. If the test results meet the preset requirements, the classification module in the deep learning network model will be set as an unsupervised clustering algorithm model, and the trained parameters will be retained. The modified deep learning network model is tested using the test dataset. If the test results meet the preset requirements, the training of the deep learning network model is completed. If the test results do not meet the preset requirements, the deep learning network model is trained again until the test results meet the preset requirements.
6. A radar individual open set identification device based on deep learning, characterized in that, include: The acquisition module is used to acquire the radar individual signals to be identified; The preprocessing module is used to preprocess the radar individual signal to be identified to obtain several segments of sequence signal; An identification module is used to input the several sequence signals into a deep learning network model to obtain the category of the radar individual signal to be identified; wherein, the deep learning network model includes: The feature extraction module is used to extract feature information from the plurality of sequence signals; The Long Short-Term Memory (LSTM) network module is used to learn the characteristic relationships between the feature information of several sequence signals. A multilayer perceptron module is used to reduce the dimensionality of the features output by the long short-term memory network module; The classification module is used to identify individual radar signals based on the reduced-dimensional features, and to obtain the category of the individual radar signal to be identified. The loss function of the deep learning network model is: loss = lam * CNLoss(pred, label) + (1 - lam) * CNLoss(pred, label) randon In this context, `loss` represents the loss function value of the deep learning network model, `lam` represents the samples drawn from the β distribution, `CNLoss()` is the cross-entropy loss function with predefined weights, `pred` is the predicted signal category, and `label` is the true signal category. randon This refers to the category of signals after their order has been randomly shuffled.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the deep learning-based radar individual open set identification method as described in any one of claims 1-5.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the deep learning-based radar individual open set identification method as described in any one of claims 1-5.
Citation Information
Patent Citations
Robust communication radiation source intelligent identification method based on deep learning
CN112613443A