A method and apparatus for radar target recognition under main lobe interference based on prior information assistance

Abstract

This application relates to a radar target recognition method and apparatus under main lobe interference based on prior information assistance. The method includes: constructing a point scatter function under interference capable of characterizing interference features and HRRP distortion patterns to effectively extract prior interference information; and combining a multi-scale feature extraction network layer with a cross-attention fusion mechanism to achieve deep fusion of HRRP features and prior interference information, thereby significantly improving the target recognition accuracy and robustness of the model under typical interference conditions such as intermittent sampling and forwarding. This solves the problems of poor robustness and decreased accuracy in high-resolution range image recognition of radar targets under typical interference conditions in existing technologies.

Description

Technical Field

[0001] This application relates to the field of automatic radar target recognition technology, and in particular to a radar target recognition method and device under main lobe interference based on prior information assistance. Background Technology

[0002] Radar target recognition is one of the core functions of radar systems and is widely used in both military and civilian fields. High-Resolution Range Profile (HRRP) radar, with its advantages of fast imaging speed, low processing complexity, and rich structural information, is an important approach to radar target recognition. Radar HRRP target recognition methods can be mainly divided into two categories: traditional recognition methods and deep learning-based recognition methods.

[0003] Traditional research on HRRP target recognition mainly focuses on designing suitable feature extraction methods. While these methods extract highly interpretable features, they are highly dependent on the researcher's design experience, and the computational complexity of feature extraction limits their application in real-world scenarios. Deep learning-based HRRP recognition primarily employs deep neural networks for feature extraction and target classification. Depending on the network model used, this method can be categorized into three main types: models based on Convolutional Neural Networks (CNNs), models based on Recurrent Neural Networks (RNNs), and models based on Transformers. With the rapid development of artificial intelligence technology, deep learning-based HRRP target recognition methods have achieved excellent recognition results and gained widespread application.

[0004] In recent years, active electronic interference has posed a severe challenge to radar. Insufficient radar anti-jamming capability, especially anti-main lobe interference capability, has increasingly become a core bottleneck affecting radar detection capability. Theoretical research and practical applications show that typical radar interference, such as intermittent sampling and relay, will inevitably lead to distortion of target recognition features, resulting in a significant decrease in actual target recognition accuracy or even complete failure. Under interference conditions, the robustness of radar target recognition is seriously insufficient.

[0005] In summary, existing radar target identification methods exhibit poor robustness and reduced accuracy in identifying high-resolution range images of radar targets under typical interference conditions such as intermittent sampling and forwarding, which urgently needs to be addressed. Summary of the Invention

[0006] This application provides a radar target recognition method and apparatus based on prior information-assisted main lobe interference to solve the problems of poor robustness and decreased recognition accuracy of existing radar target recognition methods under typical interference conditions such as intermittent sampling and forwarding.

[0007] The first aspect of this application provides a radar target recognition method under main lobe interference based on prior information, applied in the offline training stage, comprising the following steps: acquiring the echo signal and baseband waveform of the radar target under interference conditions, extracting interference features using the echo signal, generating an interference suppression operator, and performing echo interference suppression and target detection operations based on the echo signal, the baseband waveform, and the interference suppression operator to obtain interference-suppressed echo and target information, and generating a high-resolution range image of the target based on the interference-suppressed echo and the target information; performing baseband waveform equivalent processing based on the baseband waveform, the target information, and the interference suppression operator to generate an interference-based point scatter function, and constructing interference prior information corresponding to the high-resolution range image of the target based on the interference-based point scatter function; and performing the target detection based on a pre-built radar target recognition model. Multi-scale feature extraction is performed on the high-resolution range image of the target and the prior information of the interference to obtain a first multi-scale feature corresponding to the high-resolution range image of the target and a second multi-scale feature corresponding to the prior information of the interference. Preset channel attention and spatial attention mechanisms are used to enhance the first and second multi-scale features to generate corresponding first and second enhanced features. Based on a preset cross-attention mechanism, the first and second enhanced features are fused to generate corresponding fused features. A corresponding hybrid loss function is constructed based on a preset cross-entropy loss function and a supervised contrast loss function. The hybrid loss function is used to constrain the fused features to train the radar target recognition model, so that the trained radar target recognition model can be used to perform corresponding radar target recognition operations during the online recognition phase.

[0008] Optionally, in one embodiment of this application, the step of extracting multi-scale features from the high-resolution range image of the target and the prior information of interference based on a pre-built radar target recognition model to obtain a first multi-scale feature corresponding to the high-resolution range image of the target and a second multi-scale feature corresponding to the prior information of interference, and using a preset channel attention mechanism and spatial attention mechanism to enhance the first multi-scale feature and the second multi-scale feature to generate corresponding first enhanced features and second enhanced features, includes: inputting the high-resolution range image of the target and the prior information of interference into a pre-built multi-scale feature network layer, and using convolution kernels of different scales in parallel on the high-resolution range image of the target and the prior information of interference based on the multi-scale feature network layer to obtain the first multi-scale feature and the second multi-scale feature; and enhancing the first multi-scale feature and the second multi-scale feature in the spatial dimension respectively. A global average pooling operation is performed on the first multi-scale feature and the second multi-scale feature to obtain the corresponding channel weight features. These channel weight features are then input into a preset multilayer perceptron to model the nonlinear relationship between channels. Based on this nonlinear relationship and a preset activation function, corresponding channel attention weights are generated. The channel attention weights are then multiplied element-wise with both the first and second multi-scale features to obtain the corresponding channel attention output features. The channel attention output features are then average pooled along the channel dimension to obtain the corresponding spatial weight features. Based on these spatial weight features, the multilayer perceptron, and the preset activation function, corresponding spatial attention weights are generated. Finally, the spatial attention weights are multiplied element-wise with the channel attention output features to generate the first enhancement feature and the second enhancement feature.

[0009] Optionally, in one embodiment of this application, the step of fusing the first enhanced feature and the second enhanced feature based on a preset cross-attention mechanism to generate a corresponding fused feature includes: performing a linear mapping on the first enhanced feature and the second enhanced feature to convert the first enhanced feature and the second enhanced feature into corresponding query vectors, key vectors, and value vectors, respectively; adding preset sine and cosine position codes to the query vectors, the key vectors, and the value vectors, respectively, to obtain corresponding position-coded query vectors, position-coded key vectors, and position-coded value vectors; calculating an attention matrix based on the position-coded query vectors and the position-coded key vectors; multiplying the attention matrix by the position-coded value vectors to obtain fused attention-aware features; performing residual connections and layer normalization operations on the fused attention-aware features and the first enhanced feature to obtain corresponding normalization results; and sending the normalization results to a preset feedforward network to output the fused feature.

[0010] Optionally, in one embodiment of this application, the mathematical expression of the hybrid loss function is:

[0011] L total = (1-λ)·L CEL +λ·L SCL

[0012] Among them, L total L represents the hybrid loss function; CEL L represents the cross-entropy loss function; SCL Let represent the supervised contrastive loss function; 1-λ represent the hyperparameters corresponding to the cross-entropy loss function; λ represent the hyperparameters corresponding to the supervised contrastive loss function.

[0013] A second aspect of this application provides a radar target recognition method under main lobe interference based on prior information, applied in the online recognition stage, comprising the following steps: acquiring the target high-resolution range image and radar echo signal of the target radar; extracting key interference features corresponding to the radar echo signal to construct corresponding target interference prior information based on the key interference features; inputting the target high-resolution range image and target interference prior information of the target radar into a pre-trained radar target recognition model to output the corresponding radar target recognition result.

[0014] A third aspect of this application provides a radar target recognition device under main lobe interference based on prior information, applied in the offline training phase, comprising: an interference suppression module, used to acquire the echo signal and baseband waveform of the radar target under interference conditions, extract interference features using the echo signal, generate an interference suppression operator, and perform echo interference suppression and target detection operations based on the echo signal, the baseband waveform, and the interference suppression operator to obtain interference-suppressed echo and target information, and generate a high-resolution range image of the target based on the interference-suppressed echo and the target information; a baseband waveform equivalent processing module, used to perform baseband waveform equivalent processing based on the baseband waveform, the target information, and the interference suppression operator to generate an interference-based point scatter function, and construct interference prior information corresponding to the high-resolution range image of the target based on the interference-based point scatter function; and a feature extraction module, used to extract prior information based on a pre-constructed radar target recognition feature. The radar target recognition model is configured to perform multi-scale feature extraction on the high-resolution range image of the target and the prior information of the interference to obtain a first multi-scale feature corresponding to the high-resolution range image of the target and a second multi-scale feature corresponding to the prior information of the interference. A preset channel attention mechanism and a spatial attention mechanism are used to enhance the first and second multi-scale features to generate corresponding first and second enhanced features. A feature fusion module is used to fuse the first and second enhanced features based on a preset cross-attention mechanism to generate corresponding fused features. A training module is used to construct a corresponding hybrid loss function based on a preset cross-entropy loss function and a supervised contrast loss function, and to use the hybrid loss function to constrain the fused features to train the radar target recognition model. The trained radar target recognition model is then used to perform corresponding radar target recognition operations during the online recognition phase.

[0015] Optionally, in one embodiment of this application, the feature extraction module includes: a convolution unit, used to input the target high-resolution range image and the interference prior information into a pre-constructed multi-scale feature network layer, so as to perform convolution operations on the target high-resolution range image and the interference prior information in parallel using convolution kernels of different scales based on the multi-scale feature network layer, so as to obtain the first multi-scale feature and the second multi-scale feature; a first pooling unit, used to perform global average pooling on the first multi-scale feature and the second multi-scale feature in the spatial dimension respectively, so as to obtain the corresponding channel weight features, and input the channel weight features into a preset multilayer perceptron, so as to model the nonlinear relationship between channels, and based on the channel... The system employs a nonlinear relationship and a preset activation function to generate corresponding channel attention weights. A first multiplication unit performs element-wise multiplication of the channel attention weights with the first multi-scale feature and the second multi-scale feature to obtain corresponding channel attention output features. A second pooling unit performs average pooling on the channel attention output features along the channel dimension to obtain corresponding spatial weight features, and generates corresponding spatial attention weights based on the spatial weight features, the multilayer perceptron, and the preset activation function. A second multiplication unit performs element-wise multiplication of the spatial attention weights with the channel attention output features to generate the first enhanced feature and the second enhanced feature.

[0016] Optionally, in one embodiment of this application, the feature fusion module includes: a linear mapping unit, configured to perform linear mapping on the first enhanced feature and the second enhanced feature to convert the first enhanced feature and the second enhanced feature into corresponding query vectors, key vectors and value vectors, respectively; a calculation unit, configured to add preset sine and cosine position codes to the query vector, the key vector and the value vector, respectively, to obtain corresponding position-coded query vectors, position-coded key vectors and position-coded value vectors, and calculate an attention matrix based on the position-coded query vector and the position-coded key vector; a normalization unit, configured to multiply the attention matrix with the position-coded value vector to obtain fused attention-aware features, and perform residual connection and layer normalization operations on the fused attention-aware features and the first enhanced feature to obtain corresponding normalization results; and an output unit, configured to send the normalization results to a preset feedforward network to output the fused features.

[0017] Optionally, in one embodiment of this application, the mathematical expression of the hybrid loss function is:

[0018] L total = (1-λ)·L CEL +λ·L SCL

[0019] Among them, L total L represents the hybrid loss function; CEL L represents the cross-entropy loss function; SCL Let represent the supervised contrastive loss function; 1-λ represent the hyperparameters corresponding to the cross-entropy loss function; λ represent the hyperparameters corresponding to the supervised contrastive loss function.

[0020] The fourth aspect of this application provides a radar target recognition device under main lobe interference based on prior information, applied in the online recognition stage, comprising: an acquisition module for acquiring the target high-resolution range image and radar echo signal of the target radar corresponding to the target radar; a construction module for extracting key interference features corresponding to the radar echo signal to construct corresponding target interference prior information based on the key interference features; and a recognition module for inputting the target high-resolution range image and target interference prior information of the target radar into a pre-trained radar target recognition model to output the corresponding radar target recognition result.

[0021] A fifth aspect of this application provides 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 program to implement the radar target recognition method under main lobe interference based on prior information assistance as described in the above embodiments.

[0022] A sixth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the radar target identification method under main lobe interference based on prior information assistance as described above.

[0023] A seventh aspect of this application provides a computer program product, including a computer program that is executed to implement the above-described radar target recognition method under main lobe interference based on prior information assistance.

[0024] Therefore, the embodiments of this application have the following beneficial effects:

[0025] The embodiments of this application can acquire the echo signal and baseband waveform of a radar target under interference conditions, extract interference features using the echo signal, generate an interference suppression operator, and perform echo interference suppression and target detection operations based on the echo signal, baseband waveform, and interference suppression operator to obtain interference-suppressed echo and target information. A high-resolution range profile of the target is then generated based on the interference-suppressed echo and target information. Based on the baseband waveform, target information, and interference suppression operator, equivalent processing of the baseband waveform is performed to generate an interference sub-point scatter function, and interference prior information corresponding to the high-resolution range profile of the target is constructed based on the interference sub-point scatter function. Finally, based on a pre-built radar target recognition model, multi-scale feature extraction is performed on the high-resolution range profile of the target and the interference prior information. This paper describes a radar target recognition model. The model obtains first multi-scale features corresponding to the high-resolution range image of the target and second multi-scale features corresponding to the prior information of interference. It then uses preset channel attention and spatial attention mechanisms to enhance these features, generating corresponding first and second enhanced features. Based on a preset cross-attention mechanism, the first and second enhanced features are fused to generate corresponding fused features. A corresponding hybrid loss function is constructed based on preset cross-entropy loss and supervised contrast loss functions, and the hybrid loss function is used to constrain the fused features. This is used to train the radar target recognition model, which is then used to perform corresponding radar target recognition operations during the online recognition phase. This application constructs a point spread function that can characterize interference characteristics and HRRP distortion patterns, effectively extracting prior information of interference. By combining a multi-scale feature extraction module with a cross-attention fusion mechanism, it achieves deep fusion of HRRP features and prior information of interference, thereby significantly improving the target recognition accuracy and robustness of the model under typical interference conditions such as intermittent sampling and forwarding. This solves the problems of poor robustness and decreased accuracy in high-resolution range image recognition of radar targets under typical interference conditions such as intermittent sampling and forwarding in existing radar target recognition methods.

[0026] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0027] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0028] Figure 1 This is a flowchart of a radar target recognition method based on prior information-assisted main lobe interference, provided according to an embodiment of this application, for use in the offline training phase;

[0029] Figure 2This is a schematic diagram of a radar echo signal processing and target recognition process according to an embodiment of this application;

[0030] Figure 3 This is a schematic diagram of HRRP and interference prior information provided according to an embodiment of this application;

[0031] Figure 4 This is a schematic diagram of the logical architecture of a radar target recognition method under main lobe interference based on prior information assistance, according to an embodiment of this application.

[0032] Figure 5 This is a schematic diagram illustrating the effect of supervised comparison loss according to an embodiment of this application;

[0033] Figure 6 This is a schematic diagram illustrating the identification of a confusion matrix using a test set of different algorithms according to embodiments of this application.

[0034] Figure 7 This is a schematic diagram of the generalization confusion matrix of different algorithms provided according to embodiments of this application;

[0035] Figure 8 This is a flowchart of a radar target recognition method based on prior information-assisted main lobe interference, provided according to an embodiment of this application, for use in the online recognition stage;

[0036] Figure 9 This is an example diagram of a radar target recognition device based on prior information-assisted main lobe interference applied in the offline training phase according to an embodiment of this application;

[0037] Figure 10 This is an example diagram of a radar target identification device based on prior information-assisted main lobe interference applied in the online identification stage according to an embodiment of this application;

[0038] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0039] Among them, 10-a radar target recognition device based on prior information-assisted main lobe interference applied in the offline training stage; 20-a radar target recognition device based on prior information-assisted main lobe interference applied in the online recognition stage; 101-interference suppression module; 102-baseband waveform equivalent processing module; 103-feature extraction module; 104-feature fusion module; 105-training module; 201-acquisition module; 202-construction module; 203-recognition module; 1101-memory; 1102-processor; 1103-communication interface. Detailed Implementation

[0040] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0041] The following describes a radar target recognition method and apparatus based on prior information-assisted main lobe interference according to embodiments of this application, with reference to the accompanying drawings. Addressing the problems mentioned in the background art, this application provides a radar target recognition method based on prior information-assisted main lobe interference. In this method, the echo signal and baseband waveform of the radar target under interference conditions are acquired, and interference features are extracted using the echo signal. An interference suppression operator is generated, and based on the echo signal, baseband waveform, and interference suppression operator, echo interference suppression and target detection operations are performed to obtain interference-suppressed echo and target information. A high-resolution range profile of the target is generated based on the interference-suppressed echo and target information. Based on the baseband waveform, target information, and interference suppression operator, equivalent processing of the baseband waveform is performed to generate a point scatter function under interference. Prior information on interference corresponding to the high-resolution range profile of the target is constructed based on the interference point scatter function. Based on a pre-built radar target recognition model, the target is identified... This application employs multi-scale feature extraction from high-resolution range images and prior interference information to obtain first multi-scale features corresponding to the target's high-resolution range image and second multi-scale features corresponding to the prior interference information. Pre-defined channel attention and spatial attention mechanisms are used to enhance these first and second multi-scale features, generating corresponding enhanced features. Based on a pre-defined cross-attention mechanism, the first and second enhanced features are fused to generate corresponding fused features. A hybrid loss function is constructed based on pre-defined cross-entropy loss and supervised contrast loss functions, and this hybrid loss function is used to constrain the fused features to train a radar target recognition model. This trained model is then used to perform corresponding radar target recognition operations during the online recognition phase. This application effectively extracts prior interference information by constructing a point spread function that characterizes interference characteristics and HRRP distortion patterns. By combining a multi-scale feature extraction module with a cross-attention fusion mechanism, it achieves deep fusion of HRRP features and prior interference information, significantly improving the model's target recognition accuracy and robustness under typical interference conditions such as intermittent sampling and forwarding. This solves the problems of poor robustness and decreased accuracy in high-resolution range image recognition of radar targets under typical interference conditions such as intermittent sampling and forwarding in existing radar target recognition methods.

[0042] Specifically, Figure 1 This is a flowchart illustrating a radar target recognition method based on prior information-assisted main lobe interference, provided as an embodiment of this application, for use in the offline training phase.

[0043] like Figure 1 As shown, the radar target recognition method under main lobe interference based on prior information includes the following steps:

[0044] In step S101, the echo signal and baseband waveform of the radar target under interference conditions are acquired, and the interference features are extracted using the echo signal. An interference suppression operator is generated, and based on the echo signal, baseband waveform and interference suppression operator, echo interference suppression and target detection operations are performed to obtain the interference-suppressed echo and target information. A high-resolution range image of the target is generated based on the interference-suppressed echo and target information.

[0045] In step S102, based on the baseband waveform, target information, and interference suppression operator, equivalent processing of the baseband waveform is performed to generate the interference sub-point scatter function, and interference prior information corresponding to the target high-resolution range image is constructed according to the interference sub-point scatter function.

[0046] The embodiments of this application can first obtain the target high-resolution range image corresponding to the preset radar target under interference conditions; in addition, since the HRRP feature is easily affected by interference distortion under intermittent sampling and forwarding interference, the embodiments of this application also need to introduce interference prior information to assist radar target identification.

[0047] Those skilled in the art should understand that, in intermittent sampling and relay jamming scenarios, the echo signal received by the radar includes the target scattered echo and the corresponding jamming signal.

[0048] like Figure 2 As shown, to construct prior information about interference, this embodiment first analyzes the radar echo signal and extracts its key interference features, including the interference sampling period and the forwarding period, to construct an interference suppression operator. Secondly, based on the extracted key interference features, this embodiment establishes an ideal point target frequency domain response model under corresponding interference, thereby simulating and generating a point spread function (PSF) to describe the range-direction response distortion characteristics of the ideal point target under the current interference conditions.

[0049] In the embodiments of this application, the point scatter function generation process is as follows:

[0050] 1. Obtain the baseband waveform of the radar transmitted signal:

[0051]

[0052] in, B is a rectangular pulse with an amplitude of 1 and a pulse width of τ; B is the bandwidth of the signal.

[0053] Performing a Fourier transform on the baseband waveform yields:

[0054]

[0055] 2. Virtual interference suppression of baseband waveforms using interference suppression operators. In practical applications, interference suppression operators may have different forms. This application provides an example of an interference suppression operator under typical intermittent sampling and forwarding interference. Let the frequency domain form of the interference suppression operator be J(f). According to the typical intermittent sampling and forwarding interference mechanism, J(f) is a periodic rectangular window function with a finite duration. Let its period be f1, and the duration of the rectangular window within a single period be τ1. Then, the spectrum of the baseband waveform affected by the interference is:

[0056]

[0057] The pulse compression result is:

[0058]

[0059] 3. To Performing the inverse Fourier transform yields the point scatter function. The inverse Fourier transform result of Y(f) is Indicates amplitude:

[0060]

[0061] in

[0062] It is understandable that the point scatter function The response distortion pattern is the same as the distortion pattern of the real target's HRRP under interference environment, and can accurately reflect the impact of interference on HRRP. Therefore, this application embodiment uses it as interference prior information. The interference-free HRRP and the HRRP under intermittent sampling and forwarding interference and their corresponding interference prior information are as follows: Figure 3 As shown.

[0063] Therefore, the embodiments of this application address the distortion of the high-resolution range image of a target caused by intermittent sampling and forwarding interference. By analyzing the modulation pattern of the interference signal in the radar echo, extracting key parameters such as the interference sampling period and forwarding period, constructing an interference suppression operator, and using this to simulate and generate the spectrum of an ideal point target under the corresponding interference conditions, a point scatter function under the corresponding interference conditions is generated. This function can accurately characterize the response distortion features of the ideal point target in the range direction, thus serving as prior information about interference for auxiliary information in HRRP identification under interference conditions.

[0064] In step S103, based on the pre-built radar target recognition model, multi-scale feature extraction is performed on the high-resolution range image of the target and the prior information of the interference to obtain the first multi-scale feature corresponding to the high-resolution range image of the target and the second multi-scale feature corresponding to the prior information of the interference. Furthermore, the first multi-scale feature and the second multi-scale feature are enhanced using a preset channel attention mechanism and a spatial attention mechanism to generate the corresponding first enhanced feature and the second enhanced feature.

[0065] Furthermore, in order to effectively extract the rich structural information in HRRP and enhance the model's ability to focus on key information in HRRP, embodiments of this application can also construct a multi-scale feature extraction network layer based on channel-spatial attention enhancement, namely MSFEM. MSFEM integrates multi-scale convolution operations and channel-spatial attention mechanisms, achieving full feature extraction of input features under different receptive fields with almost no increase in the number of model parameters.

[0066] Optionally, in one embodiment of this application, based on a pre-built radar target recognition model, multi-scale feature extraction is performed on the high-resolution range image of the target and the prior information of interference to obtain a first multi-scale feature corresponding to the high-resolution range image of the target and a second multi-scale feature corresponding to the prior information of interference. Furthermore, a preset channel attention mechanism and spatial attention mechanism are used to enhance the first and second multi-scale features to generate corresponding first and second enhanced features. This includes: inputting the high-resolution range image of the target and the prior information of interference into a pre-built multi-scale feature network layer; performing convolution operations on the high-resolution range image of the target and the prior information of interference using convolution kernels of different scales in parallel based on the multi-scale feature network layer to obtain the first and second multi-scale features; and enhancing the first and second multi-scale features in the spatial dimension respectively. Global average pooling is performed on the multi-scale features and the second multi-scale features to obtain the corresponding channel weight features. The channel weight features are then input into a preset multilayer perceptron to model the nonlinear relationship between channels. Based on the nonlinear relationship between channels and a preset activation function, corresponding channel attention weights are generated. The channel attention weights are then multiplied element-wise with the first multi-scale features and the second multi-scale features to obtain the corresponding channel attention output features. Average pooling is performed on the channel attention output features along the channel dimension to obtain the corresponding spatial weight features. Based on the spatial weight features, the multilayer perceptron, and the preset activation function, corresponding spatial attention weights are generated. Finally, the spatial attention weights are multiplied element-wise with the channel attention output features to generate the first enhancement features and the second enhancement features.

[0067] It should be noted that the multi-scale feature extraction network layer module based on channel-spatial attention enhancement in this embodiment mainly consists of three parts: a multi-scale feature extraction structure, a channel attention network layer, and a spatial attention network layer, as follows: Figure 4 As shown, the details are as follows:

[0068] 1. Multi-scale feature extraction structure:

[0069] In the embodiments of this application, the multi-scale feature extraction structure can first use convolution kernels of different scales in parallel to perform convolution operations on the input features (i.e., the high-resolution range image of the target and the prior information of interference) to obtain feature representations under different receptive fields. Let the input features be:

[0070] F∈R C×L

[0071] Where C represents the number of channels; L represents the length of the feature sequence.

[0072] The features corresponding to the high-resolution range image of the target and the prior information of the interference will be fed into n convolutional branches with different receptive fields. Specifically, each branch uses a convolutional branch of size 1×K. i One-dimensional convolution kernel (where i = 1, 2, ..., n, K) i The input features are convolved with the receptive field size of the i-th convolutional kernel to generate a feature representation at the corresponding scale.

[0073] To fuse output features from different receptive fields, embodiments of this application can combine all output feature maps {F1, F2, ..., F...} n A concatenation operation is performed along the channel dimension to form a unified intermediate feature representation (i.e., the first multi-scale feature and the second multi-scale feature). multi =Concat(F1,F2,…,F n ).

[0074] 2. Channel Attention Network Layer:

[0075] Those skilled in the art should understand that the channel attention network layer aims to explore the relative importance between different channel features. By explicitly modeling the dependencies of the channel dimensions, it guides the network to focus on feature channels that are more discriminative for the current task, thereby improving the quality of the overall feature representation.

[0076] Specifically, this channel attention network layer can first process the input features (i.e., the first multi-scale feature and the second multi-scale feature) F multi Global average pooling is performed in the spatial dimension to obtain the channel weight features. Secondly, the embodiments of this application can The data is sent to a multilayer perceptron (MLP) consisting of two fully connected layers to model the nonlinear relationships between channels, and then activated by a sigmoid function to generate channel attention weights M. c Finally, the channel attention weights are correlated with the input features F. multi Multiply each channel element by element to enhance the features of key channels and suppress redundant channels.

[0077] In summary, the output features of the channel attention network layer (i.e., the channel attention output features) are as follows:

[0078]

[0079] Where σ represents the Sigmoid activation function; AvgPool represents the average pooling in the spatial dimension; This represents the element-wise multiplication operation under the broadcast mechanism.

[0080] 3. Spatial Attention Network Layer:

[0081] Understandably, spatial attention network layers aim to focus on the salient regions of input features in spatial location. By introducing an attention mechanism in the spatial dimension, the network is guided to focus on discriminative spatial locations, thereby enhancing the network's ability to extract features from key regions.

[0082] Specifically, this spatial attention network layer can first process the input features (i.e., channel attention output features) F ′ Average pooling is performed along the channel dimension to obtain spatial weight features. Secondly Spatial attention weights M are generated after activation by a multilayer perceptron and a sigmoid function. s Finally, M s With input features (i.e., channel attention output features) F ′ Element-wise multiplication is performed positionally to enhance the features of the target region and suppress irrelevant information. Therefore, the output features of the spatial attention module (i.e., the first and second enhanced features) are:

[0083]

[0084] Here, AvgPool represents average pooling along the channel dimension.

[0085] Therefore, the embodiments of this application construct a multi-scale channel-spatial attention feature extraction network layer, extract features under different receptive fields with parallel convolutional branches, and integrate channel attention and spatial attention mechanisms to enhance the network's ability to focus on key feature regions, thereby effectively improving the precision and discriminativeness of HRRP representation, helping to suppress redundant interference features, and highlighting the effective contour information of the target.

[0086] In step S104, based on a preset cross-attention mechanism, the first enhanced feature and the second enhanced feature are fused to generate the corresponding fused feature. Based on a preset cross-entropy loss function and a supervised contrast loss function, a corresponding hybrid loss function is constructed. The hybrid loss function is used to constrain the fused feature to train the radar target recognition model. In the online recognition stage, the trained radar target recognition model is used to perform the corresponding radar target recognition operation.

[0087] Subsequently, since traditional feature fusion methods are difficult to effectively model the semantic relationship between HRRP features and interference priors, embodiments of this application can construct interference prior information and design a cross-attention mechanism to realize the dynamic interactive fusion of HRRP features and interference prior information, thereby enhancing the robustness of the model in HRRP recognition under interference environments.

[0088] Optionally, in one embodiment of this application, based on a preset cross-attention mechanism, a first enhanced feature and a second enhanced feature are fused to generate a corresponding fused feature, including: performing a linear mapping on the first enhanced feature and the second enhanced feature to convert the first enhanced feature and the second enhanced feature into corresponding query vectors, key vectors and value vectors, respectively; adding preset sine and cosine position codes to the query vectors, key vectors and value vectors, respectively, to obtain corresponding position-coded query vectors, position-coded key vectors and position-coded value vectors, and calculating an attention matrix based on the position-coded query vectors and position-coded key vectors; multiplying the attention matrix with the position-coded value vectors to obtain fused attention-aware features, and performing residual connections and layer normalization operations on the fused attention-aware features and the first enhanced feature to obtain corresponding normalization results; and sending the normalization results to a preset feedforward network to output the fused feature.

[0089] It should be noted that after the multi-scale feature extraction network layer completes the feature extraction of HRRP and interference prior information, this embodiment of the application needs to fuse the features of these two parts (i.e., the first enhanced feature and the second enhanced feature). Based on different feature fusion methods, this embodiment of the application can be divided into various operational mechanisms. The core lies in how to effectively integrate feature representations from different sources or levels while preserving semantic information.

[0090] As one possible approach, embodiments of this application may employ a cross-attention mechanism to achieve the fusion of HRRP and interference prior information. This mechanism guides the network to focus on semantically consistent location regions between HRRP and interference prior information during the fusion process by constructing dynamic dependencies between different features, thereby enhancing the expressive power of the fused features.

[0091] Specifically, let the two input features be F HRRP (i.e., the first enhancement feature) and F Prior (i.e., the second enhanced feature), in this embodiment of the application, it can first be converted into a query vector Q, a key vector K, and a value vector V respectively through linear mapping, that is:

[0092] Q = F HRRP W Q K = F Prior W K V = F Prior W V

[0093] Among them, W Q W K W V is a learnable parameter matrix.

[0094] To enhance the model's ability to perceive spatial structures, embodiments of this application can add sine and cosine position codes to Q, K, and V to obtain corresponding position code query vectors, position code key vectors, and position code value vectors, thereby capturing relative position information between different positions. The specific form of the position code is as follows:

[0095]

[0096] Secondly, the attention matrix in this application embodiment can be calculated using the following formula:

[0097]

[0098] Where, d k The dimension of the key vector is used for scaling to stabilize the gradient, avoid gradient vanishing or exploding, and thus improve the stability and convergence speed of model training.

[0099] Subsequently, the attention matrix is ​​multiplied by the position encoding value vector V to obtain the fused attention-aware features (i.e., fused attention-aware features), as shown in the following equation:

[0100]

[0101] Subsequently, embodiments of this application will fuse attention-aware features F attention With the first enhanced feature F HRRPAfter residual connections and layer normalization, the data is sent to the FFN (Feed-Forward Network) to further enhance its nonlinear modeling capability. The final output features (i.e., output fusion features) are as follows:

[0102] F1 = LayerNorm(F HRRP +F attention )

[0103] F out =LayerNorm(F1+FFN(F1))

[0104] Therefore, in order to achieve effective fusion of HRRP features and interference prior information, the embodiments of this application introduce a cross-attention mechanism to dynamically construct the semantic mapping relationship between HRRP features and interference prior information. This mechanism uses HRRP features as the query and interference prior information as the key and value. Through attention-weighted calculation, it adaptively mines the correlation between the two in the semantic space, thereby achieving more accurate feature fusion. This effectively overcomes the problem of insufficient semantic association modeling ability of traditional fusion methods when processing different information, and significantly improves the robust recognition ability of the model for HRRP in the main lobe interference environment.

[0105] Optionally, in one embodiment of this application, the mathematical expression of the hybrid loss function is:

[0106] L total = (1-λ)·L CEL +λ·L SCL

[0107] Among them, L total L represents the mixed loss function; CEL L represents the cross-entropy loss function; SCL Let represent the supervised comparison loss function; 1-λ represent the hyperparameters corresponding to the cross-entropy loss function; λ represent the hyperparameters corresponding to the supervised comparison loss function.

[0108] During the training phase, in order to improve the intra-class compactness and inter-class separability of HRRP and enhance the model's robustness in interference scenarios, this application embodiment may introduce a hybrid loss function based on cross-entropy loss and supervised contrastive loss, as described below:

[0109] 1. Cross-entropy loss function:

[0110] Those skilled in the art should understand that cross-entropy loss is one of the most commonly used supervised loss functions in classification tasks, used to measure the difference between the probability distribution of the model's output and the true label distribution. Its core idea is to improve the model's discriminative ability in classification tasks by minimizing the cross-entropy distance between the predicted and true distributions. The cross-entropy loss function is defined as:

[0111]

[0112] Where N represents the number of samples in the mini-batch; M is the total number of categories; y ic p represents the true label of the i-th sample on the c-th position; ic This represents the model's predicted label for the sample in class c. During model training, the cross-entropy loss function effectively guides the optimization of model parameters by quantifying the difference between the model's predicted distribution and the true label distribution.

[0113] The cross-entropy loss reaches its minimum when the model's predictions are completely consistent with the true labels; however, the loss increases significantly when the predictions deviate from the true labels, especially when the model has a high confidence level in the wrong category, thus imposing a stronger penalty on misjudgments.

[0114] 2. Supervised contrastive loss function:

[0115] In the field of machine learning, traditional contrastive learning, as an unsupervised learning paradigm, guides the model to learn discriminative feature representations by constructing similarity and difference relationships between samples. Traditional contrastive learning is self-supervised, utilizing the context of the data itself or augmented samples to construct positive and negative sample pairs. For example, it generates similar samples as positive examples by performing data augmentation operations such as random cropping and rotation on images, while other samples in the same batch are used as negative examples.

[0116] However, this unsupervised contrastive learning approach has limitations in semantic understanding, especially in classification tasks with clearly defined target categories. It lacks effective use of label information, making it difficult for the model to learn accurate category-discriminating features.

[0117] Supervised contrastive loss, building upon traditional contrastive learning, utilizes label information to explicitly model the semantic similarity between samples, guiding the model to optimize the feature distribution of samples in the embedding space. In supervised contrastive learning, each sample is considered an anchor point, forming positive sample pairs with other samples of the same class and negative sample pairs with samples of different classes. For a sample set I in a mini-batch, the positive sample set P(i) for each anchor point i∈I consists of all samples with the same label as the anchor point itself, while the negative sample set N(i) consists of samples from other classes. Figure 5As shown, the goal of supervised contrastive loss is to maximize the similarity between the anchor point and its positive samples, while minimizing its similarity with negative samples, thereby achieving intra-class compactness and inter-class separability in the embedding space. Its loss function is:

[0118]

[0119] Where A(i) contains all other samples except sample i; z i Let be the feature representation of sample i in the projection space; sim(·) represents the cosine similarity function, which maps the features to the interval [-1,1], and the larger the value, the higher the similarity; τ is the temperature coefficient used to adjust the smoothness of the feature distribution.

[0120] Finally, the hybrid loss function in this embodiment is a weighted combination of the cross-entropy loss function and the supervised contrastive loss function, where λ is a hyperparameter that adjusts the contribution of the two loss functions. The mathematical expression of this hybrid loss function is:

[0121] L total = (1-λ)·L CEL +λ·L SCL

[0122] Among them, L total L represents the mixed loss function; CEL L represents the cross-entropy loss function; SCL Let represent the supervised comparison loss function; 1-λ represent the hyperparameters corresponding to the cross-entropy loss function; λ represent the hyperparameters corresponding to the supervised comparison loss function.

[0123] Therefore, the embodiments of this application combine cross-entropy loss and supervised contrastive loss to construct a hybrid loss function. The cross-entropy loss is used to improve classification accuracy, while the contrastive loss strengthens the intra-class compactness and inter-class separability of the feature space by bringing similar samples closer together and pushing apart the embedding distance of dissimilar samples. This hybrid loss function can promote the model to learn more robust and discriminative feature representations in the presence of interfering prior auxiliary information, thereby further improving the robustness of the model in interference scenarios.

[0124] The recognition performance of this application is analyzed experimentally below with reference to the accompanying drawings.

[0125] To specifically compare the HRRP recognition performance of the recognition algorithm (i.e., radar target recognition model) of this application embodiment with that of existing technologies in the face of intermittent sampling and forwarding interference, this application embodiment constructed an HRRP dataset for 6 types of targets under 21 interference parameters and their corresponding interference prior information. For this dataset, this application embodiment selected HRRP data under 15 interference parameters for training and randomly divided it into training and test sets in an 8:2 ratio; at the same time, HRRP data under the other 6 interference parameters were set as an independent test set to evaluate the algorithm's generalization ability under unknown interference parameters.

[0126] For evaluation metrics, this application uses the recognition accuracy of the radar target recognition model on the test set as the main indicator to evaluate its recognition performance, and its accuracy on an independent test set as the basis for measuring the generalization ability of the radar target recognition model. Therefore, the evaluation metrics used in this application are test set accuracy and generalization.

[0127] Table 1 shows a comparison of the recognition performance of the embodiments of this application with TARAN, TACNN, SAN, Ptrans and ConvLSTM methods.

[0128] Table 1

[0129]

[0130] As shown in Table 1, the test set accuracy and generalization of the embodiments of this application are both optimal. Regarding test set accuracy, the radar target recognition model of the embodiments of this application achieves 95.01%, which is significantly better than other comparative algorithms, indicating that the radar target recognition model can maintain excellent recognition capabilities even in interference environments. Simultaneously, the radar target recognition model of the embodiments of this application achieves an accuracy of 83.97% on the independent test set, verifying its good generalization performance and demonstrating strong adaptability and robustness when facing unseen interference parameters.

[0131] To compare the recognition performance of different recognition algorithms for various target types, embodiments of this application also demonstrate the confusion matrices of different algorithms for each target type. Among them, Figure 6 A schematic diagram illustrating the identification of confusion matrices on test sets for different algorithms. Figure 7 This is a schematic diagram illustrating the generalization confusion matrix of different algorithms. For example... Figure 6 As shown, in the confusion matrix corresponding to the radar target recognition model of this application embodiment, all target categories exhibit a significant diagonal advantage, with recognition accuracy exceeding 90%, and almost none are misclassified as other categories, demonstrating a clear advantage over other comparison methods. Figure 7As shown, although the overall recognition difficulty of the model increases under unknown interference scenarios, the radar target recognition model in this application embodiment still maintains strong recognition performance, with the diagonal elements of most targets still dominating. In contrast, the comparison method exhibits more severe confusion with similar targets in generalization scenarios, indicating a stronger dependence on training data and limited generalization ability.

[0132] In summary, the embodiments of this application construct a point distribution function that can characterize interference characteristics and HRRP distortion patterns, effectively extracting prior interference information. By combining a multi-scale feature extraction module with a cross-attention fusion mechanism, a deep fusion of HRRP features and prior interference information is achieved. This significantly improves the target recognition accuracy and robustness of the model under typical interference conditions such as intermittent sampling and forwarding. The model is well-suited for high-resolution radar automatic target recognition tasks in complex electromagnetic interference environments and has broad engineering application prospects.

[0133] According to the embodiments of this application, the radar target recognition method based on prior information-assisted main lobe interference in the online recognition stage acquires the echo signal and baseband waveform of the radar target under interference conditions, extracts interference features using the echo signal, generates an interference suppression operator, and performs echo interference suppression and target detection operations based on the echo signal, baseband waveform, and interference suppression operator to obtain the interference-suppressed echo and target information. A high-resolution range profile of the target is then generated based on the interference-suppressed echo and target information. Based on the baseband waveform, target information, and interference suppression operator, equivalent processing of the baseband waveform is performed to generate a point spread function under interference, and interference prior information corresponding to the high-resolution range profile of the target is constructed based on the interference point spread function. Finally, based on a pre-built radar target recognition model, the high-resolution range profile of the target is... Multi-scale feature extraction is performed on the high-resolution range image of the target and prior interference information to obtain the first multi-scale feature corresponding to the target high-resolution range image and the second multi-scale feature corresponding to the prior interference information. Pre-set channel attention and spatial attention mechanisms are used to enhance the first and second multi-scale features to generate corresponding enhanced features. Based on a pre-set cross-attention mechanism, the first and second enhanced features are fused to generate corresponding fused features. A corresponding hybrid loss function is constructed based on a pre-set cross-entropy loss function and supervised contrast loss function, and the hybrid loss function is used to constrain the fused features to train the radar target recognition model. The trained radar target recognition model is then used to perform corresponding radar target recognition operations during the online recognition phase. This application effectively extracts prior interference information by constructing a point spread function that can characterize interference characteristics and HRRP distortion patterns. Combined with the multi-scale feature extraction module and the cross-attention fusion mechanism, it achieves deep fusion of HRRP features and prior interference information, thereby significantly improving the target recognition accuracy and robustness of the model under typical interference conditions such as intermittent sampling and forwarding.

[0134] Figure 8 This is a flowchart illustrating a radar target recognition method based on prior information-assisted main lobe interference, provided as an embodiment of this application, for use in the online recognition stage.

[0135] like Figure 8 As shown, the radar target recognition method under main lobe interference based on prior information includes the following steps:

[0136] In step S801, the high-resolution range image of the target under test and the radar echo signal corresponding to the target radar are acquired.

[0137] In step S802, key interference features corresponding to the radar echo signal are extracted to construct corresponding target interference prior information based on the key interference features.

[0138] In step S803, the high-resolution range image of the target to be tested and the prior information of the target interference are input into the pre-trained radar target recognition model to output the corresponding radar target recognition result.

[0139] The radar target recognition method based on prior information-assisted main lobe interference proposed in the online recognition stage according to the embodiments of this application obtains the target high-resolution range image and radar echo signal of the target radar; extracts key interference features corresponding to the radar echo signal to construct corresponding target interference prior information based on the key interference features; and inputs the target high-resolution range image and target interference prior information into a pre-trained radar target recognition model to output the corresponding radar target recognition result. This application effectively extracts interference prior information by constructing a point spread function that can characterize interference characteristics and HRRP distortion law, and combines a multi-scale feature extraction module with a cross-attention fusion mechanism to achieve deep fusion of HRRP features and interference prior information, thereby significantly improving the target recognition accuracy and robustness of the model under typical interference conditions such as intermittent sampling and forwarding.

[0140] Secondly, with reference to the accompanying drawings, a radar target identification device based on prior information-assisted main lobe interference according to an embodiment of this application is described.

[0141] Figure 9 This is a block diagram of a radar target recognition device based on prior information-assisted main lobe interference applied in the offline training phase according to an embodiment of this application.

[0142] like Figure 9 As shown, the radar target recognition device 10 based on prior information-assisted main lobe interference applied in the offline training stage includes: an interference suppression module 101, a baseband waveform equivalent processing module 102, a feature extraction module 103, a feature fusion module 104, and a training module 105.

[0143] The interference suppression module 101 is used to acquire the echo signal and baseband waveform of the radar target under interference conditions, extract interference features using the echo signal, generate an interference suppression operator, and perform echo interference suppression and target detection operations based on the echo signal, baseband waveform and interference suppression operator to obtain the interference-suppressed echo and target information, and generate a high-resolution range image of the target based on the interference-suppressed echo and target information.

[0144] The baseband waveform equivalent processing module 102 is used to perform baseband waveform equivalent processing based on the baseband waveform, target information and interference suppression operator to generate interference sub-point scatter function, and construct interference prior information corresponding to the target high-resolution range image based on the interference sub-point scatter function.

[0145] The feature extraction module 103 is used to perform multi-scale feature extraction on the high-resolution range image of the target and the prior information of interference based on the pre-built radar target recognition model, so as to obtain the first multi-scale feature corresponding to the high-resolution range image of the target and the second multi-scale feature corresponding to the prior information of interference. Furthermore, the first multi-scale feature and the second multi-scale feature are enhanced by the preset channel attention mechanism and spatial attention mechanism to generate the corresponding first enhanced feature and the second enhanced feature.

[0146] The feature fusion module 104 is used to fuse the first enhanced feature and the second enhanced feature based on a preset cross-attention mechanism to generate the corresponding fused feature.

[0147] The training module 105 is used to construct a corresponding hybrid loss function based on a preset cross-entropy loss function and a supervised contrast loss function, and to use the hybrid loss function to constrain the fused features in order to train the radar target recognition model, so as to perform the corresponding radar target recognition operation using the trained radar target recognition model in the online recognition stage.

[0148] Optionally, in one embodiment of this application, the feature extraction module 103 includes: a convolution unit, a first pooling unit, a first multiplication unit, a second pooling unit, and a second multiplication unit.

[0149] The convolutional unit is used to input the high-resolution range image of the target and the prior information of the interference into the pre-constructed multi-scale feature network layer. Based on the multi-scale feature network layer, convolutional kernels of different scales are used in parallel to perform convolution operations on the high-resolution range image of the target and the prior information of the interference to obtain the first multi-scale feature and the second multi-scale feature.

[0150] The first pooling unit is used to perform global average pooling on the first multi-scale feature and the second multi-scale feature in the spatial dimension to obtain the corresponding channel weight features. The channel weight features are then input into a preset multilayer perceptron to model the nonlinear relationship between channels. Based on the nonlinear relationship between channels and the preset activation function, the corresponding channel attention weights are generated.

[0151] The first multiplication unit is used to perform element-wise multiplication of the channel attention weights with the first multiscale feature and the second multiscale feature respectively to obtain the corresponding channel attention output features.

[0152] The second pooling unit is used to perform average pooling on the channel attention output features along the channel dimension to obtain the corresponding spatial weight features, and to generate the corresponding spatial attention weights based on the spatial weight features, the multilayer perceptron and the preset activation function.

[0153] The second multiplication unit is used to perform position-wise element-wise multiplication of the spatial attention weights and the channel attention output features to generate the first and second enhanced features.

[0154] Optionally, in one embodiment of this application, the feature fusion module 104 includes: a linear mapping unit, a calculation unit, a normalization unit, and an output unit.

[0155] The linear mapping unit is used to perform linear mapping on the first enhanced feature and the second enhanced feature, so as to convert the first enhanced feature and the second enhanced feature into corresponding query vector, key vector and value vector respectively.

[0156] The calculation unit is used to add the preset sine and cosine position codes to the query vector, key vector, and value vector respectively to obtain the corresponding position code query vector, position code key vector, and position code value vector, and to calculate the attention matrix based on the position code query vector and position code key vector.

[0157] The normalization unit is used to multiply the attention matrix with the position encoding value vector to obtain the fused attention-aware feature, and to perform residual connection and layer normalization operations on the fused attention-aware feature and the first enhanced feature to obtain the corresponding normalization result.

[0158] The output unit is used to send the normalization result to a preset feedforward network to output the fused features.

[0159] Optionally, in one embodiment of this application, the mathematical expression of the hybrid loss function is:

[0160] L total = (1-λ)·L CEL +λ·L SCL

[0161] Among them, L total L represents the mixed loss function; CEL L represents the cross-entropy loss function; SCL Let represent the supervised comparison loss function; 1-λ represent the hyperparameters corresponding to the cross-entropy loss function; λ represent the hyperparameters corresponding to the supervised comparison loss function.

[0162] It should be noted that the foregoing explanation of the radar target recognition method based on prior information-assisted main lobe interference applied in the offline training stage also applies to the radar target recognition device based on prior information-assisted main lobe interference applied in the offline training stage of this embodiment, and will not be repeated here.

[0163] According to the embodiments of this application, a radar target recognition device based on prior information assisted under main lobe interference for offline training includes an interference suppression module 101, used to acquire the echo signal and baseband waveform of the radar target under interference conditions, extract interference features using the echo signal, generate an interference suppression operator, and perform echo interference suppression and target detection operations based on the echo signal, baseband waveform, and interference suppression operator to obtain interference-suppressed echo and target information, and generate a high-resolution range image of the target based on the interference-suppressed echo and target information; a baseband waveform equivalent processing module 102, used to perform baseband waveform equivalent processing based on the baseband waveform, target information, and interference suppression operator to generate an interference-based point scatter function, and construct interference prior information corresponding to the high-resolution range image of the target based on the interference-based point scatter function; and a feature extraction module 103, used to extract radar target recognition information based on a pre-constructed feature extraction module. The model extracts multi-scale features from the high-resolution range image of the target and prior information about interference to obtain the first multi-scale features corresponding to the high-resolution range image of the target and the second multi-scale features corresponding to the prior information about interference. It then uses preset channel attention and spatial attention mechanisms to enhance the first and second multi-scale features, generating corresponding first and second enhanced features. A feature fusion module 104 fuses the first and second enhanced features based on a preset cross-attention mechanism to generate corresponding fused features. A training module 105 constructs a corresponding hybrid loss function based on preset cross-entropy loss and supervised contrast loss functions, and uses the hybrid loss function to constrain the fused features to train the radar target recognition model. This trained radar target recognition model is then used to perform corresponding radar target recognition operations during the online recognition phase. This application constructs a point distribution function that can characterize interference features and HRRP distortion patterns, effectively extracting prior information about interference. By combining a multi-scale feature extraction module with a cross-attention fusion mechanism, it achieves deep fusion of HRRP features and prior information about interference, thereby significantly improving the target recognition accuracy and robustness of the model under typical interference conditions such as intermittent sampling and forwarding.

[0164] Figure 10 This is a block diagram of a radar target recognition device based on prior information-assisted main lobe interference applied in the online pre-stage of this application.

[0165] like Figure 10 As shown, the radar target identification device 20 based on prior information-assisted main lobe interference applied in the online identification stage includes: an acquisition module 201, a construction module 202, and an identification module 203.

[0166] The acquisition module 201 is used to acquire the target high-resolution range image and radar echo signal of the target radar corresponding to the target radar.

[0167] The construction module 202 is used to extract key interference features corresponding to radar echo signals, so as to construct corresponding target interference prior information based on the key interference features.

[0168] The identification module 203 is used to input the high-resolution range image of the target to be tested and the prior information of the target interference into the pre-trained radar target identification model, so as to output the corresponding radar target identification result.

[0169] It should be noted that the foregoing explanation of the radar target recognition method based on prior information-assisted main lobe interference applied in the online recognition stage also applies to the radar target recognition device based on prior information-assisted main lobe interference applied in the online recognition stage of this embodiment, and will not be repeated here.

[0170] According to the embodiments of this application, a radar target recognition device based on prior information-assisted main lobe interference for the online recognition stage includes an acquisition module 201 for acquiring the target high-resolution range image and radar echo signal of the target radar; a construction module 202 for extracting key interference features corresponding to the radar echo signal to construct corresponding target interference prior information based on the key interference features; and a recognition module 203 for inputting the target high-resolution range image and target interference prior information of the target radar into a pre-trained radar target recognition model to output the corresponding radar target recognition result. This application effectively extracts interference prior information by constructing a point spread function that can characterize interference characteristics and HRRP distortion patterns. Combined with a multi-scale feature extraction module and a cross-attention fusion mechanism, it achieves deep fusion of HRRP features and interference prior information, thereby significantly improving the target recognition accuracy and robustness of the model under typical interference conditions such as intermittent sampling and forwarding.

[0171] Figure 11 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0172] The memory 1101, the processor 1102, and the computer program stored on the memory 1101 and executable on the processor 1102.

[0173] When the processor 1102 executes the program, it implements the radar target recognition method under main lobe interference based on prior information assistance provided in the above embodiments.

[0174] Furthermore, electronic devices also include:

[0175] Communication interface 1103 is used for communication between memory 1101 and processor 1102.

[0176] The memory 1101 is used to store computer programs that can run on the processor 1102.

[0177] The memory 1101 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0178] If the memory 1101, processor 1102, and communication interface 1103 are implemented independently, then the communication interface 1103, memory 1101, and processor 1102 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be divided into address buses, data buses, control buses, etc. For ease of representation, Figure 11 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0179] Optionally, in a specific implementation, if the memory 1101, processor 1102, and communication interface 1103 are integrated on a single chip, then the memory 1101, processor 1102, and communication interface 1103 can communicate with each other through an internal interface.

[0180] The processor 1102 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0181] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the radar target recognition method under main lobe interference based on prior information assistance as described above.

[0182] This application also provides a computer program product, including a computer program, which, when executed, is used to implement the above-described radar target recognition method under main lobe interference based on prior information assistance.

[0183] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0184] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0185] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0186] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0187] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0188] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0189] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0190] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for radar target recognition under main-lobe interference based on prior information assistance, applied to an offline training stage, characterized in that, Includes the following steps: The system acquires the echo signal and baseband waveform of the radar target under interference conditions, and extracts interference features using the echo signal, including the interference sampling period and the forwarding period, to generate an interference suppression operator. Based on the echo signal, the baseband waveform, and the interference suppression operator, the system performs echo interference suppression and target detection operations to obtain the interference-suppressed echo and target information. The system then generates a high-resolution range profile of the target based on the interference-suppressed echo and the target information. Based on the baseband waveform, the target information, and the interference suppression operator, equivalent processing of the baseband waveform is performed to generate a point scatter function under interference. The point scatter function under interference is used to describe the response distortion characteristics of an ideal point target in the range direction under the current interference conditions, and interference prior information corresponding to the high-resolution range image of the target is constructed according to the point scatter function under interference. Based on a pre-built radar target recognition model, multi-scale feature extraction is performed on the high-resolution range image of the target and the prior information of the interference to obtain the first multi-scale feature corresponding to the high-resolution range image of the target and the second multi-scale feature corresponding to the prior information of the interference. Furthermore, the first multi-scale feature and the second multi-scale feature are enhanced using a preset channel attention mechanism and spatial attention mechanism to generate the corresponding first enhanced feature and second enhanced feature. Based on a preset cross-attention mechanism, the first enhanced feature and the second enhanced feature are fused to generate a corresponding fused feature. Based on a preset cross-entropy loss function and a supervised contrast loss function, a corresponding hybrid loss function is constructed. The hybrid loss function is used to constrain the fused feature to train the radar target recognition model. In the online recognition stage, the trained radar target recognition model is used to perform the corresponding radar target recognition operation.

2. The method of claim 1, wherein, The pre-built radar target recognition model extracts multi-scale features from the high-resolution range image of the target and the prior information of the interference to obtain a first multi-scale feature corresponding to the high-resolution range image of the target and a second multi-scale feature corresponding to the prior information of the interference. Furthermore, it uses preset channel attention and spatial attention mechanisms to enhance the first and second multi-scale features to generate corresponding first and second enhanced features, including: The target high-resolution range image and the interference prior information are input into a pre-constructed multi-scale feature network layer. Based on the multi-scale feature network layer, convolution operations of different scales are performed on the target high-resolution range image and the interference prior information in parallel to obtain the first multi-scale feature and the second multi-scale feature. Global average pooling is performed on the first multi-scale feature and the second multi-scale feature in the spatial dimension to obtain the corresponding channel weight features. The channel weight features are then input into a preset multilayer perceptron to model the nonlinear relationship between channels. Based on the nonlinear relationship between channels and a preset activation function, the corresponding channel attention weights are generated. The channel attention weights are multiplied element-wise with the first multi-scale feature and the second multi-scale feature respectively to obtain the corresponding channel attention output features; The channel attention output features are averaged in the channel dimension to obtain the corresponding spatial weight features, and the corresponding spatial attention weights are generated based on the spatial weight features, the multilayer perceptron and the preset activation function. The spatial attention weights are multiplied element-wise with the channel attention output features to generate the first enhanced feature and the second enhanced feature.

3. The method of claim 1, wherein, The method based on a preset cross-attention mechanism fuses the first enhanced feature and the second enhanced feature to generate a corresponding fused feature, including: A linear mapping is performed on the first enhanced feature and the second enhanced feature to convert the first enhanced feature and the second enhanced feature into corresponding query vector, key vector and value vector, respectively; Preset sine and cosine position codes are added to the query vector, the key vector, and the value vector respectively to obtain the corresponding position code query vector, position code key vector, and position code value vector. Based on the position code query vector and the position code key vector, the attention matrix is ​​calculated. The attention matrix is ​​multiplied by the position encoding value vector to obtain the fused attention-aware feature, and residual connection and layer normalization operations are performed on the fused attention-aware feature and the first enhanced feature to obtain the corresponding normalization result. The normalization result is sent to a preset feedforward network to output the fused features.

4. The method of claim 1, wherein, The mathematical expression for the hybrid loss function is: in, This represents the hybrid loss function; This represents the cross-entropy loss function; Represents the supervised contrastive loss function; 1- This represents the hyperparameters corresponding to the cross-entropy loss function; This represents the hyperparameters corresponding to the supervised comparison loss function.

5. A radar target recognition method under main lobe interference based on prior information assistance, applied to an online recognition stage, characterized in that, The radar target recognition method based on prior information-assisted main lobe interference, as described in any one of claims 1-4, is adopted for the offline training phase, wherein the method applied to the online recognition phase includes the following steps: Acquire the high-resolution range image and radar echo signal of the target target corresponding to the target radar; Extract the key interference features corresponding to the radar echo signal, and construct the corresponding target interference prior information based on the key interference features; The high-resolution range image of the target under test and the prior information of the target interference are input into the pre-trained radar target recognition model to output the corresponding radar target recognition result.

6. An apparatus for radar target recognition under main-lobe interference based on priori information assistance, applied to an offline training stage, characterized in that, include: The interference suppression module is used to acquire the echo signal and baseband waveform of the radar target under interference conditions, and to extract interference features using the echo signal, including the interference sampling period and the forwarding period, to generate an interference suppression operator. Based on the echo signal, the baseband waveform and the interference suppression operator, the module performs echo interference suppression and target detection operations to obtain the interference-suppressed echo and target information, and generates a high-resolution range profile of the target based on the interference-suppressed echo and the target information. The baseband waveform equivalent processing module is used to perform baseband waveform equivalent processing based on the baseband waveform, the target information and the interference suppression operator to generate an interference-based point scatter function. The interference-based point scatter function is used to describe the response distortion characteristics of an ideal point target in the range direction under the current interference conditions, and to construct the interference prior information corresponding to the high-resolution range image of the target based on the interference-based point scatter function. The feature extraction module is used to perform multi-scale feature extraction on the high-resolution range image of the target and the interference prior information based on a pre-built radar target recognition model, so as to obtain the first multi-scale feature corresponding to the high-resolution range image of the target and the second multi-scale feature corresponding to the interference prior information. The module also uses a preset channel attention mechanism and spatial attention mechanism to enhance the first multi-scale feature and the second multi-scale feature to generate the corresponding first enhanced feature and second enhanced feature. The feature fusion module is used to fuse the first enhanced feature and the second enhanced feature based on a preset cross-attention mechanism to generate a corresponding fused feature; The training module is used to construct a corresponding hybrid loss function based on a preset cross-entropy loss function and a supervised contrast loss function, and to use the hybrid loss function to constrain the fused features in order to train the radar target recognition model, so as to perform the corresponding radar target recognition operation using the trained radar target recognition model in the online recognition stage.

7. An apparatus for radar target recognition under main-lobe interference based on priori information assistance, applied to an online recognition stage, characterized in that, The radar target identification device under main lobe interference as described in claim 6, applied in the offline training phase, wherein the device applied in the online identification phase includes: The acquisition module is used to acquire the target high-resolution range image and radar echo signal of the target target corresponding to the target radar; A construction module is used to extract key interference features corresponding to the radar echo signal, so as to construct corresponding target interference prior information based on the key interference features; The identification module is used to input the high-resolution range image of the target under test and the prior information of the target interference into the pre-trained radar target identification model, so as to output the corresponding radar target identification result.

8. An electronic device, comprising: include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the radar target identification method under main lobe interference based on prior information assistance as described in any one of claims 1-4 or 5.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the radar target recognition method under main lobe interference based on prior information assistance as described in any one of claims 1-4 or 5.

10. A computer program product, comprising a computer program, characterized in that, The computer program is executed to implement the radar target identification method under main lobe interference based on prior information assistance as described in any one of claims 1-4 or 5.

Images