Radar radiation source individual identification method and device, computer equipment and storage medium

Abstract

The invention relates to a radar radiation source individual identification method and device, computer equipment and a storage medium, and the method comprises the steps: carrying out the local feature extraction of a preprocessed pulse signal through employing an improved deep residual shrinkage network, obtaining a local feature, carrying out the enhancement of the local feature in a space dimension and a time dimension through employing a space-time fusion network, and carrying out the recognition of a radar radiation source individual. Fusing the enhanced features to obtain fused features, performing dynamic weight calibration on channel dimension features and time dimension features of the fused features through a combined channel calibration mechanism and time calibration by using a dynamic feature calibration network, and fusing the features after weight calibration with local features to obtain features to be identified; and finally, radar radiation source individual identification is carried out by using the classification head according to the to-be-identified features. By adopting the method, the reliability and effectiveness of radar radiation source individual identification in a complex scene can be improved.

Description

Technical Field

[0001] The present application relates to the field of radar signal processing and identification technology, and in particular to a method, apparatus, computer equipment, and storage medium for identifying individual radar radiation sources. Background Art

[0002] In modern electronic countermeasures and intelligent monitoring scenarios, individual identification of radar emitters, as a key technology, plays a decisive role in target monitoring and situational awareness in complex electromagnetic environments.

[0003] However, traditional methods face significant challenges in environments with low signal-to-noise ratios (SNRs) and dynamic interference. For one thing, the traditional one-dimensional radar pulse signal processing process relies on complex mathematical operations such as the Wiener distribution and wavelet transform. This not only requires extremely high technical expertise but also requires significant computational resources and labor costs during the data preprocessing phase, resulting in inefficient implementation. Furthermore, due to the lack of adaptive mechanisms for complex environments, traditional algorithms are susceptible to multipath effects, noise interference, and man-made electromagnetic interference. This makes it difficult to guarantee recognition accuracy and system robustness in practical applications, severely limiting the reliability and effectiveness of radar emitter identification technology in complex scenarios. Summary of the Invention

[0004] Based on this, it is necessary to provide a radar emitter individual identification method, device, computer equipment and storage medium that can improve the reliability and effectiveness in complex scenarios to address the above technical problems.

[0005] A method for identifying an individual radar emitter, the method comprising: Acquire a pulse signal emitted by a radar radiation source, and preprocess the pulse signal to obtain a preprocessed pulse sequence; Using an improved deep residual shrinkage network, extracting local features from the preprocessed pulse sequence to obtain local features; Using a spatiotemporal fusion network, the local features are enhanced in both spatial and temporal dimensions, and then the enhanced features are fused to obtain fused features; Using a dynamic feature calibration network, the channel dimension features and time dimension features of the fusion feature are dynamically weighted by a joint channel calibration mechanism and time calibration, and then the weighted calibrated features are fused with the local features to obtain the features to be identified; The classification head is used to perform individual identification of radar emitters according to the features to be identified, and an identification result is obtained.

[0006] In one embodiment, when preprocessing the pulse signal, the one-dimensional radar pulse signal sequence is mapped to text content in natural language processing, wherein each radar pulse signal envelope is analogous to a sentence in the text, and each amplitude point on the envelope serves as the basic word element that constitutes the sentence.

[0007] In one embodiment, preprocessing the pulse signal includes: performing standardization processing on the pulse signal, locating the main pulse area, extracting the signal and performing data enhancement in sequence, converting the pulse signal into a tensor of the channel dimension, and obtaining the preprocessed pulse sequence.

[0008] In one embodiment, in the improved deep residual shrinkage network: The initial convolution layer uses a specific convolution kernel to extract features from the preprocessed pulse sequence to obtain initial features; A residual shrinkage unit is used to suppress noise from the initial features by dynamically generating a channel-level shrinkage threshold, and a multi-scale fusion unit is used to jointly extract multi-resolution features in the time-frequency domain through parallel convolution paths; The local features are generated according to the time-frequency domain multi-resolution features using a batch normalization layer and an activation function.

[0009] In one embodiment, in the spatiotemporal fusion network: Utilizing CNN network and Transformer network to extract spatial features and temporal features of the local features respectively; The spatial features and temporal features are enhanced respectively by a spatial attention mechanism and a temporal attention mechanism to obtain spatial enhanced features and temporal enhanced features; The spatial enhancement feature and the temporal enhancement feature are dynamically fused using the learned gating parameters to obtain the fused feature.

[0010] In one embodiment, in the dynamic feature calibration network: Channel calibration captures the channel-level statistical features of the fused features through global average pooling and generates a channel weight matrix; Time calibration uses a lightweight deep separable convolutional network to analyze the local temporal contextual relationship of the fused features and generate a time dimension attention weight matrix; Multiplying the channel weight matrix by the time dimension attention weight matrix element by element to generate a spatiotemporal joint attention weight, and using the spatiotemporal joint attention weight to perform point-wise weighting on the fusion feature to obtain a calibrated feature; The calibrated features are residually fused with the local features to obtain the features to be identified.

[0011] In one embodiment, in the classification header: The features to be identified are compressed in the time dimension through global average pooling, and then mapped to category prediction probabilities through a fully connected layer.

[0012] The present application also provides a radar radiation source individual identification device, the device comprising: A signal preprocessing module is used to obtain the pulse signal emitted by the radar radiation source and preprocess the pulse signal to obtain a preprocessed pulse sequence; A feature extraction module is used to extract local features from the preprocessed pulse sequence using an improved deep residual shrinkage network to obtain local features; A spatiotemporal feature fusion module is used to enhance the local features in both spatial and temporal dimensions using a spatiotemporal fusion network, and then fuse the enhanced features to obtain fused features; A dynamic feature calibration module is used to use a dynamic feature calibration network to perform dynamic weight calibration on the channel dimension features and time dimension features of the fusion feature through a joint channel calibration mechanism and time calibration, and then fuse the weight-calibrated features with the local features to obtain the features to be identified; The classification and recognition module is used to use the classification head to perform individual identification of radar radiation sources according to the features to be identified and obtain an identification result.

[0013] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above-mentioned radar radiation source individual identification method when executing the computer program.

[0014] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the above-mentioned radar radiation source individual identification method.

[0015] The above-mentioned radar emitter individual identification method, apparatus, computer device, and storage medium extract local features from preprocessed pulse signals using an improved deep residual shrinkage network. These local features are then enhanced in both spatial and temporal dimensions using a spatiotemporal fusion network. The enhanced features are then fused to form fused features. A dynamic feature calibration network then dynamically weights the channel and temporal dimensions of the fused features using a combined channel calibration mechanism and temporal calibration. The weighted features are then fused with the local features to form the features to be identified. Finally, a classification head is used to identify individual radar emitters based on the features to be identified. This method can improve the reliability and effectiveness of radar emitter individual identification in complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 1 is a flow chart of a method for identifying individual radar emitters in one embodiment; Figure 2 A schematic diagram of a three-stage hybrid architecture network model in one embodiment; Figure 3 Schematic diagram of the structure of an improved deep residual shrinkage network in one embodiment; Figure 4 Schematic diagram of dual feature fusion in a spatiotemporal fusion network in one embodiment; Figure 5 Schematic diagram of the processing process of the spatial attention mechanism and the temporal attention mechanism in the spatiotemporal fusion network in one embodiment, wherein: Figure 5 (a) is a schematic diagram showing the processing of the spatial attention mechanism. Figure 5 (b) Schematic diagram showing the processing process of the temporal attention mechanism; Figure 6 A schematic diagram of a channel calibration process in a dynamic feature calibration network according to an embodiment; Figure 7 A schematic diagram of a time calibration process in a dynamic feature calibration network according to an embodiment; Figure 8 This is a diagram of the DTnet training and testing curves for a navigation radar dataset in an experiment. Figure 8 (a) is a schematic diagram of the loss curve. Figure 8 (b) Schematic diagram of the accuracy curve; Figure 9 This is a diagram of the VGG training and testing curves for a navigation radar dataset in an experiment. Figure 9 (a) is a schematic diagram of the loss curve. Figure 9 (b) Schematic diagram of the accuracy curve; Figure 10 This is a diagram of the Resnet training and testing curves for a navigation radar dataset in an experiment. Figure 10 (a) is a schematic diagram of the loss curve. Figure 10 (b) Schematic diagram of the accuracy curve; Figure 11 This is a diagram of the Alexnet training and testing curves for a navigation radar dataset in an experiment. Figure 11 (a) is a schematic diagram of the loss curve. Figure 11 (b) Schematic diagram of the accuracy curve; Figure 12 A schematic diagram of DTnet features under a navigation radar dataset in an experiment; Figure 13 A schematic diagram of Resnet features in a navigation radar dataset from an experiment; Figure 14 A schematic diagram of VGG features in a navigation radar dataset in an experiment; Figure 15 A schematic diagram of Alexnet features in a navigation radar dataset in an experiment; Figure 16 is a structural block diagram of a radar radiation source individual identification device in one embodiment; Figure 17 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0018] In this application, the existing radar emitter individual identification method has poor adaptability to complex environments and is easily interfered with, resulting in low recognition accuracy and insufficient robustness. Figure 1 As shown, a method for identifying individual radar emitters is provided, which specifically includes the following steps: Step S100: acquiring a pulse signal emitted by a radar radiation source, and preprocessing the pulse signal to obtain a preprocessed pulse sequence.

[0019] Step S110 , using the improved deep residual shrinkage network, extracting local features from the preprocessed pulse sequence to obtain local features.

[0020] In step S120 , the local features are enhanced in both spatial and temporal dimensions using a spatiotemporal fusion network, and the enhanced features are then fused to obtain fused features.

[0021] In step S130, a dynamic feature calibration network is used to perform dynamic weight calibration on the channel dimension features and time dimension features of the fusion features through a joint channel calibration mechanism and time calibration, and then the weight-calibrated features are fused with the local features to obtain the features to be identified.

[0022] Step S140: Using the classification head, the radar emitter is individually identified according to the features to be identified, and an identification result is obtained.

[0023] In this embodiment, a three-stage hybrid architecture network model was designed by integrating techniques such as deep residual shrinkage, multi-scale feature interaction, spatiotemporal attention mechanism, and dynamic feature calibration to effectively address the aforementioned issues. This method can achieve high-precision identification of individual radar emitters in complex environments, with an accuracy rate exceeding 97.5%, and can maintain the original recognition performance even with reduced sample size. This technology can be used for radar emitter identification tasks in complex environments. It directly acts on radar pulse signals, accurately identifying individual emitters while preserving key signal features, and reducing or eliminating the problems of low radar recognition accuracy and poor robustness caused by the limitations of traditional methods.

[0024] In step S100, based on the characteristics of one-dimensional radar pulse signals, this method introduces innovative ideas and the thinking of natural language processing semantic analysis. The envelope of each radar signal pulse corresponds to an independent sentence unit in the text, thereby revealing the commonality between the two in structural organization and information transmission. The various amplitude points on the envelope are regarded as the basic word elements that constitute the sentence. This establishes a new signal analysis perspective. In this way, the preprocessed data can be migrated to the recurrent neural network (RNN) and Transformer architecture for radar signal processing in this method, providing new ideas for tasks such as signal feature extraction and pattern recognition, while significantly reducing the cost and time of algorithm development.

[0025] In this embodiment, preprocessing the pulse signal includes: standardizing the pulse signal in sequence, locating the main pulse area, extracting the signal and enhancing the data, converting the pulse signal into a tensor of the channel dimension, and obtaining a preprocessed pulse sequence.

[0026] Specifically, radar pulse signal amplitudes are normalized to bring signals of varying intensities to the same scale. Next, the main pulse region is located, using an algorithm to precisely identify and lock onto the pulse region containing key information to remove redundant signals. Decimation is then performed, reducing the data volume through downsampling to improve signal processing efficiency. Data augmentation is then performed, expanding the dataset through transformations such as translation and scaling to enhance the generalization performance of subsequent processing models. Finally, the processed data is converted into a tensor with channel dimensions, which serves as input to the subsequent processing network, providing the data foundation for high-precision identification of individual radar emitters.

[0027] In this embodiment, a three-stage hybrid architecture radar emitter individual recognition model is designed, such as Figure 2 As shown, it specifically includes the improved deep residual shrinkage network in step S110, the spatiotemporal fusion network in step S120, and the dynamic feature calibration network in step S130.

[0028] To address the difficulty of feature extraction in low signal-to-noise ratio environments in radar emitter individual identification tasks, traditional residual shrinkage networks (DRSNets) face limitations in radar signal processing: fixed-threshold shrinkage mechanisms are difficult to adapt to dynamic noise environments, single-scale convolution kernels cannot capture the multi-scale features of RF fingerprints, and shallow networks are insufficiently sensitive to pulse detail features. Therefore, an improved Deep Residual Shrinkage Network (DRSNet) is proposed. This network first embeds an attention mechanism within the residual block to dynamically generate channel-level shrinkage thresholds; second, constructs parallel convolution pathways based on multi-scale feature fusion to jointly extract multi-resolution features in the time-frequency domain; and third, employs a multi-stage feature calibration strategy to enhance the network's perception of subtle individual features. These three innovative designs effectively improve feature representation capabilities in noisy environments.

[0029] The overall architecture of the DRSNet network consists of the initial convolutional layer, the residual contraction unit (RSBU_CW) and the multi-scale fusion module (MultiScaleFusion). Each module works together to achieve efficient feature learning in a noisy environment. Figure 3 shown.

[0030] Specifically, the number of specific convolution kernels in the initial convolution layer is 1, the kernel size is 7, and the stride is 2.

[0031] Specifically, in the improved deep residual shrinkage network, dynamic soft thresholds are generated through channel attention to suppress noise and extract multi-scale local features, thereby improving the network's feature representation capabilities in noisy environments.

[0032] like Figure 4 As shown in the figure, in the spatiotemporal fusion network, input fusion is performed by weighted summing of the spatial and temporal features of the input data, and the spatial-temporal feature weights are balanced by learnable gating parameters, thereby achieving efficient modeling of complex spatiotemporal patterns in time series data.

[0033] In this embodiment, in the spatiotemporal fusion network, the spatial features and temporal features of local features are extracted using the CNN network and the Transformer network respectively, and then the spatial features and temporal features are enhanced respectively through the spatial attention mechanism and the temporal attention mechanism to obtain spatial enhanced features and temporal enhanced features, such as Figure 5 As shown in the figure, the learned gating parameters are used to dynamically fuse the spatial enhancement features and the temporal enhancement features to obtain the fused features. The dynamic feature calibration mechanism adaptively enhances the response strength of key feature channels and important time nodes while preserving the original feature distribution, thereby improving the recognition performance of the model.

[0034] Specifically, spatial attention uses per-channel weighting to highlight key local regions within spatial features, resulting in spatially enhanced features. Temporal attention, on the other hand, uses a multi-head self-attention mechanism to establish global long-range dependencies to enhance temporal features, resulting in temporally enhanced features. Learnable gating parameters are then used to dynamically combine the advantages of both, enabling efficient modeling of complex spatiotemporal features in time series data, resulting in fused features.

[0035] In this embodiment, in the dynamic feature calibration network, as Figure 6 As shown in , channel calibration captures the channel-level statistical features of the fusion features through global average pooling, and then generates the channel weight matrix through sigmoid and ReLU functions. Figure 7 As shown, time calibration generates a time dimension attention weight matrix by analyzing the local temporal contextual relationship of the fused features through a lightweight deep separable convolutional network.

[0036] Furthermore, the channel weight matrix is multiplied element-by-element by the time dimension attention weight matrix to generate the spatiotemporal joint attention weight, and the spatiotemporal joint attention weight is used to weight the fused features point-by-point to obtain the calibrated features. Finally, the calibrated features are residually fused with the local features to obtain the features to be identified.

[0037] Specifically, the calibrated features are residually fused with local features through a learnable residual scaling coefficient to obtain robust features to be identified.

[0038] In step S140, in the classification head, the features to be identified are compressed in the time dimension through global average pooling, and then mapped to the category prediction probability through the fully connected layer.

[0039] In this paper, comparative experiments are also conducted to verify the performance of the proposed method. Figure 8-11 As shown in the figure, the experimental results show that the recognition performance of this method is significantly superior to that of the traditional method, and it can still maintain the original recognition performance when the samples are reduced; compared with the classic deep learning network, it has higher accuracy, better stability and generalization ability; the ablation experiment verifies the effectiveness of each module.

[0040] Furthermore, the discriminant features learned by different networks are visually analyzed, such as Figure 12-15 As shown in the figure, a registered forward hook is used to collect features from the penultimate layer of the model and record their labels. Finally, the features and labels are merged and the t-SNE algorithm is used to reduce the dimensionality to two dimensions for visualization analysis. The results show that the features of different radar emitters after training are significantly different. They are dispersed from each other, while the features of the same individuals are clustered together, verifying the credibility of the recognition results.

[0041] Among the aforementioned radar emitter identification methods, a Transformer-based radar emitter identification technique was proposed. This innovative integration of deep residual shrinkage, multi-scale feature interaction, spatiotemporal attention, and dynamic feature calibration addresses the challenge of radar signals being susceptible to interference in complex battlefield environments. While preserving key signal characteristics, this method achieves high-precision identification of individual radar emitters, improving recognition accuracy by over 15% compared to traditional methods and maintaining a stable recognition performance of 97.5% even with a reduced sample size.

[0042] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0043] In one embodiment, Figure 16 As shown, a radar emitter individual identification device is provided, comprising: a signal preprocessing module 200, a feature extraction module 210, a spatiotemporal feature fusion module 220, a dynamic feature calibration module 230, and a classification and identification module 240, wherein: The signal preprocessing module 200 is used to obtain the pulse signal emitted by the radar radiation source and preprocess the pulse signal to obtain a preprocessed pulse sequence; A feature extraction module 210 is configured to extract local features from the preprocessed pulse sequence using an improved deep residual shrinkage network to obtain local features; The spatiotemporal feature fusion module 220 is configured to enhance the local features in both spatial and temporal dimensions using a spatiotemporal fusion network, and then fuse the enhanced features to obtain fused features. A dynamic feature calibration module 230 is configured to utilize a dynamic feature calibration network to perform dynamic weight calibration on the channel dimension features and the time dimension features of the fused features through a joint channel calibration mechanism and time calibration, and then fuse the weight-calibrated features with the local features to obtain features to be identified; The classification and recognition module 240 is configured to utilize a classification head to perform individual recognition of radar emitters according to the features to be recognized and obtain recognition results.

[0044] The specific definitions of the radar emitter individual identification device can be found in the definitions of the radar emitter individual identification method above and will not be repeated here. Each module in the aforementioned radar emitter individual identification device can be implemented in whole or in part via software, hardware, or a combination thereof. Each of these modules can be embedded in or independent of a processor in a computer device in hardware form, or stored in a computer device memory in software form, allowing the processor to call and execute the corresponding operations of each module.

[0045] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 17 As shown. The computer device includes a processor, memory, network interface, display screen and input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for identifying an individual radar emitter is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a key, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse.

[0046] Those skilled in the art will understand that Figure 17 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0047] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented: Acquire a pulse signal emitted by a radar radiation source, and preprocess the pulse signal to obtain a preprocessed pulse sequence; Using an improved deep residual shrinkage network, extracting local features from the preprocessed pulse sequence to obtain local features; Using a spatiotemporal fusion network, the local features are enhanced in both spatial and temporal dimensions, and then the enhanced features are fused to obtain fused features; Using a dynamic feature calibration network, the channel dimension features and time dimension features of the fusion feature are dynamically weighted by a joint channel calibration mechanism and time calibration, and then the weighted calibrated features are fused with the local features to obtain the features to be identified; The classification head is used to perform individual identification of radar emitters according to the features to be identified, and an identification result is obtained.

[0048] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: Acquire a pulse signal emitted by a radar radiation source, and preprocess the pulse signal to obtain a preprocessed pulse sequence; Using an improved deep residual shrinkage network, extracting local features from the preprocessed pulse sequence to obtain local features; Using a spatiotemporal fusion network, the local features are enhanced in both spatial and temporal dimensions, and then the enhanced features are fused to obtain fused features; Using a dynamic feature calibration network, the channel dimension features and time dimension features of the fusion feature are dynamically weighted by a joint channel calibration mechanism and time calibration, and then the weighted calibrated features are fused with the local features to obtain the features to be identified; The classification head is used to perform individual identification of radar emitters according to the features to be identified, and an identification result is obtained.

[0049] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0050] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0051] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for identifying individual radar radiation sources, characterized in that: The method comprises: Acquire a pulse signal emitted by a radar radiation source, and preprocess the pulse signal to obtain a preprocessed pulse sequence; Using an improved deep residual shrinkage network, extracting local features from the preprocessed pulse sequence to obtain local features; Using a spatiotemporal fusion network, the local features are enhanced in both spatial and temporal dimensions, and then the enhanced features are fused to obtain fused features; Using a dynamic feature calibration network, the channel dimension features and time dimension features of the fusion feature are dynamically weighted by a joint channel calibration mechanism and time calibration, and then the weighted calibrated features are fused with the local features to obtain the features to be identified; The classification head is used to perform individual identification of radar emitters according to the features to be identified, and an identification result is obtained.

2. The radar radiation source individual identification method according to claim 1, characterized in that: When preprocessing the pulse signal, the one-dimensional radar pulse signal sequence is mapped to text content in natural language processing, wherein each radar pulse signal envelope is analogous to a sentence in the text, and each amplitude point on the envelope serves as the basic word element that constitutes the sentence.

3. The radar radiation source individual identification method according to claim 2, characterized in that: Preprocessing the pulse signal includes: performing standardization processing on the pulse signal in sequence, locating the main pulse area, using signal extraction and data enhancement, converting the pulse signal into a tensor of the channel dimension, and obtaining the preprocessed pulse sequence.

4. The radar radiation source individual identification method according to claim 3, characterized in that: In the improved deep residual shrinkage network: The initial convolution layer uses a specific convolution kernel to extract features from the preprocessed pulse sequence to obtain initial features; A residual shrinkage unit is used to suppress noise from the initial features by dynamically generating a channel-level shrinkage threshold, and a multi-scale fusion unit is used to jointly extract multi-resolution features in the time-frequency domain through parallel convolution paths; The local features are generated according to the time-frequency domain multi-resolution features using a batch normalization layer and an activation function.

5. The radar radiation source individual identification method according to claim 4, characterized in that: In the spatiotemporal fusion network: Utilizing CNN network and Transformer network to extract spatial features and temporal features of the local features respectively; The spatial features and temporal features are enhanced respectively by a spatial attention mechanism and a temporal attention mechanism to obtain spatial enhanced features and temporal enhanced features; The spatial enhancement feature and the temporal enhancement feature are dynamically fused using the learned gating parameters to obtain the fused feature.

6. The radar radiation source individual identification method according to claim 4, characterized in that: In the dynamic feature calibration network: Channel calibration captures the channel-level statistical features of the fused features through global average pooling and generates a channel weight matrix; Time calibration uses a lightweight deep separable convolutional network to analyze the local temporal contextual relationship of the fused features and generate a time dimension attention weight matrix; Multiplying the channel weight matrix by the time dimension attention weight matrix element by element to generate a spatiotemporal joint attention weight, and using the spatiotemporal joint attention weight to perform point-wise weighting on the fusion feature to obtain a calibrated feature; The calibrated features are residually fused with the local features to obtain the features to be identified.

7. The radar radiation source individual identification method according to claim 6, characterized in that: In the classification header: The features to be identified are compressed in the time dimension through global average pooling, and then mapped to category prediction probabilities through a fully connected layer.

8. A radar radiation source individual identification device, characterized in that: The device comprises: A signal preprocessing module is used to obtain the pulse signal emitted by the radar radiation source and preprocess the pulse signal to obtain a preprocessed pulse sequence; A feature extraction module is used to extract local features from the preprocessed pulse sequence using an improved deep residual shrinkage network to obtain local features; A spatiotemporal feature fusion module is used to enhance the local features in both spatial and temporal dimensions using a spatiotemporal fusion network, and then fuse the enhanced features to obtain fused features; A dynamic feature calibration module is used to use a dynamic feature calibration network to perform dynamic weight calibration on the channel dimension features and time dimension features of the fusion feature through a joint channel calibration mechanism and time calibration, and then fuse the weight-calibrated features with the local features to obtain the features to be identified; The classification and recognition module is used to use the classification head to perform individual identification of radar radiation sources according to the features to be identified and obtain an identification result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images