Radar target significant scattering center extraction method, device, equipment and medium

Abstract

The present application relates to a method, apparatus, device, and medium for extracting significant scattering centers of radar targets. The method generates simulated ISAR images of the target at different angles based on a scattering center model, then uses an image feature extraction network to extract features from multiple simulated ISAR images corresponding to the same target in an ISAR image dataset to obtain multiple feature images. Visual class activation mapping is used to convert each feature image into a CAM image, and each CAM image is rotated and angle-weighted to obtain a CAM-weighted image. The attention contribution value of each attention region is calculated on the CAM-weighted image. A preset number of key attention regions are selected based on the attention contribution values, and the simulated scattering centers corresponding to the selected key attention regions are used as significant scattering centers. This method can extract a small number of key simulated scattering centers, and these key simulated scattering centers can be used to simulate a simulated image with target characteristics.

Description

Technical Field

[0001] The present application relates to the field of radar signal processing technology, and in particular to a method, device, equipment and medium for extracting significant scattering centers of radar targets. Background Art

[0002] Radar, one of the most important sensors in remote sensing, uses a modulated waveform to concentrate electromagnetic wave energy within a target area to search for targets. The target's scattering data can be used to determine its characteristics. The echo signal of a broadband radar can generally be considered to be the coherent synthesis of the echo signals from all the target's scattering centers. Therefore, the target's echo signal can be described by constructing a reasonable scattering center model. Research on scattering center models has accompanied the development of radar technology. Scattering centers not only provide information such as the amplitude, location, polarization, and structure of the local scattering source, but also provide a concise and complete description of the target's overall scattering field. The target's scattering field can then be reconstructed based on the information parameters in the scattering center model. Therefore, scattering center models are crucial for studying target electromagnetic scattering characteristics, playing a vital role in target recognition, data compression and fusion, and radar cross section (RCS) extrapolation. Therefore, selecting a reasonable scattering center model is key to determining the relationship between scattering centers and the scattered field.

[0003] Four scattering center models are commonly used in radar research: the point scattering center model, the damped exponential model (DE), the geometrical theory of diffraction (GTD), and the attributed scattering center model (ASCM). The attributed scattering center model (ASCM) is often used to extract salient features from scattering centers. This model, based on the GTD and PO perspectives, utilizes multiple parameter combinations to characterize the frequency and angle dependence of different typical scattering structures, providing more complete information about the electromagnetic and geometric properties of the scattering center. However, the attributed scattering center model has high parameter dimensionality, complex computations, and difficult parameter estimation. Furthermore, most current parameter estimation algorithms are suitable for small viewing angles and are difficult to apply to multiple viewing angles, resulting in weak scattering centers being overwhelmed by strong ones. Alternatively, image domain segmentation methods can be used to extract salient features from scattering centers. The response of a scattering center appears as multiple "bright spots" in an image, representing multiple independent scattering centers. This characteristic ensures the feasibility of image segmentation. However, in practice, the sidelobes of one scattering center may be coupled with the mainlobes of another scattering center, resulting in energy leakage when segmenting and decoupling scattering centers. Summary of the Invention

[0004] Based on this, it is necessary to provide a radar target significant scattering center extraction method, device, equipment and medium that can enable the scattering center model to represent richer discriminative features to address the above technical problems.

[0005] A method for extracting significant scattering centers of radar targets, the method comprising:

[0006] Acquire an ISAR image dataset, wherein the ISAR image dataset includes simulated ISAR images of a target at different angles generated based on a scattering center model;

[0007] Using an image feature extraction network to extract features from a plurality of simulated ISAR images corresponding to the same target in the ISAR image dataset to obtain a plurality of feature images;

[0008] Using a visual class activation map to convert each of the feature images into a CAM image, and rotating each CAM image so that the target in each CAM image presents the same posture, performing angle weighting on the rotated CAM image to obtain a CAM weighted image;

[0009] An attention contribution value of each attention area is calculated on the CAM weighted image, a preset number of key attention areas are selected according to the attention contribution value, and simulated scattering centers corresponding to the selected key attention areas are used as significant scattering centers.

[0010] In one embodiment, when generating simulated ISAR images of a target at different angles based on a scattering center model:

[0011] The target's initial attitude angle is arbitrarily set to 0°. The target's attitude angle is then rotated 360° around the rotation center from the initial state. Simulated ISAR images are generated at the corresponding angles after each rotation by a preset interval. The target's attitude angle is the sum of the rotation angles.

[0012] In one embodiment, when rotating the CAM images so that the objects in the CAM images present the same posture, a rotation matrix is ​​used to rotate the CAM images. The rotation matrix is ​​expressed as:

[0013]

[0014] In the above formula, m represents the posture angle of the target in the CAM image.

[0015] In one embodiment, the angle weighting is performed on the rotated CAM image to obtain the CAM weighted image using the following formula:

[0016]

[0017] In the above formula, M represents the total number of angles, represents the CAM weighted image obtained after angle weighting, A m Represents the CAM diagram when the attitude angle is m.

[0018] In one embodiment, calculating the attention contribution value of each attention area on the CAM image and selecting a preset number of key attention areas according to the attention contribution value includes:

[0019] The attention contribution values ​​of the attention areas are sorted from large to small, and the attention areas with the highest attention contribution values ​​are selected as the key attention areas according to the preset number.

[0020] In one embodiment, when calculating the attention contribution value of each attention area on the CAM weighted image:

[0021] In the CAM weighted image, calculating an attention region bounding box of each simulated scattering center according to a preset size;

[0022] The attention area of ​​each of the simulated scattering centers is calculated according to the attention area bounding box.

[0023] In one embodiment, the following formula is used to calculate the attention contribution value in the attention area:

[0024]

[0025] In the above formula, c i (x,y) represents the attention area c i The value in [x,y] T Indicates that the position range in the attention area is x i1 ≤x≤x i2 ,y i1 ≤y≤y i2 , P i Represents the attention contribution value of the i-th simulated scattering center.

[0026] The present application provides a device for extracting significant scattering centers of radar targets, the device comprising:

[0027] An ISAR image simulation module is used to obtain an ISAR image dataset, wherein the ISAR image dataset includes simulated ISAR images of the target at different angles generated based on a scattering center model;

[0028] An ISAR image feature extraction module is used to extract features from multiple simulated ISAR images corresponding to the same target in the ISAR image dataset using an image feature extraction network to obtain multiple feature images;

[0029] a CAM image angle weighting module, configured to convert each of the feature images into a CAM image using a visual class activation map, rotate each CAM image so that the objects in each CAM image present the same posture, and perform angle weighting on the rotated CAM images to obtain a CAM weighted image;

[0030] The salient scattering center extraction module is used to calculate the attention contribution value of each attention area on the CAM weighted image, select a preset number of key attention areas based on the attention contribution value, and use the simulated scattering centers corresponding to the selected key attention areas as the salient scattering centers.

[0031] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0032] Acquire an ISAR image dataset, wherein the ISAR image dataset includes simulated ISAR images of a target at different angles generated based on a scattering center model;

[0033] Using an image feature extraction network to extract features from a plurality of simulated ISAR images corresponding to the same target in the ISAR image dataset to obtain a plurality of feature images;

[0034] Using a visual class activation map to convert each of the feature images into a CAM image, and rotating each CAM image so that the target in each CAM image presents the same posture, performing angle weighting on the rotated CAM image to obtain a CAM weighted image;

[0035] An attention contribution value of each attention area is calculated on the CAM weighted image, a preset number of key attention areas are selected according to the attention contribution value, and simulated scattering centers corresponding to the selected key attention areas are used as significant scattering centers.

[0036] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0037] Acquire an ISAR image dataset, wherein the ISAR image dataset includes simulated ISAR images of a target at different angles generated based on a scattering center model;

[0038] Using an image feature extraction network to extract features from a plurality of simulated ISAR images corresponding to the same target in the ISAR image dataset to obtain a plurality of feature images;

[0039] Using a visual class activation map to convert each of the feature images into a CAM image, and rotating each CAM image so that the target in each CAM image presents the same posture, performing angle weighting on the rotated CAM image to obtain a CAM weighted image;

[0040] An attention contribution value of each attention area is calculated on the CAM weighted image, a preset number of key attention areas are selected according to the attention contribution value, and simulated scattering centers corresponding to the selected key attention areas are used as significant scattering centers.

[0041] The above-mentioned radar target salient scattering center extraction method, device, equipment, and medium generate simulated ISAR images of the target at different angles based on a scattering center model, use an image feature extraction network to extract features from multiple simulated ISAR images corresponding to the same target in an ISAR image dataset, and obtain multiple feature images. Each feature image is converted into a CAM image using a visual class activation map. Each CAM image is rotated and angle-weighted to obtain a CAM-weighted image. The attention contribution value of each attention region is calculated on the CAM-weighted image. A preset number of key attention regions are selected based on the attention contribution values, and the simulated scattering centers corresponding to the selected key attention regions are used as salient scattering centers. This method can extract a small number of key simulated scattering centers, and use these key simulated scattering centers to simulate a simulated image with target characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 1 is a flow chart of a method for extracting significant scattering centers of radar targets according to an embodiment;

[0043] Figure 2 A diagram of a dataset of 6 types of aircraft in an experiment;

[0044] Figure 3 A schematic diagram of the VGG16 migration network structure in an experiment;

[0045] Figure 4 CAM diagram of the same target at different posture angles in an experiment;

[0046] Figure 5 A schematic diagram of CAM weighting in an experiment;

[0047] Figure 6 Extraction results of significant scattering centers in an experiment;

[0048] Figure 7 The CAM image when only significant scattering centers are retained in an experiment;

[0049] Figure 8Schematic diagram of experimental results when only significant scattering centers are retained in an experiment;

[0050] Figure 9 A schematic diagram of the steps of a method for extracting target significant scattering centers in another embodiment;

[0051] Figure 10 1. It is a structural block diagram of a radar target significant scattering center extraction device in one embodiment;

[0052] Figure 11 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0053] 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.

[0054] Under current technology, the application of neural networks to synthetic aperture radar (SAR) target recognition often faces a significant challenge: the limited availability of measured SAR images. Consequently, more data must be generated through simulations based on scattering center models. However, the images generated using these simulation methods often fail to fully meet the requirements for neural network training and present numerous challenges. First, simulated images generated using scattering center models are often overly idealized and exhibit significant differences from real, measured images. These differences include, but are not limited to, variations in image texture, noise, and lighting conditions, which can lead to poor performance of neural networks in real applications. Second, even when using simulated data for training, the overly detailed simulated images can lead to overfitting of the neural network to these simulated data, reducing its generalization ability in real scenes. This further weakens the neural network's recognition performance in real applications. Therefore, current technical limitations lie in the lack of sufficient, realistic, measured SAR image data and appropriate methods for processing simulated data. Addressing these issues will help improve the effectiveness of neural networks in SAR target recognition and provide new breakthroughs for technological development in related fields.

[0055] In response to the above problems, Figure 1 As shown, a method for extracting significant scattering centers of radar targets is provided, comprising the following steps:

[0056] Step S100: Acquire an ISAR image dataset, where the ISAR image dataset includes simulated ISAR images of a target at different angles generated based on a scattering center model.

[0057] Step S110 , using an image feature extraction network to perform feature extraction on multiple simulated ISAR images corresponding to the same target in the ISAR image dataset to obtain multiple feature images.

[0058] In step S120 , each feature image is converted into a CAM image using a visual class activation map, and each CAM image is rotated so that the targets in each CAM image present the same posture. The rotated CAM images are angle-weighted to obtain a CAM weighted image.

[0059] Step S130 , calculating the attention contribution value of each attention area on the CAM weighted image, selecting a preset number of key attention areas according to the attention contribution value, and using the simulated scattering centers corresponding to the selected key attention areas as significant scattering centers.

[0060] In this example, we apply the Class Activation Map (CAM), originally used for visualization in convolutional neural networks, to scattering center feature extraction. We propose a radar target scattering center extraction method based on CAM guidance. This method effectively extracts the most significant discriminative features from the scattering center model, which can be used to streamline the model. This method calculates the attention contribution of scattering centers through CAM to select significant scattering centers.

[0061] In step S100, an ISAR image dataset of a target at different postures and frequencies is obtained by using a scattering center model forward acquisition technique.

[0062] In this embodiment, due to the wide bandwidth of high-resolution radar, the target's scattering characteristics, acting on the optical region based on the relative relationship between target size and radar operating wavelength, can be equated to a frequency- and azimuth-dependent scattering center model. Traditional methods for simulating ISAR images typically pre-select a target's scattering center model and then simulate and generate radar echoes based on a turntable model. This process requires acquiring the target's scattering characteristics in all its attitudes.

[0063] In this embodiment, when generating simulated ISAR images of a target at different angles based on the scattering center model, the target's initial attitude angle is arbitrarily set to 0°. The target's attitude angle is then rotated 360° around the rotation center from the initial state. Simulated ISAR images at corresponding angles are generated after each rotation by a preset interval. The target's attitude angle is the sum of the rotation angles.

[0064] Specifically, the target's attitude angle is rotated 360° around the rotation center, with each rotation interval of 0.5°. The rotation matrix (counterclockwise) is expressed as:

[0065]

[0066] In formula (1), m represents the attitude angle.

[0067] It should be noted that the rotation interval angle is customized according to the application scenario and is not limited to the above 0.5°.

[0068] It is worth noting that the scattering amplitude of each scattering center is usually randomly generated within a small range, and the relationship between the scattering center intensity and the target distance is rarely considered. The known radar equation is:

[0069]

[0070] In formula (2), P r and P t Represent the receiving and transmitting power respectively, G represents the antenna gain, λ represents the operating wavelength, σ represents the scattering center amplitude, R represents the target distance, and it is easy to get the received power P r Inversely proportional to the fourth power of the target distance R.

[0071] Assuming that the distance between the effective rotation center of the target and the radar is R0, then By Taylor expansion, we can approximate:

[0072]

[0073] Therefore, in order to better simulate the scattering characteristics of radar targets, a range-direction mask can be added when simulating radar echoes:

[0074]

[0075] In formula (4), r represents the relative distance between the scattering center and the effective rotation center.

[0076] After obtaining the radar echo, conventional ISAR imaging processing is performed to obtain the ISAR image. The carrier frequency, bandwidth, pulse width and pulse repetition frequency of the simulated radar signal are set to 10 GHz, 600 MHz, 100 μs and 200 Hz respectively. In order to match the size of the actual measured ISAR data, the simulation scene size is set to N r ×N a , where N r =256 is the distance dimension, N a =256 is the azimuthal dimension.

[0077] In one embodiment, when constructing the ISAR image dataset, Figure 2 It is constructed based on the six types of aircraft targets in the .

[0078] In this embodiment, when performing ISAR image simulation, the ISAR simulation image can also be obtained by other means.

[0079] After performing image simulations with different types of aircraft targets, the dataset was classified based on attitude angle. The target's attitude angle was rotated 360° around the center of rotation, with each rotation interval being 0.5°. This yielded 720 ISAR images for each target type. The training set consisted of 360 ISAR images with attitude angles ranging from 0.5° to 359.5° and 1° intervals, while the test set consisted of 360 ISAR images with attitude angles ranging from 0° to 359° and 1° intervals.

[0080] In step S110, an image feature extraction network is used to extract a feature map of the simulated image. Here, the image feature extraction network can select a convolutional neural network, a lightweight network, a self-attention mechanism network, and a network of the feature extraction part in the pyramid network.

[0081] In this embodiment, the image feature extraction network selected is from the VGG16 migration learning network, and the network structure is as follows: Figure 3 shown.

[0082] During the training of the feature extraction network, the mini-batch size is set to 64, and the Adam optimizer with a learning rate of 0.0001 is used for training.

[0083] After extracting features, traditional CNNs typically connect several fully connected layers before performing softmax classification. However, these multiple fully connected layers consume the majority of the model's parameters. CAM uses global average pooling (GAP) instead of fully connected layers to reduce the number of parameters. Specifically, a GAP operation is performed on all feature maps obtained from the last convolutional layer to obtain the image's feature vector, which is then directly connected to the output layer. Using GAP, the weights corresponding to a label in the output layer are mapped to the convolutional feature map output by the previous layer, revealing the importance of each feature map to the output result. This weighted summation of feature maps is CAM.

[0084] In step S120, after obtaining feature images of the same target at multiple angles through the image feature extraction network, each feature image is converted into a CAM image.

[0085] Specifically, several feature maps are extracted from the CNN part of the model and used Indicates that H, W, and K represent the height, width, and number of feature maps, respectively. After GAP, the feature vector of the input image is obtained and connected to the output layer (fully connected layer) of the model. Represents the weight of the nth neuron (classification result) in the output layer connected to the kth feature map. At this time, the CAM map corresponding to the nth class can be expressed as:

[0086]

[0087] In formula (5), F k represents the kth feature map.

[0088] The feature map can be activated by a corresponding visual pattern (such as airplane wings, etc.). The more important the pattern is for the recognition result, the higher the weight corresponding to the feature map. The larger the feature map, the larger the image should be. CAM is the linear weighted sum of these feature maps, followed by upsampling the input image and performing minimum and maximum normalization. The value of each position in the map represents its importance to the recognition result. Therefore, CAM can capture the most critical attention region for a particular category. It was originally applied to CNN visualization and weakly supervised object localization. This method utilizes it to generate scattering center extraction guided by key attention.

[0089] After obtaining the multi-angle CAM, it is necessary to perform angle weighting. However, due to the target's posture changes, the CAM cannot be mapped to the original scattering center model, so a rotation transformation is required before weighting.

[0090] In this embodiment, when rotating each CAM image so that the target in each CAM image presents the same posture, a rotation matrix is ​​used to rotate each CAM image. The rotation matrix is ​​expressed as:

[0091]

[0092] In formula (6), m represents the pose angle of the target in the CAM image.

[0093] After rotating the CAM image:

[0094]

[0095] In formula (7), It's A m The coordinates of the scattering center in yes The coordinates of the scattering center in A m The CAM diagram represents the posture angle m. Represents the rotated CAM image.

[0096] Furthermore, the rotated CAM image is angle-weighted to obtain the CAM weighted image using the following formula:

[0097]

[0098] In formula (8), M represents the total number of angles, A represents the CAM weighted image obtained after angle weighting, and Am Represents the CAM diagram when the posture angle is m.

[0099] like Figure 4 As shown in the figure, the CAM images at different posture angles in an experiment are the CAM weighted images after multi-angle weighting. Figure 5 shown.

[0100] In step S130, the attention contribution value of each attention area is calculated on the CAM weighted image, and a preset number of key attention areas are selected according to the attention contribution value, including: sorting the attention contribution value of each attention area from large to small, and selecting the attention areas with the highest attention contribution value according to the preset number as the key attention areas.

[0101] In this embodiment, in the CAM weighted image, if the original target scattering center model has a total of I scattering centers, the position of the i-th scattering center is The attention region under each scattering center in the angle-weighted CAM image is calculated using these coordinate positions. Specifically, each simulated scattering center in each CAM image has an attention region in the CAM-weighted image. This region is calculated by calculating the attention region bounding box of each simulated scattering center in the CAM-weighted image according to a preset size. The attention region of each simulated scattering center is then calculated based on the attention region bounding box.

[0102] Specifically, if X is a CAM weighted image with height and width h and w respectively, and p is the padding size, then the calculation formula for the coordinates of the required rectangular attention region bounding box is:

[0103]

[0104] In formula (9), x i1 、x i2 、y i1 、y i2 They represent the left boundary, right boundary, lower boundary and upper boundary of the attention region of the i-th scattering center in a CAM weighted image.

[0105] Then, the attention area of ​​each scatter center is calculated based on the bounding box using the following formula:

[0106] c i =X[x i1 :x i2 ,y i1 :y i2 ] (10)

[0107] Furthermore, the following formula is used to calculate the attention contribution value in the attention area:

[0108]

[0109] In formula (11), c i (x,y) represents the attention area c i The value in [x,y] T Indicates that the position range in the attention area is x i1 ≤x≤x i2 ,y i1 ≤y≤y i2 , P i Represents the attention contribution value of the i-th simulated scattering center.

[0110] Next, the attention contribution value of each scattering center is ranked. The higher the ranking, the more significant the feature. The required number of scattering centers are selected as significant scattering centers.

[0111] like Figure 6 The following is the experimental result of significant scattering center extraction. When only significant scattering centers are retained, the test set is generated in the same way. Figure 7 The CAM map is generated when only significant scattering centers are retained during the experiment. Figure 8 The network recognition test results are saved in the form of confusion matrix. The experimental results show that using only significant scattering centers as the target model has no effect on the network recognition results, proving that significant scattering centers can well characterize the characteristics of the original target scattering center model.

[0112] The experimental results show that the proposed method can achieve good scattering center extraction effect while maintaining a high recognition rate of the network. Figure 9 Compared to traditional methods that perform calculations on ISAR images, this method uses an end-to-end network to generate CAMs as the basis for calculation, making it more reliable. Furthermore, this method uses CAM images to guide scattering center extraction, opening up new approaches for scattering center feature extraction.

[0113] In this method for extracting significant scattering centers from radar targets, a scattering center model forward acquisition technique is used to obtain an ISAR image dataset of the target at different poses and frequencies. The dataset is then divided by imaging angle, which serves as the training and test sets for the network. A CAM algorithm is then used to identify key attention regions in the ISAR images. These regions are then effectively weighted at multiple angles. The resulting weighted attention image helps the scattering center model extract more discriminative key local features. Furthermore, an attention contribution value is calculated for each scattering center based on a preset number of scattering centers and the weighted attention image. Finally, the attention contribution values ​​are ranked by numerical value, and the top-ranked scattering centers are extracted as significant scattering centers. When the basic scattering center model is known, this method can effectively select scattering centers with more discriminative features, thereby reducing the total number of scattering centers. In practical applications, the CAM image is used to guide scattering center extraction, opening up new avenues for scattering center feature extraction. In terms of implementation, compared to traditional methods that perform calculations on ISAR images, the proposed method, which generates CAMs through an end-to-end network as a basis for calculation, is more reliable.

[0114] At the same time, by further extracting key scattering center points using this method and simulating the extracted points to obtain a simulated image, the number of scattering center points can be reduced while effectively retaining the key features of the target, thereby enabling real-time simulation under limited hardware conditions, and the simulated image is closer to the measured ISAR image, thereby improving the robustness of the neural network when using the simulated image to train it.

[0115] 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.

[0116] In one embodiment, Figure 10 As shown, a radar target significant scattering center extraction device is provided, comprising: an ISAR image simulation module 200, an ISAR image feature extraction module 210, a CAM image angle weighting module 220 and a significant scattering center extraction module 230, wherein:

[0117] An ISAR image simulation module 200 is configured to obtain an ISAR image dataset, wherein the ISAR image dataset includes simulated ISAR images of a target at different angles generated based on a scattering center model;

[0118] An ISAR image feature extraction module 210 is configured to extract features from multiple simulated ISAR images corresponding to the same target in the ISAR image dataset using an image feature extraction network to obtain multiple feature images;

[0119] A CAM image angle weighting module 220 is configured to convert each of the feature images into a CAM image using a visual class activation map, rotate each CAM image so that the objects in each CAM image present the same posture, and perform angle weighting on the rotated CAM images to obtain a CAM weighted image;

[0120] The significant scattering center extraction module 230 is used to calculate the attention contribution value of each attention area on the CAM weighted image, select a preset number of key attention areas based on the attention contribution value, and use the simulated scattering centers corresponding to the selected key attention areas as significant scattering centers.

[0121] The specific definition of the radar target significant scattering center extraction device can be found in the definition of the radar target significant scattering center extraction method above, and will not be repeated here. The various modules in the above-mentioned radar target significant scattering center extraction device can be implemented in whole or in part through software, hardware, or a combination thereof. The above-mentioned modules can be embedded in or independent of the processor of the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each of the above modules.

[0122] 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 11As 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 extracting significant scattering centers of radar targets 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.

[0123] Those skilled in the art will understand that Figure 11 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.

[0124] 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:

[0125] An ISAR image dataset is acquired, where the ISAR image dataset includes simulated ISAR images of a target at different angles generated based on a scattering center model.

[0126] The image feature extraction network is used to extract features from multiple simulated ISAR images corresponding to the same target in the ISAR image dataset to obtain multiple feature images.

[0127] Each feature image is converted into a CAM image using visual class activation mapping, and each CAM image is rotated so that the targets in each CAM image present the same posture. The rotated CAM images are angle-weighted to obtain CAM weighted images.

[0128] The attention contribution value of each attention area is calculated on the CAM weighted image, and a preset number of key attention areas are selected according to the attention contribution value. The simulated scattering centers corresponding to the selected key attention areas are used as significant scattering centers.

[0129] 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:

[0130] An ISAR image dataset is acquired, where the ISAR image dataset includes simulated ISAR images of a target at different angles generated based on a scattering center model.

[0131] The image feature extraction network is used to extract features from multiple simulated ISAR images corresponding to the same target in the ISAR image dataset to obtain multiple feature images.

[0132] Each feature image is converted into a CAM image using visual class activation mapping, and each CAM image is rotated so that the targets in each CAM image present the same posture. The rotated CAM images are angle-weighted to obtain CAM weighted images.

[0133] The attention contribution value of each attention area is calculated on the CAM weighted image, and a preset number of key attention areas are selected according to the attention contribution value. The simulated scattering centers corresponding to the selected key attention areas are used as significant scattering centers.

[0134] Those skilled in the art will appreciate 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 can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0135] 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.

[0136] 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 extracting significant scattering centers of radar targets, characterized in that: The method comprises: Acquire an ISAR image dataset, wherein the ISAR image dataset includes simulated ISAR images of a target at different angles generated based on a scattering center model; Using an image feature extraction network to extract features from a plurality of simulated ISAR images corresponding to the same target in the ISAR image dataset to obtain a plurality of feature images; The visual class activation mapping is used to convert each feature image into a CAM image, and each CAM image is rotated so that the target in each CAM image presents the same posture. The rotated CAM image is angle-weighted to obtain a CAM weighted image. The angle-weighted CAM image is angle-weighted to obtain a CAM weighted image using the following formula: In the above formula, Indicates the total number of angles, represents the CAM weighted image obtained after angle weighting, Indicates the attitude angle is CAM diagram when Calculate the attention contribution value of each attention area on the CAM weighted image, select a preset number of key attention areas according to the attention contribution value, and use the simulated scattering centers corresponding to the selected key attention areas as significant scattering centers, wherein the attention contribution value in the attention area is calculated using the following formula: In the above formula, Indicates the attention area The value in Indicates the size of the position range in the attention area is , , Indicates the The attention contribution value of the simulated scattering center.

2. The method for extracting significant scattering centers of radar targets according to claim 1, characterized in that: When generating simulated ISAR images of targets at different angles based on the scattering center model: The attitude angle of the target in a certain initial state is arbitrarily set to 0°, and then the attitude angle of the target is rotated 360° around the rotation center from the initial state. After each rotation of the preset interval degree, a simulated ISAR image at the corresponding angle is generated. The attitude angle of the target is the sum of the rotation angles.

3. The method for extracting significant scattering centers of radar targets according to claim 2, characterized in that: When rotating each CAM image so that the target in each CAM image presents the same posture, a rotation matrix is ​​used to rotate each CAM image. The rotation matrix is ​​expressed as: In the above formula, Indicates the posture angle of the target in the CAM image.

4. The method for extracting significant scattering centers of radar targets according to claim 1, wherein: Calculating the attention contribution value of each attention area on the CAM image and selecting a preset number of key attention areas according to the attention contribution value include: The attention contribution values ​​of the attention areas are sorted from large to small, and the attention areas with the highest attention contribution values ​​are selected as the key attention areas according to the preset number.

5. The method for extracting significant scattering centers of radar targets according to claim 4, characterized in that: When calculating the attention contribution value of each attention area on the CAM weighted image: In the CAM weighted image, calculating an attention region bounding box of each simulated scattering center according to a preset size; The attention area of ​​each of the simulated scattering centers is calculated according to the attention area bounding box.

6. A radar target significant scattering center extraction device, characterized in that: The device comprises: An ISAR image simulation module is used to obtain an ISAR image dataset, wherein the ISAR image dataset includes simulated ISAR images of the target at different angles generated based on a scattering center model; An ISAR image feature extraction module is used to extract features from multiple simulated ISAR images corresponding to the same target in the ISAR image dataset using an image feature extraction network to obtain multiple feature images; The CAM image angle weighting module is used to convert each of the feature images into a CAM image using a visual class activation map, and rotate each CAM image so that the targets in each CAM image present the same posture, and perform angle weighting on the rotated CAM image to obtain a CAM weighted image. The angle weighting of the rotated CAM image is performed to obtain a CAM weighted image using the following formula: In the above formula, Indicates the total number of angles, represents the CAM weighted image obtained after angle weighting, Indicates the attitude angle is CAM diagram when The salient scattering center extraction module is used to calculate the attention contribution value of each attention area on the CAM weighted image, select a preset number of key attention areas based on the attention contribution value, and use the simulated scattering centers corresponding to the selected key attention areas as salient scattering centers. The attention contribution value in the attention area is calculated using the following formula: In the above formula, Indicates the attention area The value in Indicates the size of the position range in the attention area is , , Indicates the The attention contribution value of the simulated scattering center.

7. 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 5 are implemented.

8. 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 5 are implemented.

Images