Scattering center adaptive redundancy removal method, device and equipment based on feature saliency

Abstract

The invention relates to a scattering center adaptive redundancy removal method, device and equipment based on feature saliency. The method comprises the steps of extracting multi-layer feature maps of an ISAR target image through a target recognition convolutional neural network, calculating contribution scores of the feature maps of all layers to a prediction classification result of the target recognition convolutional neural network, taking the contribution scores of the feature maps of all layers as weights, and performing point-by-point multiplication on the feature maps of all layers to obtain a class activation thermodynamic diagram; the method comprises the following steps of: accurately matching a class activation thermodynamic diagram with a target original model in an ISAR target image through a calibration method, calculating a feature saliency value of each scattering center based on the matched class activation thermodynamic diagram, dynamically selecting the scattering center with a saliency feature according to the feature saliency value of each scattering center, and finally obtaining the ISAR target image. Therefore, adaptive redundancy removal of the scattering center is realized. By adopting the method, the scattering center in the ISAR image can be subjected to high-efficiency redundancy elimination, and key feature information of the image after redundancy elimination is ensured to be reserved.

Description

Technical Field

[0001] The present application relates to the technical field of inverse synthetic aperture radar electronic countermeasures, and in particular to a method, device and equipment for adaptively removing redundancy of scattering centers based on feature significance. Background Art

[0002] The inverse synthetic aperture radar (ISAR) deception jamming technology based on digital radio frequency memory (DRFM) has become a research hotspot in the field of electronic countermeasures due to its advantages such as strong reconfigurability and high jamming fidelity. This technology uses DRFM to perform high-speed sampling-storage-modulation-forwarding links on the target's ISAR signal to generate a false signal that is highly coherent with the real target echo, thereby forming a highly realistic false target in ISAR imaging or destroying the imaging process. The core of the technology lies in accurately reproducing the time-frequency phase characteristics of the radar signal, combining the ISAR imaging mechanism to jointly modulate the translation / rotation parameters and scattering characteristics of the false target, and finally generating a high-resolution false image with target recognition confusion capability at the receiving end.

[0003] In template generation deception jamming technology, the key to effective jamming is to use a high-resolution false target electromagnetic scattering template (scattering model) based on pre-generated or real-time synthesis. However, the electromagnetic scattering template needs to store a large amount of scattering point data (including spatial position, amplitude phase and other parameters), which leads to a computing bottleneck on the FPGA platform. Secondly, the time consumption of false image generation is positively correlated with the complexity of the model, which makes it difficult to meet the millisecond-level real-time requirements of ISAR deception jamming. Therefore, how to achieve efficient model redundancy while maintaining key scattering characteristics has become a bottleneck in the development of real-time deception jamming systems.

[0004] Traditional scattering model construction methods rely on electromagnetic simulation or measured data inversion, but there is a large amount of redundant scattering data, which may cause false ISAR image distortion and affect the interference effect. Although point cloud de-redundancy technology provides ideas for solving this problem, due to the physical differences between radar scattering and optical imaging, traditional methods are difficult to apply directly. In recent years, deep learning technology has provided a new solution for point cloud de-redundancy. Deep neural networks have demonstrated the ability to deeply characterize scattering features in radar target recognition, but the feature extraction of existing methods presents black box characteristics, and the feature resolution extracted by existing network visualization technology is insufficient, which makes it difficult to meet the accuracy requirements of scattering center de-redundancy. Summary of the invention

[0005] Based on this, it is necessary to provide a scattering center adaptive de-redundancy method, device and equipment based on feature significance that can efficiently realize scattering center de-redundancy in order to solve the above technical problems.

[0006] A scattering center adaptive de-redundancy method based on feature significance, the method comprising: Acquire ISAR target images; Extracting a multi-layer feature map of the ISAR target image through a target recognition convolutional neural network, and calculating a contribution score of each layer of the feature map to a prediction classification result of the target recognition convolutional neural network; The contribution score of the feature map of each layer is used as the weight, and the feature map of each layer is multiplied point by point to obtain a class activation heat map; Accurately matching the class activation heat map with the target original model in the ISAR target image by a calibration method; Based on the matched class activation heat map, the feature significance value of each scattering center is calculated, and according to the feature significance value of each scattering center, the scattering center with significant features is dynamically selected to achieve adaptive redundancy removal of the scattering center.

[0007] In one embodiment, before calculating the contribution score of the feature map of each layer, the feature map of each layer is normalized.

[0008] In one embodiment, when calculating the contribution score of each layer of feature map to the target recognition convolutional neural network prediction classification result: After upsampling the normalized feature maps of each layer, smoothing is performed to obtain the mask corresponding to the activation map of each layer; After covering the ISAR target image with the masks of each layer, the contribution score of the feature map of each layer is obtained by using the target recognition convolutional neural network and a preset reference matrix.

[0009] In one embodiment, after covering the ISAR target image with the masks of each layer, the contribution score of the activation map of each layer is obtained by using the target recognition convolutional neural network and a preset reference matrix, including: Inputting the ISAR target image covered by each layer of mask into the target recognition convolutional neural network to obtain the prediction results of the feature maps corresponding to each layer; Inputting the preset reference matrix into the target recognition convolutional neural network to obtain a reference prediction result; The prediction results are subtracted from the benchmark prediction results to obtain the contribution scores of the feature maps corresponding to each layer.

[0010] In one embodiment, when obtaining the class activation heat map, the feature maps of each layer are multiplied point by point based on the weights through a ReLU activation function.

[0011] In one embodiment, based on the matched class activation heat map, the feature significance value of each scattering center is calculated using the following formula:

[0012] In the above formula, represents the number of single-side reference units, represents the coordinates of the scattering center, and Represent the range and azimuth resolutions, respectively. and Respectively represent the relative offset on the x and y axes.

[0013] In one embodiment, dynamically selecting a scattering center with significant features according to the feature significance value of each scattering center comprises: According to the feature significance value, the scattering centers are arranged in descending order, and a preset number of scattering centers are dynamically selected as scattering centers with significant features.

[0014] The present application also provides a scattering center adaptive de-redundancy device based on feature significance, the device comprising: ISAR image acquisition module, used to acquire ISAR target images; A contribution score calculation module, used to extract a multi-layer feature map of the ISAR target image through a target recognition convolutional neural network, and calculate the contribution score of each layer of the feature map to the prediction classification result of the target recognition convolutional neural network; A class activation heat map obtaining module is used to use the contribution scores of the feature maps of each layer as weights, multiply the feature maps of each layer point by point, and obtain a class activation heat map; A matching module, used for accurately matching the class activation thermal map with the target original model in the ISAR target image by a calibration method; The scattering center adaptive de-redundancy module is used to calculate the feature significance value of each scattering center based on the matched class activation heat map, and dynamically select the scattering center with significant features according to the feature significance value of each scattering center to achieve adaptive de-redundancy of the scattering center.

[0015] A computer device comprises 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: Acquire ISAR target images; Extracting a multi-layer feature map of the ISAR target image through a target recognition convolutional neural network, and calculating a contribution score of each layer of the feature map to a prediction classification result of the target recognition convolutional neural network; The contribution scores of the feature maps of each layer are used as weights, and the feature maps of each layer are multiplied point by point to obtain a class activation heat map; Accurately matching the class activation heat map with the target original model in the ISAR target image by a calibration method; Based on the matched class activation heat map, the feature significance value of each scattering center is calculated, and according to the feature significance value of each scattering center, the scattering center with significant features is dynamically selected to achieve adaptive redundancy removal of the scattering center.

[0016] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps: Acquire ISAR target images; Extracting a multi-layer feature map of the ISAR target image through a target recognition convolutional neural network, and calculating a contribution score of each layer of the feature map to a prediction classification result of the target recognition convolutional neural network; The contribution scores of the feature maps of each layer are used as weights, and the feature maps of each layer are multiplied point by point to obtain a class activation heat map; Accurately matching the class activation heat map with the target original model in the ISAR target image by a calibration method; Based on the matched class activation heat map, the feature significance value of each scattering center is calculated, and according to the feature significance value of each scattering center, the scattering center with significant features is dynamically selected to achieve adaptive redundancy removal of the scattering center.

[0017] The above-mentioned method, device and equipment for adaptive de-redundancy of scattering centers based on feature significance extracts multi-layer feature maps of ISAR target images through target recognition convolutional neural networks, and calculates the contribution scores of each layer of feature maps to the prediction classification results of target recognition convolutional neural networks, uses the contribution scores of each layer of feature maps as weights, multiplies each layer of feature maps point by point to obtain a class activation heat map, accurately matches the class activation heat map with the original target model in the ISAR target image through a calibration method, calculates the feature significance value of each scattering center based on the matched class activation heat map, and dynamically selects scattering centers with significant features according to the feature significance values ​​of each scattering center to achieve adaptive de-redundancy of scattering centers. This method can be used to efficiently remove redundancy of scattering centers in ISAR images. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 1 is a flow chart of a scattering center adaptive redundancy removal method based on feature significance in one embodiment; Figure 2 A schematic diagram of a process for obtaining a class activation heat map in an embodiment; Figure 3 A schematic diagram of a process flow of an adaptive scattering center de-redundancy strategy based on a class activation heat map in one embodiment; Figure 4 This is a simulation experiment using this method to remove redundancy results of 6 types of targets at a 50% redundancy removal rate; Figure 5 This is a schematic diagram of the redundancy removal results of three types of targets under clustering and feature methods using this method at a redundancy removal rate of 50% in a simulation experiment; Figure 6 This is a schematic diagram of the redundancy removal results of another three types of targets under clustering and feature methods when the redundancy removal rate is 50% using this method in a simulation experiment; Figure 7 This is the confusion matrix of clustering and feature methods in a simulation experiment using this method at a 50% redundancy removal rate; Figure 8 is a structural block diagram of a scattering center adaptive de-redundancy device based on feature significance in one embodiment; Fig. 9 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0020] In view of various problems existing in the ISAR image scattering center de-redundancy method in the prior art, in one embodiment, Figure 1 As shown, a scattering center adaptive de-redundancy method based on feature significance is provided, which specifically includes the following steps: Step S100, acquiring an ISAR target image.

[0021] Step S110, extracting multi-layer feature maps of the ISAR target image through the target recognition convolutional neural network, and calculating the contribution score of each layer of the feature map to the prediction classification result of the target recognition convolutional neural network.

[0022] Step S120, using the contribution score of each layer of feature maps as a weight, multiplying the feature maps of each layer point by point to obtain a class activation heat map.

[0023] Step S130, accurately matching the class activation heat map with the target original model in the ISAR target image by a calibration method.

[0024] Step S140, based on the matched class activation heat map, calculate the feature significance value of each scattering center, and dynamically select the scattering center with significant features according to the feature significance value of each scattering center to achieve adaptive redundancy removal of the scattering center.

[0025] In this embodiment, Score Based Layers-CAM (SL-CAM), i.e., class activation heat map, is first extracted for ISAR target images, and then an adaptive scattering center de-redundancy strategy based on SL-CAM is proposed. When extracting the corresponding class activation heat map according to the ISAR target image, the feature map output by the deep network convolution kernel is superimposed and mapped to the input space in combination with the advantages of multi-layer feature maps, so as to obtain high-precision significant regions that play a key role in network decision-making, solving the problem of insufficient feature resolution extracted by existing network visualization technology. According to the difference in the significance values ​​of these regions, the scattering center is dynamically selected to obtain a scattering center model with more significant feature representation, solving the problem of a large amount of redundant data in the scattering model.

[0026] In this embodiment, the above steps S100 to S120 are the part of extracting the class activation heat map using the target recognition convolutional neural network, and the process is as follows: Figure 2 As shown, steps S130 to S140 are the parts for removing redundancy based on the class activation heat map by using the feature significance value. The process is as follows Figure 3 shown.

[0027] In step S100, the target in the ISAR target image to be subjected to scatter center de-redundancy can be set according to a specific application background, such as an airplane, a vehicle, an aircraft, etc.

[0028] In step S110, the target recognition convolutional neural network is an existing neural network that has been trained to predict the category of the target in the input ISAR target image. The prediction process can be expressed as: ,in, represents the landmark recognition convolutional neural network, and the input vector, i.e., the ISAR target image, is represented as , The prediction vector representing the output of the model is the predicted target category.

[0029] In this embodiment, the target recognition convolutional neural network includes multiple convolutional layers with different convolutional kernel sizes, and the multi-layer network features of the ISAR target image can be extracted through each convolutional layer. ,in, Shidi Tier Network feature map of channel output.

[0030] In this embodiment, before calculating the contribution score of the feature map of each layer, the feature map of each layer is normalized to The normalized formula is:

[0031] In this embodiment, when calculating the contribution score of each layer of feature maps to the target recognition convolutional neural network prediction classification result: the normalized feature maps of each layer are upsampled and then smoothed to obtain masks corresponding to the feature maps of each layer. After the masks of each layer are covered on the ISAR target image, the target recognition convolutional neural network and the preset reference matrix are used to obtain the contribution score of the feature maps of each layer.

[0032] In this embodiment, after covering the ISAR target image with each layer of mask, the contribution score of each layer of feature map is obtained by using the target recognition convolutional neural network and a preset reference matrix, including: inputting the ISAR target image covered by each layer of mask into the target recognition convolutional neural network to obtain the prediction result of the feature map corresponding to each layer, inputting the preset reference matrix into the target recognition convolutional neural network to obtain the reference prediction result, and subtracting each prediction result from the reference prediction result to obtain the contribution score of the feature map corresponding to each layer.

[0033] Specifically, when calculating the contribution score of each layer of feature map, it is necessary to assume that there is a known output benchmark: , Usually a zero matrix, we want to calculate No. item For model output The contribution of replace The output change after the item is evaluated and then the change of network output before and after the change is compared to obtain the classification category Therefore, the feature map For model output Contribution It is expressed as:

[0034] In the above formula, the symbol It stands for Hadamard Product, that is, the corresponding elements of the matrix are multiplied respectively, where the mask of each layer of feature image Should Have the same size, so the masks of the feature images of each layer Through the following formula:

[0035] In the above formula, Represents upsampling to The size is consistent, is a smoothing equation.

[0036] In step S120, when the class activation heat map is obtained, the activation maps of each layer are multiplied point by point based on the weights through the ReLU activation function.

[0037] In this embodiment, the ReLU activation function is used to remove neurons that are useless for responding to the class of interest. Consider the model The convolutional layer in , and given a category of interest , class activation heatmap of SL-CAM Can be defined as:

[0038] in,

[0039] In the above formula, Representative feature map Contribution score.

[0040] Since the distance-azimuth dimension of the ISAR image is different from the actual physical coordinates, it needs to be converted into the same coordinate system to specifically correspond to each scattering center. Therefore, in step S130, the class activation heat map and the target original model in the ISAR target image are accurately matched by a calibration method.

[0041] Specifically, the calibration formula of ISAR image is:

[0042] In the above formula, and Represent the range and azimuth resolutions respectively.

[0043] Furthermore, the class activation heat map is converted from the distance-azimuth coordinate system to the Cartesian coordinate system using the following formula:

[0044] In the above formula, , , and They represent the row, column, and row and column maximum values ​​of the distance-azimuth dimension coordinate system respectively.

[0045] In step S140, the coordinates of a scattering center of the target model in the ISAR target image are , the coordinate position is defined as the detection unit. In view of the fact that the class activation heat map may have a certain degree of diffusion effect on feature representation, an additional reference unit is set to more comprehensively evaluate the feature significance. is the number of reference units on one side, then the total number of The characteristic significance value of the scattering center is calculated using the weighted method using the following formula:

[0046] In the above formula, represents the number of single-side reference units, represents the coordinates of the scattering center, and Represent the range and azimuth resolutions, respectively. and Respectively represent the relative offset on the x and y axes.

[0047] In this embodiment, dynamically selecting scattering centers with significant features according to the feature significance values ​​of each scattering center includes: arranging the scattering centers in descending order according to the feature significance values, and dynamically selecting a preset number of scattering centers as scattering centers with significant features.

[0048] Specifically, the scattering centers are ordered according to the significance value. After sorting, the scattering centers with higher significance values ​​are regarded as more critical and significant, while those with lower significance values ​​are regarded as relatively minor or insignificant. Subsequently, according to the difference in these significance values, the scattering centers are dynamically selected to obtain a scattering center model with more significant feature representation.

[0049] In order to verify the effectiveness of the redundancy removal process of this method, two simulation experiments are carried out to illustrate it.

[0050] The experimental data set is based on the corresponding ISAR echo data generated by the aircraft scattering center model and turntable model simulation of 6 different targets. After receiving these echo signals at the simulated receiving end, they are compressed in the range and azimuth directions to obtain ISAR images. The experimental recognition network uses a transfer learning network based on ResNet18. In the experiment, only the last fully connected layer of the network is modified, and the number of classifications is adjusted to match the number of categories in the experiment. The network is trained using the complete model before de-redundancy, and the recognition accuracy is 100%. The de-redundancy models generated in the simulation experiment are all used for imaging recognition tests based on this network.

[0051] The simulation experiment 1 adopted a 50% de-redundancy rate as the experimental standard, and performed de-redundancy on the aircraft scattering center models of 6 different types of targets to verify the feasibility of the present invention. Figure 4The intermediate process and results of de-redundancy of six types of aircraft scatter center models are shown in the figure. The de-redundancy method proposed in the present invention effectively reduces redundant data.

[0052] Simulation experiment 2 conducts an in-depth comparative analysis of clustering-based methods (including Gaussian mixture model Gmm and Kmeans algorithm) and feature-based methods (SL-CAM). Figure 6 and Figure 7 The redundancy removal results of the six types of models in the Gmm model, Kmeans model and SL-CAM algorithm are shown respectively. Figure 6 and Figure 7 As shown in Figure 2, there are significant differences in the performance levels of different methods when processing ISAR image processing tasks after redundancy removal. Clustering methods mainly focus on extracting the contours of the target as salient features, while feature methods can adaptively extract salient features of the target based on the focus of the network.

[0053] In order to more accurately evaluate the performance of these methods, Structural Similarity Index (SSIM) was used as an evaluation index. SSIM can measure the structural similarity between two images, thus providing us with a means to quantitatively compare the performance of algorithms. SSIM is a number between 0 and 1. The larger the SSIM, the smaller the difference between the output image and the undistorted image, that is, the better the image quality. Table 1 shows the SSIM values ​​of the Gmm model, Kmeans model, and SL-CAM algorithm at a de-redundancy rate of 50%. As shown in Table 1, for most targets, SL-CAM has a higher SSIM value, indicating that the algorithm proposed in the present invention can better extract the key feature structures in the original image.

[0054] Table 1 SSIM values ​​of Gmm model, Kmeans model and SL-CAM algorithm at 50% deduplication rate

[0055] At the same time, the network recognition rate is used as an auxiliary verification method, and the confusion matrix is ​​used to intuitively demonstrate the performance of different methods in the target recognition task. Figure 5 The confusion matrices generated by the Gmm model, Kmeans model, and SL-CAM algorithm at a 50% redundancy removal rate are used for target recognition. Figure 5 As shown in the figure, the two clustering methods have similar recognition rates, with Gmm performing slightly better than kmeans. However, SL-CAM has a recognition rate of 76.25%, significantly exceeding the recognition level of Gmm by about 10%, fully demonstrating the excellent performance of the present invention.

[0056] In the above-mentioned scattering center adaptive deduplication method based on feature significance, the multi-layer activation map of the ISAR target image is extracted by the target recognition convolutional neural network, and the contribution score of each layer of activation map to the prediction classification result of the target recognition convolutional neural network is calculated. The contribution score of each layer of activation map is used as the weight, and the activation map of each layer is multiplied point by point to obtain the class activation heat map. The class activation heat map and the original target model in the ISAR target image are accurately matched by the calibration method. Based on the matched class activation heat map, the feature significance value of each scattering center is calculated. According to the feature significance value of each scattering center, the scattering center with significant features is dynamically selected to realize the adaptive deduplication of the scattering center. The method can effectively reduce redundant data while ensuring that the deduplication image retains key feature information.

[0057] 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. Moreover, 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.

[0058] In one embodiment, Figure 8 As shown, a scattering center adaptive de-redundancy device based on feature significance is provided, comprising: an ISAR image acquisition module 200, a contribution score calculation module 210, a class activation heat map acquisition module 220, a matching module 230 and a scattering center adaptive de-redundancy module 240, wherein: An ISAR image acquisition module 200 is used to acquire an ISAR target image; A contribution score calculation module 210 is used to extract a multi-layer feature map of the ISAR target image through a target recognition convolutional neural network, and calculate the contribution score of each layer of the feature map to the prediction classification result of the target recognition convolutional neural network; A class activation heat map obtaining module 220 is used to use the contribution scores of the feature maps of each layer as weights, and multiply the feature maps of each layer point by point to obtain a class activation heat map; A matching module 230, configured to accurately match the class activation heat map with the target original model in the ISAR target image by a calibration method; The scattering center adaptive deduplication module 240 is used to calculate the feature significance value of each scattering center based on the matched class activation heat map, and dynamically select the scattering center with significant features according to the feature significance value of each scattering center to achieve adaptive deduplication of the scattering center.

[0059] For the specific definition of the scattering center adaptive de-redundancy device based on feature significance, please refer to the definition of the scattering center adaptive de-redundancy method based on feature significance in the above text, which will not be repeated here. Each module in the above-mentioned scattering center adaptive de-redundancy device based on feature significance can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0060] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig. 9 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, 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 the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a scattering center adaptive de-redundancy method based on feature significance 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 covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0061] Those skilled in the art will understand that Fig. 9 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 those shown in the figure, or combine certain components, or have a different arrangement of components.

[0062] 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 ISAR target images; Extracting a multi-layer feature map of the ISAR target image through a target recognition convolutional neural network, and calculating a contribution score of each layer of the feature map to a prediction classification result of the target recognition convolutional neural network; The contribution scores of the feature maps of each layer are used as weights, and the feature maps of each layer are multiplied point by point to obtain a class activation heat map; Accurately matching the class activation heat map with the target original model in the ISAR target image by a calibration method; Based on the matched class activation heat map, the feature significance value of each scattering center is calculated, and according to the feature significance value of each scattering center, the scattering center with significant features is dynamically selected to achieve adaptive redundancy removal of the scattering center.

[0063] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented: Acquire ISAR target images; Extracting a multi-layer feature map of the ISAR target image through a target recognition convolutional neural network, and calculating a contribution score of each layer of the feature map to a prediction classification result of the target recognition convolutional neural network; The contribution scores of the feature maps of each layer are used as weights, and the feature maps of each layer are multiplied point by point to obtain a class activation heat map; Accurately matching the class activation heat map with the target original model in the ISAR target image by a calibration method; Based on the matched class activation heat map, the feature significance value of each scattering center is calculated, and according to the feature significance value of each scattering center, the scattering center with significant features is dynamically selected to achieve adaptive redundancy removal of the scattering center.

[0064] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and 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 (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0065] The technical features of the above embodiments may 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.

[0066] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A scattering center adaptive de-redundancy method based on feature significance, characterized in that: The method comprises: Acquire ISAR target images; Extracting a multi-layer feature map of the ISAR target image through a target recognition convolutional neural network, and calculating a contribution score of each layer of the feature map to a prediction classification result of the target recognition convolutional neural network; The contribution scores of the feature maps of each layer are used as weights, and the feature maps of each layer are multiplied point by point to obtain a class activation heat map; Accurately matching the class activation heat map with the target original model in the ISAR target image by a calibration method; Based on the matched class activation heat map, the feature significance value of each scattering center is calculated, and according to the feature significance value of each scattering center, the scattering center with significant features is dynamically selected to achieve adaptive redundancy removal of the scattering center.

2. The scattering center adaptive de-redundancy method based on feature significance according to claim 1, characterized in that: Before calculating the contribution score of the feature map of each layer, the feature map of each layer is also normalized.

3. The scattering center adaptive de-redundancy method based on feature significance according to claim 2, characterized in that: When calculating the contribution score of each layer of feature map to the target recognition convolutional neural network prediction classification result: After upsampling the normalized feature maps of each layer, smoothing is performed to obtain masks corresponding to the feature maps of each layer; After covering the ISAR target image with the masks of each layer, the contribution score of the feature map of each layer is obtained by using the target recognition convolutional neural network and a preset reference matrix.

4. The scattering center adaptive de-redundancy method based on feature significance according to claim 3, characterized in that: After covering the ISAR target image with the masks of each layer, the contribution scores of the feature maps of each layer are obtained by using the target recognition convolutional neural network and a preset reference matrix. Inputting the ISAR target image covered by each layer of mask into the target recognition convolutional neural network to obtain the prediction results of the feature maps corresponding to each layer; Inputting the preset reference matrix into the target recognition convolutional neural network to obtain a reference prediction result; The prediction results are subtracted from the benchmark prediction results to obtain the contribution scores of the feature maps corresponding to each layer.

5. The scattering center adaptive de-redundancy method based on feature significance according to claim 4, characterized in that: When obtaining the class activation heat map, the feature maps of each layer are multiplied point by point based on the weights through the ReLU activation function.

6. The scattering center adaptive de-redundancy method based on feature significance according to claim 4, characterized in that: Based on the matched class activation heat map, the feature significance value of each scattering center is calculated using the following formula: In the above formula, represents the number of single-side reference units, represents the coordinates of the scattering center, and Represent the range and azimuth resolutions, respectively. and Respectively represent the relative offset on the x and y axes.

7. The scattering center adaptive de-redundancy method based on feature significance according to claim 4, characterized in that: The dynamically selecting a scattering center with significant features according to the feature significance value of each scattering center comprises: According to the feature significance value, the scattering centers are arranged in descending order, and a preset number of scattering centers are dynamically selected as scattering centers with significant features.

8. A scattering center adaptive de-redundancy device based on feature significance, characterized in that: The device comprises: ISAR image acquisition module, used to acquire ISAR target images; A contribution score calculation module, used to extract a multi-layer activation map of the ISAR target image through a target recognition convolutional neural network, and calculate the contribution score of each layer of the activation map to the prediction classification result of the target recognition convolutional neural network; A class activation heat map obtaining module is used to use the contribution score of the activation map of each layer as a weight, multiply the activation map of each layer point by point, and obtain a class activation heat map; A matching module, used for accurately matching the class activation thermal map with the target original model in the ISAR target image by a calibration method; The scattering center adaptive de-redundancy module is used to calculate the feature significance value of each scattering center based on the matched class activation heat map, and dynamically select the scattering center with significant features according to the feature significance value of each scattering center to achieve adaptive de-redundancy of the scattering center.

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.

Images