SAR Target Identification Method and Device Based on Structured Electromagnetic Scattering Features

Abstract

This application relates to a SAR target recognition method and apparatus based on structured electromagnetic scattering features. The method includes: extracting strong scattering points of targets from each SAR sample image, constructing graph structure data based on the extracted strong scattering points, and then using the SAR sample images and the corresponding graph structure data to form sample pairs for training a target recognition model. The target recognition model includes a first feature extraction network that extracts structured electromagnetic scattering features from the graph structure data, a second feature extraction network that extracts depth features from the SAR sample images, and a feature fusion network that performs target recognition based on the structured electromagnetic scattering features and depth features. This allows the trained target recognition model obtained using this method to effectively utilize the electromagnetic scattering characteristics of targets in SAR images, thereby improving the target recognition rate.

Description

Technical Field

[0001] This application relates to the field of image target recognition technology, and in particular to a SAR target recognition method and apparatus based on structured electromagnetic scattering features. Background Technology

[0002] Synthetic Aperture Radar (SAR) is widely used in many fields, such as military, geography, and natural disaster detection, due to its all-weather, all-day imaging capabilities. Target recognition is a fundamental problem in SAR image interpretation and has always been a research hotspot. Given the important role of aircraft in civilian and military missions, such as flight management and military reconnaissance, aircraft identification has become a crucial issue in SAR target identification. Especially with the rapid development of SAR imaging technology, high-resolution SAR images are easier to acquire than ever before, providing more research opportunities for challenging fine-grained SAR aircraft identification tasks.

[0003] SAR aircraft target images typically exhibit several unique characteristics, including discrete target appearance and angle sensitivity. First, SAR aircraft targets display a highly discrete appearance. Compared to vehicle and ship targets in SAR images, aircraft images show less and more discrete scattering information, making them difficult for the human eye to distinguish. This is because aircraft targets have relatively simple structures and smooth surfaces, resulting in less distinct geometric features (such as contours) and texture characteristics. Electromagnetic scattering is one of the most important characteristics of aircraft. Second, because SAR is an active imaging method, the image of a target varies greatly at different imaging angles, which is the angle-sensitive nature of SAR images.

[0004] Early SAR target recognition methods primarily relied on manual feature extraction, such as geometric shape features and transform domain features. Manual feature extraction methods require extensive expert experience for parameter design and have weak generalization capabilities across various situations, making them unsuitable for current high-resolution SAR target recognition tasks. In recent years, with the continuous development of deep learning methods, attention has gradually shifted to this type of method that can automatically extract features. Currently, most SAR aircraft recognition methods based on convolutional neural networks directly borrow from deep learning methods in optical images, failing to effectively utilize the electromagnetic scattering characteristics inherent in SAR's unique imaging mechanism, such as geometric diffraction models, attribute scattering centers, and strong scattering point features. Furthermore, SAR aircraft target images exhibit unique discrete appearance characteristics; using convolutional neural network models makes it difficult to effectively capture the correlations between different components of the aircraft target, failing to fully extract the effective features of the SAR aircraft target, thus limiting the SAR aircraft target recognition capability. Summary of the Invention

[0005] Therefore, it is necessary to provide a SAR target recognition method and device based on structured electromagnetic scattering features that can effectively improve the target recognition rate, addressing the aforementioned technical problems.

[0006] A SAR target identification method based on structured electromagnetic scattering features, the method comprising:

[0007] Acquire a SAR image sample set, which includes multiple SAR sample images of the same target type;

[0008] Extract the strong scattering points of the target in each of the SAR sample images, and construct graph structure data based on the extracted strong scattering points;

[0009] Each SAR sample image in the SAR image sample set is paired with its corresponding graph structure data to form multiple sample pairs, and a training sample set is constructed based on these multiple sample pairs.

[0010] The target recognition model is trained based on the sample pairs in the training sample set to obtain a trained target recognition model. The target recognition model includes a first feature extraction network that extracts structured electromagnetic scattering features based on graph structure data in the sample pairs, a second feature extraction network that extracts depth features based on SAR sample images in the same sample pair, and a feature fusion network that performs target recognition based on the structured electromagnetic scattering features and depth features.

[0011] Acquire a SAR target image to be identified, input the SAR target image into a pre-trained target recognition model, and output the target recognition result to achieve the identification of the target in the SAR target image.

[0012] In one embodiment, extracting the strong scattering points of the target in each of the SAR sample images includes:

[0013] Based on the Gaussian mixture distribution model, a solution formula is constructed to characterize the distribution of strong scattering points of targets in the SAR sample images.

[0014] After minimizing the formula to be solved, the strong scattering point features of the target in the SAR sample image are obtained by iterative calculation using the expectation-maximization algorithm.

[0015] In one embodiment, the target in the SAR sample image includes multiple strong scattering points, and the features of each strong scattering point include the local spatial coordinates and amplitude attributes of the strong scattering point.

[0016] In one embodiment, the step of constructing the graph structure data based on the extracted strong scattering points includes:

[0017] The multiple strong scattering point features of the target in the SAR sample image are used as nodes in the graph structure;

[0018] The local spatial relationships between strong scattering points are taken as edges between corresponding nodes, and the edges are described by an adjacency matrix.

[0019] In one embodiment, the first feature extraction network is a graph convolutional neural network;

[0020] The graph convolutional neural network includes a series of stacked GCN layers, a concatenated global average pooling layer, a global max pooling layer, and a fully connected layer.

[0021] In one embodiment, when the graph convolutional neural network extracts structured electromagnetic scattering features from graph-structured data, it includes:

[0022] The nodes and edges of the graph structure data are used as input to the graph convolutional neural network, and the high-level semantic features and structural features of the electromagnetic scattering features of the target are extracted through the stacked multi-layer GCN layers.

[0023] The advanced semantic features and structural features are obtained as global features through the concatenation of the global average pooling layer and the global max pooling layer;

[0024] The global features are used to obtain the structured electromagnetic scattering features through the fully connected layer.

[0025] In one embodiment, the second feature extraction network is an improved convolutional neural network.

[0026] In one embodiment, training the target recognition model includes:

[0027] After inputting the sample pairs into the target recognition model, the target recognition estimation result is obtained;

[0028] Calculate the loss function based on the target identification estimation results and sample labels;

[0029] The parameters in the target recognition model are adjusted based on the calculation results until the loss function converges, thus obtaining the trained target recognition model.

[0030] In one embodiment, the SAR target recognition method is applied to target recognition of SAR images in which the target is an aircraft.

[0031] A SAR target identification device based on structured electromagnetic scattering characteristics, the device comprising:

[0032] An image sample set acquisition module is used to acquire a SAR image sample set, which includes multiple SAR sample images of the same target type.

[0033] The graph structure data construction module is used to extract the strong scattering points of the target in each of the SAR sample images and construct graph structure data based on the extracted strong scattering points.

[0034] The training sample set construction module is used to form multiple sample pairs based on each SAR sample image in the SAR image sample set and its corresponding graph structure data, and to construct a training sample set based on the multiple sample pairs.

[0035] The target recognition model training module is used to train the target recognition model based on the sample pairs in the training sample set to obtain a trained target recognition model. The target recognition model includes a first feature extraction network that extracts structured electromagnetic scattering features based on graph structure data in the sample pairs, a second feature extraction network that extracts depth features based on SAR sample images in the same sample pair, and a feature fusion network that performs target recognition based on the structured electromagnetic scattering features and depth features.

[0036] The SAR image target recognition module is used to acquire SAR target images to be identified, input the SAR target images into a pre-trained target recognition model, and output the target recognition results to achieve the recognition of targets in the SAR target images.

[0037] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program performing the following steps:

[0038] Acquire a SAR image sample set, which includes multiple SAR sample images of the same target type;

[0039] Extract the strong scattering points of the target in each of the SAR sample images, and construct graph structure data based on the extracted strong scattering points;

[0040] Each SAR sample image in the SAR image sample set is paired with its corresponding graph structure data to form multiple sample pairs, and a training sample set is constructed based on these multiple sample pairs.

[0041] The target recognition model is trained based on the sample pairs in the training sample set to obtain a trained target recognition model. The target recognition model includes a first feature extraction network that extracts structured electromagnetic scattering features based on graph structure data in the sample pairs, a second feature extraction network that extracts depth features based on SAR sample images in the same sample pair, and a feature fusion network that performs target recognition based on the structured electromagnetic scattering features and depth features.

[0042] Acquire a SAR target image to be identified, input the SAR target image into a pre-trained target recognition model, and output the target recognition result to achieve the identification of the target in the SAR target image.

[0043] A computer-readable storage medium having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0044] Acquire a SAR image sample set, which includes multiple SAR sample images of the same target type;

[0045] Extract the strong scattering points of the target in each of the SAR sample images, and construct graph structure data based on the extracted strong scattering points;

[0046] Each SAR sample image in the SAR image sample set is paired with its corresponding graph structure data to form multiple sample pairs, and a training sample set is constructed based on these multiple sample pairs.

[0047] The target recognition model is trained based on the sample pairs in the training sample set to obtain a trained target recognition model. The target recognition model includes a first feature extraction network that extracts structured electromagnetic scattering features based on graph structure data in the sample pairs, a second feature extraction network that extracts depth features based on SAR sample images in the same sample pair, and a feature fusion network that performs target recognition based on the structured electromagnetic scattering features and depth features.

[0048] Acquire a SAR target image to be identified, input the SAR target image into a pre-trained target recognition model, and output the target recognition result to achieve the identification of the target in the SAR target image.

[0049] The aforementioned SAR target recognition method and apparatus based on structured electromagnetic scattering features extracts strong scattering points of targets from each SAR sample image, constructs graph structure data based on the extracted strong scattering points, and then uses SAR sample images and corresponding graph structure data to form sample pairs for training the target recognition model. The target recognition model includes a first feature extraction network that extracts structured electromagnetic scattering features based on graph structure data, a second feature extraction network that extracts depth features based on SAR sample images, and a feature fusion network that performs target recognition based on structured electromagnetic scattering features and depth features. This allows the trained target recognition model to effectively utilize the electromagnetic scattering characteristics of targets when recognizing targets in SAR images, thereby improving the target recognition rate. Attached Figure Description

[0050] Figure 1This is a flowchart illustrating a SAR target identification method based on structured electromagnetic scattering features in one embodiment.

[0051] Figure 2 This is a schematic diagram illustrating the distribution of different target features extracted using different networks in one embodiment, wherein, Figure 2 (a) is a schematic diagram of the feature distribution extracted using the VGG network. Figure 2 (b) A schematic diagram showing the feature distribution extracted using the improved VGG network (DF-DCN) proposed in this paper. Figure 2 (c) is a schematic diagram of the feature distribution of electromagnetic scattering features extracted using a graph convolutional neural network. Figure 2 (d) shows the feature distribution extracted using this method;

[0052] Figure 3 This is a schematic diagram of the target recognition model in one embodiment;

[0053] Figure 4 This is a structural block diagram of a SAR target identification device based on structured electromagnetic scattering features in one embodiment;

[0054] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0056] like Figure 1 As shown, a SAR target identification method based on structured electromagnetic scattering features is provided, including the following steps:

[0057] Step S100: Obtain a SAR image sample set, which includes multiple SAR sample images of the same target type;

[0058] Step S110: Extract the strong scattering points of the target in each SAR sample image, and construct the graph structure data based on the extracted strong scattering points;

[0059] Step S120: Each SAR sample image in the SAR image sample set is paired with its corresponding graph structure data to form multiple sample pairs, and a training sample set is constructed based on the multiple sample pairs.

[0060] Step S130: Train the target recognition model based on the sample pairs in the training sample set to obtain the trained target recognition model. The target recognition model includes a first feature extraction network that extracts structured electromagnetic scattering features based on graph structure data in the sample pair, a second feature extraction network that extracts depth features based on SAR sample images in the same sample pair, and a feature fusion network that performs target recognition based on structured electromagnetic scattering features and depth features.

[0061] In step S140, the SAR target image to be identified is acquired, the SAR target image is input into the trained target recognition model, and the target recognition result is output to realize the recognition of the target in the SAR target image.

[0062] To address the issue of low recognition rates when identifying targets with simple structures and smooth surfaces in SAR images, where geometric features such as contour lights and textures are not readily apparent, this method extracts strong scattering point features from the target in the SAR image and constructs graph-structured data of these features to mine the target's structured electromagnetic scattering characteristics, fully utilizing the SAR target's electromagnetic scattering information. Then, dilated convolution is used to expand the receptive field, increasing the correlation between the discrete parts of the aircraft target and extracting the SAR target's depth features. Finally, a fusion algorithm is used to obtain more discriminative SAR target features for target recognition, effectively improving the target recognition rate of SAR images.

[0063] In this method, such as Figure 1 As shown, steps S100-S120 are for constructing a sample set for training the target recognition model, step S130 is for training the target recognition model and obtaining the trained target recognition model, and step S140 is for applying the trained target recognition model to perform target recognition on the SAR target image.

[0064] In this embodiment, the target identified in the SAR image refers to a target with simple structural features and a smooth surface, including aircraft and other targets that meet this condition. The method will be described below using an aircraft as an example.

[0065] Specifically, in step S100, the SAR image sample set includes multiple SAR sample images of various types of aircraft at different acquisition angles.

[0066] In step S110, extracting the strong scattering points of the target in each SAR sample image includes: constructing a formula to be solved based on the Gaussian mixture distribution model to characterize the distribution of strong scattering points of the target in the SAR sample image. When solving the formula, after minimizing the formula, the expectation-maximization algorithm is used for iterative calculation to obtain the strong scattering point features of the target in the SAR sample image.

[0067] Specifically, the strong scattering points of different types of aircraft have different distributions, such as... Figure 2 As shown, a single strong scattering point region can be approximated as a circular or perfectly circular region, and can be assumed to follow a two-dimensional Gaussian distribution. Therefore, the distribution of strong scattering points of an aircraft target can be characterized using a Gaussian mixture distribution model (GMM):

[0068]

[0069] In formula (1), Let μ represent a two-dimensional Gaussian distribution. k and Σ k Let denot and represent the mean and the method, respectively. Let K represent the number of two-dimensional Gaussian distributions, and ω represent the mean and the method. k Let x represent the weights of each two-dimensional Gaussian distribution, and define x. i =(x i ,y i ) is a pixel in the SAR sample image X.

[0070] Equation (1) is the formula to be solved. When solving it, for a SAR sample image X with N strong scattering points, the following can be obtained by minimizing:

[0071]

[0072] When solving formula (2), it is difficult to find the basic parameters Θ={θ1,...,θ k The maximum likelihood (ML) estimate of}, where θ k ={ω k ,μ k ,Σ k}

[0073] Therefore, the Expectation-Maximization (EM) algorithm is used to calculate the parameters of the Gaussian Mixture Model (GMM). Since the GMM algorithm is highly sensitive to initial parameters, the watershed algorithm is used in this method to determine the number of components. Each component can be fitted with a two-dimensional Gaussian distribution, and the centroid of each component is used as the initial parameter of the GMM distribution.

[0074] By iterating the EM algorithm, a more accurate Gaussian Mixture Model (GMM) distribution of strong scattering points from an aircraft can be obtained. Correspondingly, the position [x, y] of each peak point in the two-dimensional Gaussian distribution represents the position of a single strong scattering point.

[0075] Finally, the strong scattering point feature [P,x,y] of the aircraft target is obtained, where P represents the amplitude attribute of the current strong scattering point, and x and y represent the local spatial coordinates of the current strong scattering point.

[0076] In this embodiment, constructing graph structure data based on the extracted strong scattering points includes: using the features of multiple strong scattering points of the target in the SAR sample image as nodes of the graph structure, using the local spatial positional relationship between each strong scattering point as the edge between the corresponding nodes, and describing the edge using an adjacency matrix.

[0077] Specifically, a graph structure G = (V, E) consists of two crucial parts: nodes V and edges E. Naturally, the strong scattering point features of an aircraft target can be viewed as nodes V of the graph. According to the strong scattering point model, each strong scattering point includes an amplitude attribute P. i and local spatial coordinates x i y i Therefore, a node can represent dimension V. i =[P i ,x i ,y i ].

[0078] Then, the "distance" between strong scattering points is considered as the edge of the graph. Finally, the graph structure of the electromagnetic scattering characteristics of the aircraft target is completed (referred to as ESGraph).

[0079] In ESGraph, edge features are determined by the local spatial relationships between different strong scattering points. Furthermore, the closer the nodes are to each other, the stronger their intrinsic connections. Typically, an adjacency matrix is ​​used. To describe the edge information of the graph:

[0080]

[0081] In formula (3), τ represents the distance threshold, and a i,j This represents the correlation between the i-th strong scattering point and the j-th strong scattering point.

[0082] In step S120, a training sample set for training the target recognition model is constructed by forming a sample pair with SAR sample images and graph structure data constructed from strong scattering points extracted from those images. A set of sample pairs is input each time the target recognition model is trained.

[0083] In this embodiment, the target recognition model consists of a first feature extraction network, a second feature extraction network, and a feature fusion network, as follows: Figure 3As shown in the diagram. The first feature extraction network is used to extract the structured electromagnetic scattering features of the graph data, the second feature extraction network is used to extract the depth features of the SAR sample image, and finally the feature fusion network fuses the structured electromagnetic scattering features and the depth features to identify the target.

[0084] Specifically, the first feature extraction network is a graph convolutional neural network (GCN network), that is... Figure 3 The dashed box on the left represents the network structure, which uses a GCN network to mine the correlation information between strong scattering points based on the graph structure. The graph convolutional neural network consists of stacked GCN layers, concatenated global average pooling layers, global max pooling layers, and fully connected layers connected in sequence.

[0085] Furthermore, graph convolutional neural networks, when extracting structured electromagnetic scattering features from graph-structured data, include:

[0086] The nodes and edges of the graph-structured data are used as input to a graph convolutional neural network (GCN). Multiple layers of GCN are stacked to extract high-level semantic and structural features of the target's electromagnetic scattering characteristics. The GCN layers are calculated using the following formula:

[0087]

[0088] In formula (4), the feature H extracted by the l-th layer network (l) It is the input to the network layer l+1, and H (0) The node features V, W (l) σ is the weight of the l-th layer network, and σ(·) is the activation function.

[0089] A multi-layered GCN network is used to deeply mine high-level semantic and structural features of electromagnetic scattering characteristics of aircraft targets. Then, the global features of ESGraph are obtained by concatenating the results of global average pooling and global max pooling operations. Next, a fully connected layer is used to transform the feature dimension to 128 dimensions. Finally, the structured electromagnetic scattering feature h is obtained. ES Furthermore, it is the first feature in the subsequent feature fusion.

[0090] In this embodiment, the second feature extraction network is a convolutional neural network (CNN network).

[0091] Specifically, in existing technologies, using CNN networks for SAR target recognition can achieve high recognition accuracy. However, due to the small size and discrete appearance of aircraft targets, directly applying these methods presents two problems: One problem is that the large number of pooling layers in the network leads to excessive resolution loss, resulting in insufficient information in subsequent feature representations. The other problem is that the discrete appearance of aircraft targets makes it difficult for small convolutional kernels to extract the correlations between the discrete parts of the aircraft.

[0092] Therefore, in order to obtain more discriminative deep features, this embodiment proposes a DF-DCN network based on VGGNet. The specific structure of the DF-DCN network is as follows: Figure 3 As shown, that is Figure 3 The dashed box on the right represents the network structure. Finally, a fully connected layer extracts a 128-dimensional feature vector h from the DF-DCN network. DF This is the second part of subsequent feature fusion.

[0093] Finally, the structured electromagnetic scattering features h are fused using a Feature Fusion Network (FFM). ES and depth features h DF The task of identifying aircraft targets is completed. A more discriminative feature h is obtained through linear weighted fusion. The mathematical form of the final output fused feature is:

[0094] h = w ES h ES +w DF h DF (5)

[0095] In formula (5), w ES and w DF The structured electromagnetic scattering characteristics h are respectively ES and depth features h DF The weight. Compared to using h alone. ES and h DF In general, features obtained through FFM are more discriminative.

[0096] In this implementation, training the target recognition model includes: inputting sample pairs into the target recognition model to obtain target recognition estimation results; calculating the loss function based on the target recognition estimation results and sample labels; and adjusting the parameters in the target recognition model based on the calculation results until the loss function converges, thus obtaining the trained target recognition model.

[0097] In this paper, experiments were also conducted to demonstrate the effectiveness of the proposed method. Table 1 shows the target recognition results based on this method.

[0098] Table 1: Target Recognition Results

[0099]

[0100] The aforementioned SAR target recognition method based on structured electromagnetic scattering features extracts strong scattering points of targets from each SAR sample image, constructs graph structure data based on the extracted strong scattering points, and then uses SAR sample images and corresponding graph structure data to form sample pairs for training the target recognition model. The target recognition model includes a first feature extraction network that extracts structured electromagnetic scattering features from the graph structure data, a second feature extraction network that extracts depth features from the SAR sample images, and a feature fusion network that performs target recognition based on the structured electromagnetic scattering features and depth features. This allows the trained target recognition model to effectively utilize the electromagnetic scattering characteristics of targets in SAR images, improving the target recognition rate and solving the problem of inaccurate feature extraction due to the discrete appearance of SAR aircraft targets in existing technologies.

[0101] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed 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 some of the sub-steps or stages of other steps.

[0102] In one embodiment, such as Figure 4 As shown, a SAR target recognition device based on structured electromagnetic scattering features is provided, comprising: an image sample set acquisition module 200, a graph structure data construction module 210, a training sample set construction module 220, a target recognition model training module 230, and a SAR image target recognition module 240, wherein:

[0103] The image sample set acquisition module 200 is used to acquire a SAR image sample set, which includes multiple SAR sample images of the same target type.

[0104] The graph structure data construction module 210 is used to extract the strong scattering points of the target in each of the SAR sample images and construct graph structure data based on the extracted strong scattering points.

[0105] The training sample set construction module 220 is used to form multiple sample pairs based on each SAR sample image in the SAR image sample set and the corresponding graph structure data, and to construct a training sample set based on the multiple sample pairs.

[0106] The target recognition model training module 230 is used to train the target recognition model based on the sample pairs in the training sample set to obtain a trained target recognition model. The target recognition model includes a first feature extraction network that extracts structured electromagnetic scattering features based on graph structure data in the sample pairs, a second feature extraction network that extracts depth features based on SAR sample images in the same sample pair, and a feature fusion network that performs target recognition based on the structured electromagnetic scattering features and depth features.

[0107] The SAR image target recognition module 240 is used to acquire the SAR target image to be recognized, input the SAR target image into the trained target recognition model, and output the target recognition result to realize the recognition of the target in the SAR target image.

[0108] Specific limitations regarding the SAR target identification device based on structured electromagnetic scattering features can be found in the limitations of the SAR target identification method based on structured electromagnetic scattering features mentioned above, and will not be repeated here. Each module in the aforementioned SAR target identification device based on structured electromagnetic scattering features can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in a computer device, or stored in software in the memory of a computer device, so that the processor can call and execute the corresponding operations of each module.

[0109] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a SAR target recognition method based on structured electromagnetic scattering characteristics. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0110] Those skilled in the art will understand that Figure 5The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0111] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0112] Acquire a SAR image sample set, which includes multiple SAR sample images of the same target type;

[0113] Extract the strong scattering points of the target in each of the SAR sample images, and construct graph structure data based on the extracted strong scattering points;

[0114] Each SAR sample image in the SAR image sample set is paired with its corresponding graph structure data to form multiple sample pairs, and a training sample set is constructed based on these multiple sample pairs.

[0115] The target recognition model is trained based on the sample pairs in the training sample set to obtain a trained target recognition model. The target recognition model includes a first feature extraction network that extracts structured electromagnetic scattering features based on graph structure data in the sample pairs, a second feature extraction network that extracts depth features based on SAR sample images in the same sample pair, and a feature fusion network that performs target recognition based on the structured electromagnetic scattering features and depth features.

[0116] Acquire a SAR target image to be identified, input the SAR target image into a pre-trained target recognition model, and output the target recognition result to achieve the identification of the target in the SAR target image.

[0117] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0118] Acquire a SAR image sample set, which includes multiple SAR sample images of the same target type;

[0119] Extract the strong scattering points of the target in each of the SAR sample images, and construct graph structure data based on the extracted strong scattering points;

[0120] Each SAR sample image in the SAR image sample set is paired with its corresponding graph structure data to form multiple sample pairs, and a training sample set is constructed based on these multiple sample pairs.

[0121] The target recognition model is trained based on the sample pairs in the training sample set to obtain a trained target recognition model. The target recognition model includes a first feature extraction network that extracts structured electromagnetic scattering features based on graph structure data in the sample pairs, a second feature extraction network that extracts depth features based on SAR sample images in the same sample pair, and a feature fusion network that performs target recognition based on the structured electromagnetic scattering features and depth features.

[0122] Acquire a SAR target image to be identified, input the SAR target image into a pre-trained target recognition model, and output the target recognition result to achieve the identification of the target in the SAR target image.

[0123] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, 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), dual 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.

[0124] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.

[0125] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A SAR target identification method based on structured electromagnetic scattering features, characterized in that, The method includes: Acquire a SAR image sample set, which includes multiple SAR sample images of the same target type; Extracting strong scattering points of targets in each of the SAR sample images specifically includes: constructing a formula to be solved based on a Gaussian mixture distribution model to characterize the distribution of strong scattering points of targets in the SAR sample images; minimizing the formula to be solved; and then using the expectation-maximization algorithm to iteratively calculate the strong scattering point features of targets in the SAR sample images. The graph structure data is constructed based on the extracted strong scattering points, specifically including: using the features of multiple strong scattering points of the target in the SAR sample image as nodes of the graph structure, wherein the strong scattering point features include the local spatial coordinates and amplitude attributes of the strong scattering points, using the local spatial positional relationship between each strong scattering point as the edge between the corresponding nodes, and using an adjacency matrix to describe the edge. Each SAR sample image in the SAR image sample set is paired with its corresponding graph structure data to form multiple sample pairs, and a training sample set is constructed based on these multiple sample pairs. The target recognition model is trained based on the sample pairs in the training sample set to obtain a trained target recognition model. The target recognition model includes a first feature extraction network that extracts structured electromagnetic scattering features based on graph structure data in the sample pairs, a second feature extraction network that extracts depth features based on SAR sample images in the same sample pair, and a feature fusion network that performs target recognition based on the structured electromagnetic scattering features and depth features. Acquire a SAR target image to be identified, input the SAR target image into a pre-trained target recognition model, and output the target recognition result to achieve the identification of the target in the SAR target image.

2. The SAR target identification method according to claim 1, characterized in that, The SAR sample image contains multiple strong scattering points of the target.

3. The SAR target identification method according to claim 2, characterized in that, The first feature extraction network is a graph convolutional neural network; The graph convolutional neural network includes a series of stacked GCN layers, a concatenated global average pooling layer, a global max pooling layer, and a fully connected layer.

4. The SAR target identification method according to claim 3, characterized in that, When the graph convolutional neural network extracts structured electromagnetic scattering features from graph-structured data, it includes: The nodes and edges of the graph structure data are used as input to the graph convolutional neural network, and the high-level semantic features and structural features of the electromagnetic scattering features of the target are extracted through the stacked multi-layer GCN layers. The advanced semantic features and structural features are obtained as global features through the concatenation of the global average pooling layer and the global max pooling layer; The global features are used to obtain the structured electromagnetic scattering features through the fully connected layer.

5. The SAR target identification method according to claim 4, characterized in that, The second feature extraction network is an improved convolutional neural network.

6. The SAR target identification method according to claim 5, characterized in that, Training the target recognition model includes: After inputting the sample pairs into the target recognition model, the target recognition estimation result is obtained; Calculate the loss function based on the target identification estimation results and sample labels; The parameters in the target recognition model are adjusted based on the calculation results until the loss function converges, thus obtaining the trained target recognition model.

7. The SAR target identification method according to any one of claims 1-6, characterized in that, The SAR target recognition method is applied to target recognition of SAR images in which the target is an aircraft.

8. A SAR target identification device based on structured electromagnetic scattering characteristics, characterized in that, The device includes: An image sample set acquisition module is used to acquire a SAR image sample set, which includes multiple SAR sample images of the same target type. The strong scattering point extraction module is used to extract the strong scattering points of the target in each of the SAR sample images. Specifically, it includes: constructing a formula to be solved to characterize the distribution of strong scattering points of the target in the SAR sample image based on the Gaussian mixture distribution model; minimizing the formula to be solved; and using the expectation-maximization algorithm to iteratively calculate the strong scattering point features of the target in the SAR sample image. The graph structure data construction module is used to construct graph structure data based on the extracted strong scattering points. Specifically, it includes: taking the features of multiple strong scattering points of the target in the SAR sample image as nodes of the graph structure, wherein the strong scattering point features include the local spatial coordinates and amplitude attributes of the strong scattering points, taking the local spatial positional relationship between each strong scattering point as the edge between the corresponding nodes, and describing the edge using an adjacency matrix. The training sample set construction module is used to form multiple sample pairs based on each SAR sample image in the SAR image sample set and its corresponding graph structure data, and to construct a training sample set based on the multiple sample pairs. The target recognition model training module is used to train the target recognition model based on the sample pairs in the training sample set to obtain a trained target recognition model. The target recognition model includes a first feature extraction network that extracts structured electromagnetic scattering features based on graph structure data in the sample pairs, a second feature extraction network that extracts depth features based on SAR sample images in the same sample pair, and a feature fusion network that performs target recognition based on the structured electromagnetic scattering features and depth features. The SAR image target recognition module is used to acquire SAR target images to be identified, input the SAR target images into a pre-trained target recognition model, and output the target recognition results to achieve the recognition of targets in the SAR target images.

Images