SAR image target recognition methods, devices, computer equipment and storage media

Abstract

This application relates to a SAR image target recognition method, apparatus, computer device, and storage medium. When training the target recognition network, the main scattering centers are first extracted from the SAR training image. Then, the corresponding regions of the scattering centers are enhanced to obtain scattering point enhanced images. These enhanced images, along with the corresponding original SAR training images, are simultaneously input into the target recognition network. A feature extraction unit performs multi-scale feature extraction on both the scattering point enhanced images and the SAR training images to obtain corresponding scattering point enhanced feature maps and SAR training feature maps. Classification is then performed based on the fused feature maps. By minimizing the network loss function, a trained target recognition network is obtained. When recognizing SAR images, simply inputting the image into the trained target recognition network enables target recognition. This method can accurately identify targets in SAR images.

Description

Technical Field

[0001] This application relates to the fields of intelligent radar image processing and computer intelligence technology, and in particular to a SAR image target recognition method, apparatus, computer equipment, and storage medium. Background Technology

[0002] Deep learning has achieved great success in optical target classification and recognition; however, the feasibility and accuracy of this method in SAR images are limited by the imaging characteristics of SAR images themselves. In recent years, with the rise of convolutional networks, more and more algorithms applied to optical images have been proposed, and their performance on optical images is excellent.

[0003] The primary difference between SAR and optical images lies in their imaging principles and image characteristics. Optical images acquire images by recording the reflection or emission of visible light by a target, thus providing rich visual features such as color, texture, and shape. SAR images, on the other hand, form images by using a radar system to emit microwave signals and receive the echoes; their pixel values ​​represent the target's scattering intensity. Because the imaging principle of SAR images differs from that of optical images, optical image processing methods cannot be directly applied to analyze SAR images. Another important difference is the resolution of SAR and optical images. Optical images typically have high spatial resolution, clearly displaying target details and boundaries. SAR images, however, have relatively low resolution and are affected by factors such as radar system parameters and target distance, potentially exhibiting problems like blurring and overlap. This increases the difficulty of locating and detecting targets in SAR images, placing higher demands on target identification. Summary of the Invention

[0004] Therefore, it is necessary to provide a SAR image target recognition method, apparatus, computer equipment, and storage medium that can improve recognition accuracy in response to the above-mentioned technical problems.

[0005] A SAR image target recognition method, the method comprising:

[0006] Obtain a SAR training dataset, which includes multiple SAR training images and ground truth labels for each image.

[0007] The scattering centers in the SAR training image are extracted based on the attribute scattering center model to obtain multiple scattering center points. Then, the SAR training image is enhanced with scattering points based on each of the scattering center points to obtain an enhanced scattering point image.

[0008] The enhanced scattering point image and the corresponding SAR training image are simultaneously input into the target recognition network for training. The target recognition network includes a multi-scale feature extraction unit, a feature fusion unit, and a classification unit. The feature extraction unit performs multi-scale feature extraction on the enhanced scattering point image and the SAR training image to obtain corresponding enhanced scattering point feature maps and SAR training feature maps, respectively. The feature fusion unit fuses the enhanced scattering point feature maps and the SAR training feature maps to obtain a fused feature map. The classification unit obtains a prediction result based on the fused feature map.

[0009] The loss function is calculated based on the prediction results and the corresponding real labels. The parameters of the target recognition network are adjusted based on the result of the loss function until the loss function converges, thus obtaining the trained target recognition network.

[0010] Acquire a SAR image for target recognition, input the SAR image into the trained target recognition network, and obtain the target recognition result of the SAR image.

[0011] In one embodiment, the multi-scale feature extraction unit includes a first channel and a second channel for extracting features from the enhanced scattering point image and the SAR training image, respectively.

[0012] The first channel includes a multi-scale feature extraction layer for extracting a scattering point enhancement feature map from the scattering point enhancement image;

[0013] The second channel includes a multi-layer, multi-scale feature extraction layer, used to extract multiple SAR training feature maps from the SAR training image.

[0014] In one embodiment, each of the multi-scale feature extraction layers includes multiple convolutional layers with different kernel sizes.

[0015] In one embodiment, each of the multi-scale feature extraction layers includes three convolutional layers with kernel sizes of 3, 5, and 7, respectively.

[0016] The multi-scale feature extraction layer is represented as follows:

[0017] z g =w 2-1 conv3(x G )+w 2-2 conv5(x G )+w 2-3 conv7(x G )

[0018] In the above formula, z g The output of the multi-scale feature extraction layer is represented as x.G The image is represented as the enhanced image of the scattering points, where conv3 represents a kernel size of 3, conv5 represents a kernel size of 5, conv7 represents a kernel size of 7, and w 2-1 w 2-2 w 2-3 This indicates the fusion of feature information from different receptive fields within a convolutional layer.

[0019] In one embodiment, the second channel includes three multi-scale feature extraction layers connected in sequence, each of which outputs a SAR training feature map.

[0020] In one embodiment, the feature fusion unit is represented as:

[0021] Z = z g w2+z h1 w 1-1 +z h2 w 1-2 +z h3 w 1-3

[0022] In the above formula, Z represents the fused feature map, z g This represents the enhanced feature map of the scattering points, z h1 z h2 z h3 w2, w3, and w4 represent three SAR training feature maps extracted from the SAR training image by the second channel, respectively. 1-1 w 1-2 w 1-2 Indicates the weight.

[0023] In one embodiment, the classification unit uses a fully connected layer to classify targets in the fused feature map.

[0024] This application also provides a SAR image target recognition device, the device comprising:

[0025] The training data module is used to acquire the SAR training dataset, which includes multiple SAR training images and ground truth labels for each image.

[0026] The scattering point image enhancement module is used to extract the scattering center in the SAR training image based on the attribute scattering center model to obtain multiple scattering center points, and to enhance the SAR training image based on each of the scattering center points to obtain a scattering point enhanced image.

[0027] The network training module is used to simultaneously input the enhanced scattering point image and the corresponding SAR training image into the target recognition network for training. The target recognition network includes a multi-scale feature extraction unit, a feature fusion unit, and a classification unit. The feature extraction unit performs multi-scale feature extraction on the enhanced scattering point image and the SAR training image to obtain corresponding enhanced scattering point feature maps and SAR training feature maps. The feature fusion unit fuses the enhanced scattering point feature maps and the SAR training feature maps to obtain a fused feature map. The classification unit obtains a prediction result based on the fused feature map.

[0028] The trained target recognition network module is used to calculate a loss function based on the prediction results and the corresponding real labels, and to adjust the parameters of the target recognition network according to the result of the loss function until the loss function converges, thus obtaining the trained target recognition network.

[0029] The SAR image target recognition module is used to acquire the SAR image to be identified, input the SAR image into the trained target recognition network, and obtain the target recognition result of the SAR image.

[0030] 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:

[0031] Obtain a SAR training dataset, which includes multiple SAR training images and ground truth labels for each image.

[0032] The scattering centers in the SAR training image are extracted based on the attribute scattering center model to obtain multiple scattering center points. Then, the SAR training image is enhanced with scattering points based on each of the scattering center points to obtain an enhanced scattering point image.

[0033] The enhanced scattering point image and the corresponding SAR training image are simultaneously input into the target recognition network for training. The target recognition network includes a multi-scale feature extraction unit, a feature fusion unit, and a classification unit. The feature extraction unit performs multi-scale feature extraction on the enhanced scattering point image and the SAR training image to obtain corresponding enhanced scattering point feature maps and SAR training feature maps, respectively. The feature fusion unit fuses the enhanced scattering point feature maps and the SAR training feature maps to obtain a fused feature map. The classification unit obtains a prediction result based on the fused feature map.

[0034] The loss function is calculated based on the prediction results and the corresponding real labels. The parameters of the target recognition network are adjusted based on the result of the loss function until the loss function converges, thus obtaining the trained target recognition network.

[0035] Acquire a SAR image for target recognition, input the SAR image into the trained target recognition network, and obtain the target recognition result of the SAR image.

[0036] A computer-readable storage medium having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0037] Obtain a SAR training dataset, which includes multiple SAR training images and ground truth labels for each image.

[0038] The scattering centers in the SAR training image are extracted based on the attribute scattering center model to obtain multiple scattering center points. Then, the SAR training image is enhanced with scattering points based on each of the scattering center points to obtain an enhanced scattering point image.

[0039] The enhanced scattering point image and the corresponding SAR training image are simultaneously input into the target recognition network for training. The target recognition network includes a multi-scale feature extraction unit, a feature fusion unit, and a classification unit. The feature extraction unit performs multi-scale feature extraction on the enhanced scattering point image and the SAR training image to obtain corresponding enhanced scattering point feature maps and SAR training feature maps, respectively. The feature fusion unit fuses the enhanced scattering point feature maps and the SAR training feature maps to obtain a fused feature map. The classification unit obtains a prediction result based on the fused feature map.

[0040] The loss function is calculated based on the prediction results and the corresponding real labels. The parameters of the target recognition network are adjusted based on the result of the loss function until the loss function converges, thus obtaining the trained target recognition network.

[0041] Acquire a SAR image for target recognition, input the SAR image into the trained target recognition network, and obtain the target recognition result of the SAR image.

[0042] The aforementioned SAR image target recognition method, apparatus, computer equipment, and storage medium, when training the target recognition network, first extract the main scattering centers from the SAR training image, then enhance the corresponding regions of the scattering centers to obtain scattering point enhanced images, and simultaneously input these images and the corresponding original SAR training images into the target recognition network. The feature extraction unit performs multi-scale feature extraction on the scattering point enhanced images and the SAR training images respectively to obtain corresponding scattering point enhanced feature maps and SAR training feature maps, and classifies them based on the fused feature maps. By minimizing the network loss function, a trained target recognition network is obtained. When recognizing SAR images, simply inputting them into the trained target recognition network can achieve target recognition. This method can accurately identify targets in SAR images. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating a SAR image target recognition method in one embodiment;

[0044] Figure 2 The results of this method applied to 10 targets in a simulation experiment;

[0045] Figure 3 A visualization of the separability results of fused features in a simulation experiment;

[0046] Figure 4 This is a structural block diagram of a SAR image target recognition device in one embodiment;

[0047] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0048] 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.

[0049] When identifying targets in SAR images, the scattering points in the SAR image actually reflect the target's scattering characteristics, including scattering intensity and phase. By analyzing and extracting the features of the scattering points, target classification and identification can be achieved. Based on this, such as... Figure 1 As shown, this application provides a SAR image target recognition method that combines scattering points in SAR images with optical image processing methods, specifically including the following steps:

[0050] Step S100: Obtain the SAR training dataset, which includes multiple SAR training images and the ground truth labels for each image.

[0051] Step S110: Based on the attribute scattering center model, the scattering centers in the SAR training image are extracted to obtain multiple scattering center points. Then, the SAR training image is enhanced with scattering points according to each scattering center point to obtain an enhanced scattering point image.

[0052] Step S120: The scattering point enhancement image and the corresponding SAR training image are simultaneously input into the target recognition network for training. The target recognition network includes a multi-scale feature extraction unit, a feature fusion unit, and a classification unit. The feature extraction unit performs multi-scale feature extraction on the scattering point enhancement image and the SAR training image to obtain the corresponding scattering point enhancement feature map and SAR training feature map, respectively. The feature fusion unit fuses the scattering point enhancement feature map and the SAR training feature map to obtain a fused feature map. The classification unit obtains the prediction result based on the fused feature map.

[0053] Step S130: Calculate the loss function based on the prediction results and the corresponding real labels. Adjust the parameters of the target recognition network based on the result of the loss function until the loss function converges, and then obtain the trained target recognition network.

[0054] Step S140: Obtain the SAR image to be identified, input the SAR image into the trained target recognition network, and obtain the target recognition result of the SAR image.

[0055] In this embodiment, based on the attribute scattering center theory, the main scattering centers corresponding to the SAR image are extracted, and then the regions corresponding to the scattering centers are enhanced. Multi-scale features of the enhanced image and the original image are fused for network update iteration. A multi-scale convolutional network is used as the backbone network, and by minimizing the network loss function, the model can extract effective classification features from both the enhanced and original SAR images, ultimately achieving accurate classification of targets in the SAR image.

[0056] Specifically, this method proposes a two-channel multi-stage feature fusion (TMSF) model based on attributed scattering centers and a multi-scale convolutional network. This model is used to extract physical characteristics related to SAR and fused features from the original image before classification and recognition. Based on the multi-scale network model and scattering center extraction, this method builds an end-to-end training model, improving the accuracy of SAR target recognition.

[0057] In this embodiment, steps S100 to S130 are the process of training the target recognition network, and step S140 is the process of using the trained target recognition network to recognize SAR images.

[0058] In step S100, when constructing the SAR training dataset, the SAR training images can be images of the same type of target from multiple angles or poses. Targets identified in this method can be tanks, large loading vehicles, etc.

[0059] To improve the accuracy of target recognition in SAR images, due to their unique imaging method, SAR images have unique scattering point features. Therefore, in this embodiment, introducing scattering point features into image classification is more helpful for target classification tasks.

[0060] In this embodiment, before training the target recognition network using the SAR training dataset, scattering points in the SAR training images are extracted using an attribute scattering center model. A scattering center point can be modeled in the following way:

[0061]

[0062] In formula (1), A i f represents the magnitude of a complex number. c Indicates the center frequency of the radar, a p γ represents the frequency-dependent factor. i L represents the orientation dependence of the local property scattering center. i The length of the scattering center is represented by c, and the speed of light is represented by x. i and y i Represents the coordinates in the distance and azimuth directions.

[0063] The parameters in formula (1) are all known or estimated through data processing, for example, f c The given known parameters, x i and y i The parameters can be estimated by processing the data. The parameter estimation process is as follows:

[0064]

[0065] In formula (2), Representing region R k Pixel value, θ k Representing region R k Scattering point parameters in Represented by parameter θ k Image segmentation based on This represents the maximum likelihood estimate of the original image.

[0066] In this embodiment, to avoid obtaining false scattering centers, an iterative optimization method based on CLEAN is also adopted:

[0067]

[0068] In formula (3), D n-1 This represents the image obtained after erasing n-1 scattering centers from the original SAR image, where D0 represents the original SAR image. This represents the response of the scattering center at the i-th scattering center. If:

[0069] D n 2 <D n-1 2 (4)

[0070] Then the estimation is correct; otherwise, the estimated scattering center is false.

[0071] After obtaining the scattering center point in the SAR training image, the image is augmented using the following formula based on the position of the scattering center point in the SAR training image:

[0072] x G =K*I(x i ,y i ,φ i ,L i )*x H +x H (5)

[0073] In formula (5), x G Represents an enhanced image of scattering points, x H Let I(x) represent the SAR training image, K represent the coefficients used to enhance the scattering point region, and I(x) represent the SAR training image. i ,y i ,φ i ,L i () indicates the region to be enhanced, determined by the extracted scattering points.

[0074] In step S120, each SAR training image and its corresponding scattering point enhancement image in the SAR training dataset are input as a training pair into the target recognition network for training.

[0075] In this embodiment, the target recognition network includes three units: a scale feature extraction unit, a feature fusion unit, and a classification unit.

[0076] In this embodiment, the multi-scale feature extraction unit includes a first channel and a second channel for extracting features from the scattering point enhancement image and the SAR training image, respectively. The first channel includes a multi-scale feature extraction layer for extracting a single scattering point enhancement feature map from the scattering point enhancement image, while the second channel includes multiple multi-scale feature extraction layers for extracting multiple SAR training feature maps from the SAR training image.

[0077] Specifically, each multi-scale feature extraction layer includes multiple convolutional layers with different kernel sizes. The number of convolutional layers can be set according to specific circumstances.

[0078] Preferably, each multi-scale feature extraction layer includes three convolutional layers with kernel sizes of 3, 5, and 7, respectively. The multi-scale feature extraction layer can then be represented as:

[0079] z g =w 2-1 conv3(x G )+w 1-2 conv5(x G )+w 2-3 conv7(x G (6)

[0080] In formula (6), z g Represented as the output of the multi-scale feature extraction layer, x G This represents an image enhanced by scattering points. `conv3` indicates a kernel size of 3, `conv5` indicates a kernel size of 5, and `conv7` indicates a kernel size of 7. 2-1 w 2-2 w 2-3 This indicates the fusion of feature information from different receptive fields within a convolutional layer.

[0081] Here, we need to explain formula (6), in which only the scattering point enhancement image x is used. G Taking the original SAR training image as an example, when using it as input, x in formula (6) can also be... G Replace with x H .

[0082] In this embodiment, the second channel includes three multi-scale feature extraction layers connected in sequence, and each multi-scale feature extraction layer outputs a SAR training feature map.

[0083] Therefore, in enhancing the image x with scattering points G and SAR training image x H After being input into the multi-scale feature extraction unit, the corresponding enhanced image z of the scattering points is obtained. G A feature map and the corresponding SAR training image x HMultiple feature maps.

[0084] In this embodiment, a learnable parameter adaptive feature fusion unit is used to fuse multiple output feature images. The feature fusion unit is represented as follows:

[0085] Z = z g w2+z h1 w 1-1 +z h2 w 1-2 +z h3 w 1-3 (7)

[0086] In formula (7), Z represents the fused feature map, z g This represents the enhanced feature map of scattering points, z h1 z h2 z h3 These represent three SAR training feature maps extracted from the SAR training image by the second channel, w2, w3, and w4, respectively. 1-1 w 1-2 w 1-2 Indicates the weight.

[0087] In this embodiment, the classification unit uses a fully connected layer to classify the targets in the fused feature map.

[0088] In step S130, the loss function is calculated based on the prediction results obtained by the classification unit and the corresponding ground truth labels. The parameters in the target recognition network are adjusted based on the results of the cross-entropy loss function. The target recognition network is continuously trained using SAR training images and corresponding scattering point enhancement feature maps in the SAR training dataset until the loss function converges, thus obtaining the trained target recognition network.

[0089] In this embodiment, the entire target recognition network can be represented as:

[0090] Y = f c (g1(x H )Θw1+g2(x G (8)

[0091] In formula (8), g1 and g2 represent the feature extraction processes of the first and second channels, respectively, w1 and w2 represent learnable parameters, and f c This indicates the final fully connected classification.

[0092] In this paper, simulation experiments are also conducted to verify the effectiveness of the target recognition network trained by this method.

[0093] like Figure 2As shown, the results of classifying 10 different tank types in the simulation experiment demonstrate that the network provided by this method can effectively identify and classify targets.

[0094] like Figure 3 As shown in the visualization results of the fusion of 10 target features, the features extracted by this method have good separability.

[0095] As shown in Table 1, the recognition accuracy of the three targets at different pitch angles demonstrates that the proposed method has good generalization ability at different pitch angles.

[0096] Table 1. Test accuracy of this method at different pitch angles during 15° training.

[0097]

[0098] As shown in Table 2, the recognition accuracy of the three targets under different scenarios, angles, and models further demonstrates that the method has good transferability.

[0099] Table 2. Test accuracy rates in different scenarios and for different versions / models.

[0100] BRDM2 constant Scene 3 30° 0.834 T72 A04 constant constant 0.927 ZSU234 constant Scene 2 constant 0.873

[0101] The aforementioned SAR image target recognition method provides a TMSF model composed of a scattering center extraction module, a multi-scale feature extraction module, and a feature fusion module. The scattering center extraction module utilizes the extracted scattering points to construct a region-enhanced SAR image x associated with the scattering center. G The multi-scale feature extraction module enhances the image x H And the original image x H Multi-scale convolution is performed to extract feature maps containing both global and local features. The operation on the original image involves a multi-layer, multi-scale network, with each layer generating a feature map. The image enhancement uses only one multi-scale convolution layer to generate a single feature map. The feature fusion module weights the features extracted by the multi-scale feature extraction module using learnable weights, and finally, a fully connected layer is used for classification.

[0102] This method proposes a TMSF model based on attribute scattering centers, enabling an end-to-end model integrating SAR image processing and target recognition. This TMSF model effectively extracts SAR image features and scattering point-related SAR image features. It improves the accuracy of SAR image recognition and classification, enhances generalization ability across different migration scenarios, and effectively addresses the challenges of identifying special targets in military missions.

[0103] It should be understood that, although Figure 1The 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.

[0104] In one embodiment, such as Figure 4 As shown, a SAR image target recognition device is provided, comprising: a training data module 200, a scattering point image enhancement module 210, a network training module 220, a trained target recognition network acquisition module 230, and a SAR image target recognition module 240, wherein:

[0105] The training data module 200 is used to acquire a SAR training dataset, which includes multiple SAR training images and ground truth labels for each image.

[0106] The scattering point image enhancement module 210 is used to extract the scattering center in the SAR training image based on the attribute scattering center model to obtain multiple scattering center points, and to enhance the SAR training image based on each of the scattering center points to obtain a scattering point enhanced image.

[0107] The network training module 220 is used to simultaneously input the enhanced scattering point image and the corresponding SAR training image into the target recognition network for training. The target recognition network includes a multi-scale feature extraction unit, a feature fusion unit, and a classification unit. The feature extraction unit performs multi-scale feature extraction on the enhanced scattering point image and the SAR training image to obtain corresponding enhanced scattering point feature maps and SAR training feature maps. The feature fusion unit fuses the enhanced scattering point feature maps and the SAR training feature maps to obtain a fused feature map. The classification unit obtains a prediction result based on the fused feature map.

[0108] The trained target recognition network is obtained by module 230, which is used to calculate a loss function based on the prediction results and the corresponding real labels, and adjust the parameters of the target recognition network according to the result of the loss function until the loss function converges, thus obtaining the trained target recognition network.

[0109] The SAR image target recognition module 240 is used to acquire the SAR image to be identified, input the SAR image into the trained target recognition network, and obtain the target recognition result of the SAR image.

[0110] Specific limitations regarding the SAR image target recognition device can be found in the limitations of the SAR image target recognition method described above, and will not be repeated here. Each module in the aforementioned SAR image target recognition device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0111] 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 image target recognition method. 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.

[0112] Those skilled in the art will understand that Figure 5 The 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.

[0113] 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:

[0114] Obtain a SAR training dataset, which includes multiple SAR training images and ground truth labels for each image.

[0115] The scattering centers in the SAR training image are extracted based on the attribute scattering center model to obtain multiple scattering center points. Then, the SAR training image is enhanced with scattering points based on each of the scattering center points to obtain an enhanced scattering point image.

[0116] The enhanced scattering point image and the corresponding SAR training image are simultaneously input into the target recognition network for training. The target recognition network includes a multi-scale feature extraction unit, a feature fusion unit, and a classification unit. The feature extraction unit performs multi-scale feature extraction on the enhanced scattering point image and the SAR training image to obtain corresponding enhanced scattering point feature maps and SAR training feature maps, respectively. The feature fusion unit fuses the enhanced scattering point feature maps and the SAR training feature maps to obtain a fused feature map. The classification unit obtains a prediction result based on the fused feature map.

[0117] The loss function is calculated based on the prediction results and the corresponding real labels. The parameters of the target recognition network are adjusted based on the result of the loss function until the loss function converges, thus obtaining the trained target recognition network.

[0118] Acquire a SAR image for target recognition, input the SAR image into the trained target recognition network, and obtain the target recognition result of the SAR image.

[0119] 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:

[0120] Obtain a SAR training dataset, which includes multiple SAR training images and ground truth labels for each image.

[0121] The scattering centers in the SAR training image are extracted based on the attribute scattering center model to obtain multiple scattering center points. Then, the SAR training image is enhanced with scattering points based on each of the scattering center points to obtain an enhanced scattering point image.

[0122] The enhanced scattering point image and the corresponding SAR training image are simultaneously input into the target recognition network for training. The target recognition network includes a multi-scale feature extraction unit, a feature fusion unit, and a classification unit. The feature extraction unit performs multi-scale feature extraction on the enhanced scattering point image and the SAR training image to obtain corresponding enhanced scattering point feature maps and SAR training feature maps, respectively. The feature fusion unit fuses the enhanced scattering point feature maps and the SAR training feature maps to obtain a fused feature map. The classification unit obtains a prediction result based on the fused feature map.

[0123] The loss function is calculated based on the prediction results and the corresponding real labels. The parameters of the target recognition network are adjusted based on the result of the loss function until the loss function converges, thus obtaining the trained target recognition network.

[0124] Acquire a SAR image for target recognition, input the SAR image into the trained target recognition network, and obtain the target recognition result of the SAR image.

[0125] 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.

[0126] 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.

[0127] 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 image target recognition method, characterized in that, The method includes: Obtain a SAR training dataset, which includes multiple SAR training images and ground truth labels for each image. The scattering centers in the SAR training image are extracted based on the attribute scattering center model to obtain multiple scattering center points. Then, the SAR training image is enhanced with scattering points based on each of the scattering center points to obtain an enhanced scattering point image. The enhanced scattering point image and the corresponding SAR training image are simultaneously input into the target recognition network for training. The target recognition network includes a multi-scale feature extraction unit, a feature fusion unit, and a classification unit. The feature extraction unit performs multi-scale feature extraction on the enhanced scattering point image and the SAR training image to obtain corresponding enhanced scattering point feature maps and SAR training feature maps, respectively. The feature fusion unit fuses the enhanced scattering point feature maps and the SAR training feature maps to obtain a fused feature map. The classification unit obtains a prediction result based on the fused feature map. The loss function is calculated based on the prediction results and the corresponding real labels. The parameters of the target recognition network are adjusted based on the result of the loss function until the loss function converges, thus obtaining the trained target recognition network. Acquire a SAR image to be identified, input the SAR image into the trained target identification network, and obtain the target identification result of the SAR image; The SAR training image is augmented based on each of the aforementioned scattering center points to obtain an augmented scattering image: In the above formula, This represents an enhanced image of scattering points. Represents SAR training images, This represents the coefficient that enhances the scattering point region. This indicates the region to be enhanced, determined by the extracted scattering points. The multi-scale feature extraction unit includes a first channel and a second channel for extracting features from the enhanced scattering point image and the SAR training image, respectively. The first channel includes a multi-scale feature extraction layer for extracting a scattering point enhancement feature map from the scattering point enhancement image; The second channel includes a multi-layer, multi-scale feature extraction layer, used to extract multiple SAR training feature maps from the SAR training image.

2. The SAR image target recognition method according to claim 1, characterized in that, Each multi-scale feature extraction layer comprises multiple convolutional layers with different kernel sizes.

3. The SAR image target recognition method according to claim 2, characterized in that, Each multi-scale feature extraction layer includes three convolutional layers with kernel sizes of 3, 5, and 7 respectively; The multi-scale feature extraction layer is represented as follows: In the above formula, This is represented as the output of the multi-scale feature extraction layer. This is represented as the enhanced image of the scattering points, where conv3 represents a convolution kernel of size 3. This indicates that the kernel size is 5. This indicates that the kernel size is 7. , , This indicates the fusion of feature information from different receptive fields within a convolutional layer.

4. The SAR image target recognition method according to claim 3, characterized in that, The second channel includes three multi-scale feature extraction layers connected in sequence, and each of the multi-scale feature extraction layers outputs a SAR training feature map.

5. The SAR image target recognition method according to claim 4, characterized in that, The feature fusion unit is represented as follows: In the above formula, This represents the fused feature map. This represents the enhanced feature map of the scattering points. , , These represent three SAR training feature maps extracted from the SAR training image by the second channel. , , , Indicates the weight.

6. The SAR image target recognition method according to claim 5, characterized in that, The classification unit uses a fully connected layer to classify the targets in the fused feature map.

7. A SAR image target recognition device, characterized in that, The apparatus employing the SAR image target recognition method according to any one of claims 1 to 6 comprises: The training data module is used to acquire the SAR training dataset, which includes multiple SAR training images and ground truth labels for each image. The scattering point image enhancement module is used to extract the scattering center in the SAR training image based on the attribute scattering center model to obtain multiple scattering center points, and to enhance the SAR training image based on each of the scattering center points to obtain a scattering point enhanced image. The network training module is used to simultaneously input the enhanced scattering point image and the corresponding SAR training image into the target recognition network for training. The target recognition network includes a multi-scale feature extraction unit, a feature fusion unit, and a classification unit. The feature extraction unit performs multi-scale feature extraction on the enhanced scattering point image and the SAR training image to obtain corresponding enhanced scattering point feature maps and SAR training feature maps. The feature fusion unit fuses the enhanced scattering point feature maps and the SAR training feature maps to obtain a fused feature map. The classification unit obtains a prediction result based on the fused feature map. The trained target recognition network module is used to calculate a loss function based on the prediction results and the corresponding real labels, and to adjust the parameters of the target recognition network according to the result of the loss function until the loss function converges, thus obtaining the trained target recognition network. The SAR image target recognition module is used to acquire the SAR image to be identified, input the SAR image into the trained target recognition network, and obtain the target recognition result of the SAR image.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images