ISAR-RID imaging enhancement method, device, equipment and storage medium

Abstract

The present application relates to an ISAR-RID imaging enhancement method, apparatus, device and storage medium. The method comprises: performing RID imaging based on the accumulated echo data corresponding to the RDA image; performing position encoding on the ISAR image to obtain the position encoding of the ISAR image; the position encoding is used to indicate the positions of different pixel points in the ISAR image; and inputting three frames of continuous RID images and the position encoding into a trained ISAR image enhancement network to obtain an enhanced intermediate frame RID image. Compared with traditional large-angle imaging schemes, the ISAR-RID imaging enhancement method based on deep learning solves the problems of low computational efficiency and inaccurate imaging caused by inaccurate reconstructed signals in traditional imaging algorithms such as PFA and SP. This method provides a new and efficient solution for ISAR super-resolution imaging, which is of great significance for subsequent target recognition and target classification.

Description

Technical Field

[0001] The present application relates to the field of radar technology, and in particular to an ISAR-RID imaging enhancement method, apparatus, device and storage medium. Background Art

[0002] ISAR (Inter-SAR) is a radar technology that uses data processing to equate the relative motion between the radar carrier and the target to a longer antenna aperture, replacing the smaller actual antenna aperture. It is commonly used in scenarios such as missile defense, target classification, and environmental monitoring. An ISAR system transmits a signal, receives a target echo, and processes the signal to produce a two-dimensional image of the target in azimuth and range. However, imaging at large rotation angles can lead to azimuth defocusing, resulting in poor imaging quality. Therefore, improving the resolution of ISAR imaging at large rotation angles while simultaneously addressing azimuth defocusing is a critical issue that needs to be addressed.

[0003] ISAR-RID imaging offers excellent focusing but low resolution. Currently, the mainstream algorithms for high-angle ISAR imaging are the Polar Format Algorithm (PFA) and the Subspace Pursuit (SP) algorithm. The PFA uses polar coordinates to store frequency-space observation data. Through two-dimensional interpolation, the polar coordinate data is converted to rectangular coordinates to produce high-resolution images. However, interpolation typically compromises computational efficiency. The SP algorithm divides the imaging target into several small regions to avoid the MTRC phenomenon that occurs during high-angle imaging. Finally, the sub-images are stitched together to obtain a high-resolution image of the entire target, offering improved computational speed and reconstruction probability. However, signal amplitudes often fluctuate randomly. During each backtracking, the SP algorithm retains strong signal components, which can cause strong signal components to obscure weak signal components. Signal components with relatively small amplitudes cannot be correctly reconstructed, resulting in reconstruction errors. Furthermore, spurious reconstructed signals can be generated near signals with relatively large amplitudes.

[0004] As observation time accumulates, the imaging angle increases, and the ISAR azimuth resolution improves. However, if the observation angle increases too much, the target's scattering points will shift, a phenomenon known as cross-range cell migration (MTRC). This can cause one scattering point to shift to the position of another, affecting the imaging. Furthermore, the image will not only be distorted, but in severe cases, it may even become completely defocused, rendering classic ISAR imaging algorithms (such as RDA) ineffective. Summary of the Invention

[0005] Based on this, it is necessary to provide an ISAR-RID imaging enhancement method, device, equipment and storage medium to address the defocus problem of large-angle imaging.

[0006] An ISAR-RID imaging enhancement method, the method comprising:

[0007] RID imaging is performed based on the accumulated echo data corresponding to the RDA image to obtain continuous frame RID images.

[0008] Position coding is performed on the ISAR image to obtain a position code of the ISAR image; the position code is used to indicate the positions of different pixel points in the ISAR image.

[0009] The three consecutive RID images and the position code are input into a trained ISAR image enhancement network to obtain an enhanced intermediate frame RID image; the ISAR image enhancement network includes multiple convolution modules, the input of the first convolution module is the result of splicing the position code and the three consecutive RID images in the channel dimension, the input of the second to last convolution modules is the result of splicing the output of the previous convolution module and the position code in the channel dimension, and the last convolution module outputs the enhanced intermediate frame RID image.

[0010] In one embodiment, performing RID imaging based on the accumulated echo data corresponding to the RDA image to obtain multiple frames of continuous RID images includes:

[0011] The RID imaging technology is used for the accumulated echo data corresponding to the defocused RDA image to capture RID images of different frames and obtain multi-frame continuous RID images.

[0012] In one embodiment, the position coding includes radius coding and angle coding.

[0013] The radius code is used to display the distance from a pixel point to the image center in the ISAR image.

[0014] The angle code is used to mark the angle between each pixel point in the ISAR image and the positive half axis of the x-axis.

[0015] In one embodiment, the last convolution module includes a convolution layer, and the other convolution modules include convolution layers and ReLU activation functions; three consecutive RID images and the position code are input into a trained ISAR image enhancement network to obtain an enhanced intermediate frame RID image, including:

[0016] The position code and three frames of continuous RID images are spliced ​​in the channel dimension and input into the first convolution module. After being processed by the convolution layer and activated by the ReLU activation function, the first convolution feature is obtained.

[0017] The first convolution feature and the position code are spliced ​​in the channel dimension and then input into two convolution modules to obtain a second convolution feature.

[0018] For the third convolution module to the last convolution module, the input is the result of splicing the output of the previous convolution module and the position code in the channel dimension; the output of the last convolution module is the enhanced intermediate frame RID image.

[0019] In one embodiment, three consecutive RID images and the position code are input into a trained ISAR image enhancement network to obtain an enhanced intermediate RID image, and the step further includes:

[0020] A plurality of sample label pairs for training the network are generated, and the label pairs and the position encoding of the ISAR image are used as training samples; the label pairs include three consecutive RID images and an ideal RID image of an intermediate frame.

[0021] Construct an ISAR image enhancement network.

[0022] The loss function for constructing the ISAR image enhancement network is:

[0023]

[0024] in, is the loss function, is the i-th predicted value, y (i) is the i-th true value, and m is the total number of pixels.

[0025] The ISAR image enhancement network is trained according to the training samples and the unbalanced loss function to obtain a trained ISAR image enhancement network.

[0026] In one embodiment, the tag pair generating step includes:

[0027] The ideal coordinates of the target in the RID image are convolved with the Gaussian kernel function to obtain the ideal RID image, and the ideal RID image is used as the label.

[0028] In one embodiment, the method further comprises: selecting mean square error and peak signal-to-noise ratio to evaluate the enhanced intermediate frame RID image.

[0029] An ISAR-RID imaging enhancement device, comprising:

[0030] The ISAR-RID imaging module is used to perform RID imaging based on the accumulated echo data corresponding to the RDA image to obtain continuous frame RID images.

[0031] The position coding module is used to perform position coding on the ISAR image to obtain the position coding of the ISAR image; the position coding is used to indicate the positions of different pixel points in the ISAR image.

[0032] The ISAR-RID image enhancement module is used to input three consecutive RID images and the position code into a trained ISAR image enhancement network to obtain an enhanced intermediate frame RID image; the ISAR image enhancement network includes multiple convolution modules, the input of the first convolution module is the result of splicing the position code and the three consecutive RID images in the channel dimension, the input of the second to last convolution modules is the result of splicing the output of the previous convolution module and the position code in the channel dimension, and the last convolution module outputs the enhanced intermediate frame RID image.

[0033] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0034] RID imaging is performed based on the accumulated echo data corresponding to the RDA image to obtain continuous frame RID images.

[0035] Position coding is performed on the ISAR image to obtain a position code of the ISAR image; the position code is used to indicate the positions of different pixel points in the ISAR image.

[0036] The three consecutive RID images and the position code are input into a trained ISAR image enhancement network to obtain an enhanced intermediate frame RID image; the ISAR image enhancement network includes multiple convolution modules, the input of the first convolution module is the result of splicing the position code and the three consecutive RID images in the channel dimension, the input of the second to last convolution modules is the result of splicing the output of the previous convolution module and the position code in the channel dimension, and the last convolution module outputs the enhanced intermediate frame RID image.

[0037] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0038] RID imaging is performed based on the accumulated echo data corresponding to the RDA image to obtain continuous frame RID images.

[0039] Position coding is performed on the ISAR image to obtain a position code of the ISAR image; the position code is used to indicate the positions of different pixel points in the ISAR image.

[0040] The three consecutive RID images and the position code are input into a trained ISAR image enhancement network to obtain an enhanced intermediate frame RID image; the ISAR image enhancement network includes multiple convolution modules, the input of the first convolution module is the result of splicing the position code and the three consecutive RID images in the channel dimension, the input of the second to last convolution modules is the result of splicing the output of the previous convolution module and the position code in the channel dimension, and the last convolution module outputs the enhanced intermediate frame RID image.

[0041] The aforementioned ISAR-RID imaging enhancement method, apparatus, device, and storage medium include: performing RID imaging based on the accumulated echo data corresponding to the RDA image to obtain continuous-frame RID images; performing position encoding on the ISAR image to obtain position codes for the ISAR image; the position codes are used to indicate the locations of different pixels in the ISAR image; and inputting three consecutive RID image frames and the position codes into a trained ISAR image enhancement network to obtain an enhanced intermediate-frame RID image. Compared to traditional large-angle imaging schemes, the deep learning-based ISAR-RID imaging enhancement method addresses the issues of low computational efficiency and inaccurate reconstructed signals resulting from traditional imaging algorithms such as PFA and SP. This method provides a new and efficient solution for ISAR super-resolution imaging and has significant implications for subsequent target recognition and classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 1 is an overall flow chart of an ISAR-RID imaging enhancement method in one embodiment;

[0043] Figure 2 1 is a flow chart of an ISAR-RID imaging enhancement method according to an embodiment;

[0044] Figure 3 Schematic diagram of the ISAR-RID image enhancement network structure in one embodiment;

[0045] Figure 4 For radius encoding in another embodiment;

[0046] Figure 5 Angle encoding in another embodiment;

[0047] Figure 6 is a structural block diagram of an ISAR-RID imaging enhancement device in one embodiment;

[0048] Figure 7 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0050] The ISAR-RID imaging enhancement method proposed in this application first uses the RID algorithm to process ISAR echoes and obtain RID images. The real and imaginary parts of three consecutive RID images are then input into a neural network. By training with ideal RID images of intermediate frames as labels, the network can learn the optimal mapping between input and output and ultimately generate super-resolution ISAR images. The overall process of the ISAR-RID imaging enhancement method using a neural network is as follows: Figure 1 shown.

[0051] In one embodiment, Figure 2 As shown, an ISAR-RID imaging enhancement method is provided, which includes the following steps:

[0052] Step 100: performing RID imaging based on the accumulated echo data corresponding to the RDA image to obtain continuous frame RID images;

[0053] Specifically, to improve the resolution of ISAR images while preventing azimuth defocusing, RID imaging is performed on the accumulated echo data corresponding to the defocused RDA image. The RID (Range-Instantaneous Doppler) algorithm uses the same processing as the RD algorithm in the range direction (by transmitting a signal with a large time-bandwidth product to achieve high resolution in the range direction). In the azimuth direction, time-frequency transforms such as the short-time Fourier transform (SFT) are used to obtain the target's instantaneous Doppler distribution information, thereby producing a two-dimensional, high-resolution image of the target.

[0054] By utilizing RID imaging technology, RID images of different frames can be captured, each containing information about the target's different scattering points. These consecutive RID images are then fed into a network for image enhancement, allowing the network to better learn and exploit target features. To achieve high Doppler resolution in RID images, the time window length should be maximized while ensuring that the RID image remains focused in azimuth. Therefore, the echo data corresponding to the defocused RDA imaging results is divided into three parts for RID imaging. This ensures that the RDA image corresponding to the total time covered by the three-frame window is just defocused, while the single-frame RID image is not. Finally, three consecutive ISAR-RID images are simultaneously fed into the network to enhance the intermediate frame image. This approach improves the resolution and reduces sidelobes of the intermediate-frame ISAR images while also addressing the defocus issue in RDA imaging.

[0055] Step 102: Position coding is performed on the ISAR image to obtain a position code of the ISAR image; the position code is used to indicate the positions of different pixels in the ISAR image.

[0056] Specifically, for ISAR images, the greater the distance from the scene center, the greater the degree of defocus. Furthermore, defocus in the azimuth direction is more severe than defocus in the range direction. In other words, the degree of defocus in the horizontal and vertical directions of ISAR images differs. Based on these two characteristics, a position encoding can be designed to indicate the location of pixels in ISAR images, enabling the subsequently designed neural network to better learn the features of ISAR images.

[0057] Considering the different characteristics of the middle and edge parts of ISAR images, radius coding and angle coding are designed for ISAR images to indicate the positions of different pixels in ISAR images.

[0058] Step 104: Input the three consecutive RID images and the position code into the trained ISAR image enhancement network to obtain an enhanced intermediate frame RID image. The ISAR image enhancement network includes multiple convolution modules. The input of the first convolution module is the result of splicing the position code and the three consecutive RID images in the channel dimension. The input of the second to last convolution modules is the result of splicing the output of the previous convolution module and the position code in the channel dimension. The last convolution module outputs the enhanced intermediate frame RID image.

[0059] Specifically, considering the different characteristics of the middle and edge parts of the ISAR image, a position encoding is designed for the ISAR image to indicate the positions of different pixels in the ISAR image. The total number of input channels of the network is 10, including the real and imaginary parts of three consecutive RID images received by the network, and four channels for position encoding. The ISAR-RID image enhancement network is as follows: Figure 3 shown.

[0060] The aforementioned ISAR-RID imaging enhancement method includes: performing RID imaging based on the accumulated echo data corresponding to the RDA image to obtain continuous RID image frames; performing position encoding on the ISAR image to obtain position codes for the ISAR image; the position codes are used to indicate the locations of different pixels in the ISAR image; and inputting three consecutive RID image frames and the position codes into a trained ISAR image enhancement network to obtain an enhanced intermediate RID image frame. Compared to traditional large-angle imaging schemes, the deep learning-based ISAR-RID imaging enhancement method addresses the issues of low computational efficiency and inaccurate reconstructed signals resulting from traditional imaging algorithms such as PFA and SP. This method provides a new and efficient solution for ISAR super-resolution imaging and has significant implications for subsequent target recognition and classification.

[0061] In one embodiment, step 100 includes: using RID imaging technology on the accumulated echo data corresponding to the defocused RDA image, capturing RID images of different frames, and obtaining multiple frames of continuous RID images.

[0062] In one embodiment, the position coding includes: radius coding and angle coding; the radius coding is used to display the distance from the pixel point in the ISAR image to the image center; the angle coding is used to mark the angle between each pixel point in the ISAR image and the positive half axis of the x-axis.

[0063] Specifically, the position coding of ISAR images includes radius coding and angle coding. The radius coding is as follows: Figure 4 As shown, the angle encoding is as follows Figure 5 As shown in the figure, the radius encoding shows the distance of the pixel from the center of the image. Furthermore, a complementary radius encoding is required to prevent the network from using the radius as a weight during training. It is summed with the radius corresponding to the radius encoding at the same location to obtain the maximum radius value. The angle encoding marks the angle between each pixel and the positive x-axis. Similarly, a corresponding complementary angle encoding is required.

[0064] In one embodiment, the last convolution module includes a convolution layer, and the other convolution modules include convolution layers and Relu activation functions; step 104 includes: splicing the position code and three frames of continuous RID images in the channel dimension and inputting them into the first convolution module, processing them through the convolution layer and then activating them through the Relu activation function to obtain a first convolution feature; splicing the first convolution feature and the position code in the channel dimension and inputting them into the second convolution module to obtain a second convolution feature; for the third convolution module to the last convolution module, the input is the result of splicing the output of the previous convolution module and the position code in the channel dimension; the output of the last convolution module is the enhanced intermediate frame RID image.

[0065] In one embodiment, the following steps are further included before step 104:

[0066] Step 200: Generate multiple groups of sample label pairs for training the network, and use the label pairs and the position encoding of the ISAR image as training samples; the label pairs include three consecutive RID images and an ideal RID image of the intermediate frame.

[0067] Step 202: Construct an ISAR image enhancement network.

[0068] Step 204: Construct the loss function of the ISAR image enhancement network as:

[0069]

[0070] in, is the loss function, is the i-th predicted value, y (i) is the i-th true value, and m is the total number of pixels.

[0071] Specifically, in machine learning, the loss function plays a crucial role, quantifying the difference between predicted and true values. The smaller the loss value, the more accurate the model's predictions. In deep learning-based radar imaging methods, an end-to-end training system is used to minimize the loss function between predicted and true images, achieving accurate model predictions.

[0072] However, during training, the imbalance between the number of pixels occupied by the target and the number of pixels occupied by non-target areas can cause the network to favor regressing the non-target areas over the target itself. Existing loss functions cannot optimize the network parameters for this characteristic of ISAR images, thus failing to achieve better training results. Therefore, it is crucial to design an appropriate loss function for ISAR images to help the network learn the correct features, achieving faster convergence and better prediction results.

[0073] A similar imbalance problem also exists in the field of object detection. To solve the problem of class imbalance in object detection, Focal loss is used to adjust the ratio of positive and negative samples based on the cross entropy loss function. The formula of the cross entropy loss function is as follows:

[0074]

[0075] The Focal loss formula is as follows:

[0076]

[0077] Here, α is a factor used to balance the ratio of positive and negative samples, y specifies the true value category, and y'∈[0,1] is the model's estimated probability for the category y=1. γ is used to adjust the rate at which sample weights are reduced. When γ is 0, Focalloss degenerates into a cross-entropy loss function. As γ increases, the impact of the adjustment factor also increases. In the field of image regression, MSE is the most commonly used loss function, and its formula is:

[0078]

[0079] in is the predicted value, y (i) is the true value, and m is the total number of pixels. However, the MSE loss function uniformly weights all pixels in the image and cannot solve the imbalance problem faced by ISAR images as mentioned above.

[0080] Combining the concept of focal loss with the MSE loss function can address this imbalance. First, the values ​​of the pixels in the label are normalized, and then the loss function is weighted according to these normalized values. Pixels corresponding to the target are assigned higher weights, while pixels at non-target locations are assigned lower weights. This approach allows the network to focus more on accurately reconstructing the target area during training.

[0081] The improved loss function is shown in formula (1).

[0082] In order to evaluate the experimental ISAR imaging enhancement effect, mean square error (MSE) and peak signal-to-noise ratio (PSNR) are selected as measurement indicators.

[0083] Step 206: The ISAR image enhancement network is trained according to the training samples and the loss function to obtain a trained ISAR image enhancement network.

[0084] In one embodiment, the label pair generating step includes: convolving the ideal coordinates of the target in the RID image with a Gaussian kernel function to obtain an ideal RID image, and using the ideal RID image as the label.

[0085] In one embodiment, the method further comprises: selecting mean square error and peak signal-to-noise ratio to evaluate the enhanced intermediate frame RID image.

[0086] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0087] In a specific embodiment, Figure 1 As shown, an ISAR-RID imaging enhancement method is provided, comprising the following steps:

[0088] Step S1 generates sample label pairs for training the network.

[0089] 3000 sets of sample label pairs are generated as training sets and 1000 sets of sample label pairs are generated as test sets. The image size in the dataset is 256×256.

[0090] Step S2 performs ISAR image position encoding.

[0091] For ISAR images, the greater the distance from the scene center, the greater the degree of defocus. Furthermore, defocus in the azimuth direction is more severe than in the range direction. In other words, the horizontal and vertical defocus levels of ISAR images differ. To address these two characteristics, a position encoding method was designed to indicate the location of pixels in ISAR images, enabling the subsequent neural network to better learn the features of ISAR images.

[0092] ISAR image position encoding includes radius encoding and angle encoding. The radius encoding indicates the distance of a pixel from the image center. Furthermore, a complementary radius encoding is required to prevent the network from using the radius as a weight during training. This encoding is summed with the radius corresponding to the same position in the radius encoding to obtain the maximum radius value. The angle encoding indicates the angle between each pixel and the positive x-axis. Similarly, a corresponding complementary angle encoding is required.

[0093] Step S3 designs the neural network structure.

[0094] Considering the different characteristics of the middle and edge parts of the ISAR image, a position encoding is designed for the ISAR image to indicate the position of different pixels in the ISAR image. The total number of input channels of the network is 10, including the real and imaginary parts of three consecutive RID images received by the network, and four channels for position encoding. In addition, the position encoding is spliced ​​on the channel dimension after each convolutional layer. The designed ISAR-RID image enhancement network is as follows Figure 3 shown.

[0095] In step S4, a loss function as shown in formula (1) is designed according to the characteristics of the ISAR image.

[0096] Step S5 evaluates the ISAR imaging enhancement performance.

[0097] The mean square error (MSE) and peak signal-to-noise ratio (PSNR) are chosen to evaluate the predicted images for experiments with ideal images.

[0098] MSE calculates the mean squared error (MSE) between the pixel values ​​of the ideal image and the predicted image. By doing so, it can assess the degree of difference between the two images and is a primary method for objectively assessing image quality. It is given by the following formula:

[0099]

[0100] Where I represents an ideal radar image of size m×n, and P represents the image predicted by the network. m and n represent the number of pixels in the image in the horizontal and vertical directions, respectively. The smaller the MSE value, the better the network's regression performance on the image.

[0101] PSNR is the ratio of the maximum pixel value to the noise intensity, which is mainly used to measure the ability of the algorithm to remove noise. The higher the PNSR value, the better the noise suppression performance of the algorithm. The formula of PSNR is as follows:

[0102]

[0103] Preferably, a stepped frequency signal with a frequency range of 8.8-9.2 GHz, a step length of 5 MHz, and a step number of 80 is selected as the ISAR radar transmission signal, and the imaging scene of the designed point target is (-8, 8) meters in the range direction and (-6, 6) meters in the azimuth direction. Among them, the range direction refers to the axis parallel to the radar propagation direction to the target, and the azimuth direction refers to the axis perpendicular to the range direction. In this scene, a maximum of 200 point targets are randomly generated. The target rotation speed is set to 37.497 ° / s. The window length of STFT is 158, and noverlap is 79. The range of observation angles corresponding to each frame of RID image is 6 degrees, and the observation angle rotates 3 degrees between adjacent frames. Each set of sample label pairs includes three consecutive RID images and the ideal RID image of the intermediate frame. The ideal coordinates of the target in the RID image are convolved with the Gaussian kernel function to obtain the ideal RID image, i.e., the label. Its expression is as follows:

[0104] I(x,y)=∫∫T(u,v)·h(xu,yv)dudv (7)

[0105] Where T(u,v) is the target function, I(x,y) represents the synthetic reconstructed image of the target, and h(x,y) is the PSF of the ISAR imaging system. It should be noted that the Gaussian kernel function is chosen as the point spread function (PSF) because it reduces the sidelobes in the label image compared to the sinc function.

[0106] Based on the above data, the proposed deep learning method is used to enhance ISAR-RID imaging. Finally, the imaging indicators under different signal-to-noise ratio environments are obtained as shown in Table 1. Table 1 shows that the proposed method can effectively enhance the ISAR imaging effect.

[0107] Table 1. ISAR-RID enhanced imaging performance based on deep learning

[0108] Signal-to-noise ratio (dB) MSE PSNR(dB) Noiseless 0.0019 27.1167 -10 0.0019 27.1167 -20 0.0022 26.5856 -30 0.0025 26.1045 -40 0.0032 24.9574

[0109] In one embodiment, Figure 6 As shown, an ISAR-RID imaging enhancement device is provided, comprising: an ISAR-RID imaging module, a position encoding module and an ISAR-RID image enhancement module, wherein:

[0110] ISAR-RID imaging module is used to perform RID imaging based on the accumulated echo data corresponding to the RDA image to obtain continuous frame RID images;

[0111] A position coding module is used to perform position coding on the ISAR image to obtain a position code of the ISAR image; the position code is used to indicate the positions of different pixel points in the ISAR image;

[0112] The ISAR-RID image enhancement module is used to input three consecutive RID images and position codes into the trained ISAR image enhancement network to obtain an enhanced intermediate frame RID image. The ISAR image enhancement network includes multiple convolution modules. The input of the first convolution module is the result of splicing the position code and three consecutive RID images in the channel dimension. The input of the second to last convolution modules is the result of splicing the output of the previous convolution module and the position code in the channel dimension. The last convolution module outputs the enhanced intermediate frame RID image.

[0113] In one embodiment, the ISAR-RID imaging module is further configured to use RID imaging technology on the accumulated echo data corresponding to the defocused RDA image to capture RID images of different frames to obtain multiple frames of continuous RID images.

[0114] In one embodiment, the position coding in the position coding module includes: radius coding and angle coding; the radius coding is used to display the distance from the pixel point in the ISAR image to the image center; the angle coding is used to mark the angle between each pixel point in the ISAR image and the positive half axis of the x-axis.

[0115] In one embodiment, the last convolution module includes a convolution layer, and the other convolution modules include convolution layers and Relu activation functions; the ISAR-RID image enhancement module is also used to splice the position code and three frames of continuous RID images in the channel dimension and input them into the first convolution module, and after processing through the convolution layer and then activation through the Relu activation function, obtain the first convolution feature; the first convolution feature and the position code are spliced ​​in the channel dimension and input into the two convolution modules to obtain the second convolution feature; for the third convolution module to the last convolution module, the input is the result of splicing the output of the previous convolution module and the position code in the channel dimension; the output of the last convolution module is the enhanced intermediate frame RID image.

[0116] In one embodiment, the ISAR-RID image enhancement module further includes an ISAR-RID image enhancement network training module, which is configured to generate multiple sets of sample label pairs for training the network, and use the label pairs and the position encoding of the ISAR images as training samples; the label pairs include three consecutive RID images and an ideal RID image of an intermediate frame; construct an ISAR image enhancement network; and construct a loss function for the ISAR image enhancement network, as shown in Formula (1). The ISAR image enhancement network is trained based on the training samples and the loss function to obtain a trained ISAR image enhancement network.

[0117] In one embodiment, the label pair generation step in the ISAR-RID image enhancement network training module includes: convolving the ideal coordinates of the target in the RID image with a Gaussian kernel function to obtain an ideal RID image, and using the ideal RID image as a label.

[0118] In one embodiment, the method further includes a performance evaluation module for selecting mean square error and peak signal-to-noise ratio to evaluate the enhanced intermediate frame RID image.

[0119] The specific definition of the ISAR-RID imaging enhancement device can be found in the definition of the ISAR-RID imaging enhancement method above and will not be further elaborated here. Each module in the aforementioned ISAR-RID imaging enhancement device may be implemented in whole or in part via software, hardware, or a combination thereof. Each module may be embedded in or independent of a processor within a computer device in hardware form, or may be stored in a computer device memory in software form, allowing the processor to invoke and execute the corresponding operations of each module.

[0120] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, an ISAR-RID imaging enhancement method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a key, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse.

[0121] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0122] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiment when executing the computer program.

[0123] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiment are implemented.

[0124] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0125] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0126] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. An ISAR-RID imaging enhancement method, characterized in that: The method comprises: Perform RID imaging based on the accumulated echo data corresponding to the RDA image to obtain continuous frame RID images; Performing position coding on the ISAR image to obtain a position code of the ISAR image; wherein the position code is used to indicate the positions of different pixel points in the ISAR image; Generating multiple sets of sample label pairs for training the network, and using the label pairs and the position codes of the ISAR images as training samples; the label pair generation step includes: convolving the ideal coordinates of the target in the RID image with a Gaussian kernel function to obtain an ideal RID image, and using the ideal RID image as a label; the label pair includes three consecutive RID images and an ideal RID image of an intermediate frame; Construct ISAR image enhancement network; The loss function for constructing the ISAR image enhancement network is: in, is the loss function, It is i predicted values, It is i A true value, is the total number of pixels; Training the ISAR image enhancement network according to the training samples and the loss function to obtain a trained ISAR image enhancement network; Inputting three consecutive RID images and the position code into a trained ISAR image enhancement network to obtain an enhanced intermediate frame RID image; the ISAR image enhancement network includes multiple convolution modules, wherein the input of the first convolution module is the result of splicing the position code and the three consecutive RID images in the channel dimension, the input of the second to last convolution modules is the result of splicing the output of the previous convolution module and the position code in the channel dimension, and the last convolution module outputs the enhanced intermediate frame RID image; Among them, the convolution module includes convolution layer and Relu activation function; Inputting three consecutive RID images and the position code into the trained ISAR image enhancement network to obtain an enhanced intermediate frame RID image, including: The position code and three consecutive RID images are spliced ​​in the channel dimension and input into the first convolution module. After being processed by the convolution layer and activated by the ReLU activation function, the first convolution feature is obtained. Concatenating the first convolution feature and the position code in the channel dimension and inputting the convolution feature into two convolution modules to obtain a second convolution feature; For the third convolution module to the last convolution module, the input is the result of splicing the output of the previous convolution module and the position code in the channel dimension; the output of the last convolution module is the enhanced intermediate frame RID image.

2. The method according to claim 1, characterized in that RID imaging is performed based on the accumulated echo data corresponding to the RDA image to obtain multiple frames of continuous RID images, including: The RID imaging technology is used for the accumulated echo data corresponding to the defocused RDA image to capture RID images of different frames and obtain multi-frame continuous RID images.

3. The method according to claim 1, characterized in that The position coding includes: radius coding and angle coding; The radius code is used to display the distance from the pixel point to the image center in the ISAR image; The angle code is used to mark the angle between each pixel point in the ISAR image and the positive half axis of the x-axis.

4. The method according to claim 1, wherein The method further comprises: The mean square error and peak signal-to-noise ratio are selected to evaluate the enhanced intermediate frame RID image.

5. An ISAR-RID imaging enhancement device, characterized in that: The device comprises: ISAR-RID imaging module is used to perform RID imaging based on the accumulated echo data corresponding to the RDA image to obtain continuous frame RID images; A position coding module is used to perform position coding on the ISAR image to obtain a position code of the ISAR image; the position code is used to indicate the positions of different pixels in the ISAR image; The ISAR-RID image enhancement network training module is used to generate multiple sets of sample label pairs for training the network, and use the label pairs and the position encoding of the ISAR image as training samples; the label pair generation step includes: convolving the ideal coordinates of the target in the RID image with a Gaussian kernel function to obtain an ideal RID image, and using the ideal RID image as a label; the label pair includes three consecutive RID images and the ideal RID image of the intermediate frame; constructing the ISAR image enhancement network; the loss function for constructing the ISAR image enhancement network is: in, is the loss function, It is i predicted values, It is i A true value, is the total number of pixels; Training the ISAR image enhancement network according to the training samples and the loss function to obtain a trained ISAR image enhancement network; An ISAR-RID image enhancement module is configured to input three consecutive RID images and the position code into a trained ISAR image enhancement network to obtain an enhanced intermediate RID image. The ISAR image enhancement network includes multiple convolution modules, wherein the input of the first convolution module is the result of splicing the position code and the three consecutive RID images in the channel dimension, the input of the second to last convolution modules is the result of splicing the output of the previous convolution module and the position code in the channel dimension, and the last convolution module outputs the enhanced intermediate RID image. Among them, the convolution module includes a convolution layer and a Relu activation function; the ISAR-RID image enhancement module is also used to splice the position code with three frames of continuous RID images in the channel dimension and input them into the first convolution module, and after processing through the convolution layer and then activation by the Relu activation function, obtain the first convolution feature; the first convolution feature and the position code are spliced ​​in the channel dimension and input into the second convolution module to obtain the second convolution feature; for the third convolution module to the last convolution module, its input is the result of splicing the output of the previous convolution module with the position code in the channel dimension; the output of the last convolution module is the enhanced intermediate frame RID image.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images