Self-supervised deep learning sidelobe and clutter suppression method for ship targets in sar images

Abstract

The application relates to a self-supervised deep learning sidelobe and clutter suppression method for a SAR image ship target, and relates to the technical field of image processing. The existing sidelobe and noise suppression technology often sacrifices the image detail information while reducing the noise, thereby affecting the resolution of the image. The application can suppress the sidelobe and the clutter while retaining the ship target details, and the influence on the image resolution is avoided.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a self-supervised deep learning method for suppressing sidelobes and clutter in SAR images of ship targets. Background Technology

[0002] Synthetic Aperture Radar (SAR) remote sensing technology has become an important tool for marine monitoring due to its ability to penetrate clouds and rain curtains, effectively monitoring ships at sea under various weather conditions. However, the imaging characteristics of SAR images, such as the sidelobe effect of strong scattering points and the accompanying speckle noise, severely affect image quality. These problems not only reduce image usability but also pose challenges to subsequent processing tasks such as image segmentation, target detection, and recognition. Therefore, exploring effective speckle noise suppression methods is of great significance for improving SAR image quality.

[0003] Currently, traditional sidelobe and noise suppression techniques often sacrifice image detail while reducing noise, thus affecting image resolution. Summary of the Invention

[0004] The purpose of this invention is to address the problem that existing sidelobe and noise suppression techniques often sacrifice image detail while reducing noise, thus affecting image resolution. The invention proposes a self-supervised deep learning sidelobe and clutter suppression method for ship targets in SAR images.

[0005] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0006] Self-supervised deep learning methods for suppressing sidelobes and clutter in SAR imagery of ship targets include:

[0007] Step 1: Acquire SAR images containing ship targets;

[0008] Step 2: Process the SAR image using tone mapping technology;

[0009] Step 3: Use a clustering algorithm to extract the target rectangle contour from the image processed by tone mapping technology;

[0010] Step 4: Use a weighted function to filter and cluster the extracted target rectangle contours to obtain the contours of the ship targets, and map the contours of the ship targets onto the SAR image from Step 1 to obtain the mapped image.

[0011] Step 5: Using the SAR image and the mapped image from Step 1, obtain the loss function, and update the supervised deep learning model using the loss function until the loss function converges, resulting in an image that has been denoised and sidelobe suppressed.

[0012] Furthermore, the SAR image containing the ship target in step one is a preprocessed image, and the preprocessing includes complex data conversion, multi-view processing, filtering processing, and geocoding.

[0013] Furthermore, the specific steps of step two are as follows:

[0014] Step 21: Calculate the logarithmic mean Y of the SAR image containing the ship target. av :

[0015] Step 22: Obtain the key value using the brightness value and logarithmic mean of the pixels in the image;

[0016] Steps two and three: Partition the SAR image using the key value. The specific steps are as follows:

[0017] The threshold L for low-light regions in a brightness image is obtained using the key value. T Threshold L for high illumination regions in brightness images H And in the SAR image, the pixel brightness is less than L T Yes, it is classified as a low-light area, and the pixel brightness is greater than L. H Yes, it falls into the high-illuminance area, and the pixel brightness is between L T and L H Those areas between these levels are classified as medium illumination zones.

[0018] Step 24: Compress the partitioned SAR image using an adaptive logarithmic method to obtain the display brightness after tone mapping.

[0019] Furthermore, the logarithmic mean Y of the SAR image containing the ship target av Represented as:

[0020]

[0021] Where Y(x) represents the luminance channel in the XYZ space when the image is converted, W is the total number of pixels in the image, and δ represents a positive number, with δ being 10. -4 , where x represents the pixel index in the image.

[0022] Furthermore, the key value is represented as:

[0023]

[0024] Among them, Y max Y represents the maximum brightness value of a pixel. min This represents the minimum brightness value of a pixel.

[0025] Furthermore, the L T and L H Represented as:

[0026] L T =L max -(0.9+0.1key)(L max -L min )

[0027] L H =L min +[0.6+0.4×(1-key)](L max -L min )

[0028] Among them, L max and L min These represent the maximum and minimum values ​​of the image after normalization to a value between 0 and 1.

[0029] Furthermore, the display brightness after tone mapping is expressed as:

[0030]

[0031] Among them, L i (x,y) represents the display brightness after tone mapping, L w (x,y) and L w,max These represent the original brightness and the maximum value of the original brightness, respectively, where q and k are constants. i ∈[1,∞],k i ∈[1,∞]; i=T,W,H; j=1,2,j=1,S1=10,j=2,S2=2,S represents the base of the logarithmic mapping.

[0032] Furthermore, the clustering algorithm in step three is the K-means clustering algorithm, specifically expressed as follows:

[0033]

[0034] Where, x i Let j represent the i-th sample, i.e., the i-th pixel in the image, and let j represent x. i clusters, p j denoted by , where n represents the center point of the cluster and n represents the total number of samples.

[0035] Furthermore, the specific steps of step four are as follows:

[0036] Step 41: Process the extracted target rectangular contour using a bilateral filtering algorithm;

[0037] Step 42: Perform clustering processing on the bilaterally filtered image to obtain the refined target contour;

[0038] Step 43: Map the refined target contour back to the original SAR image to obtain the mapped image.

[0039] Furthermore, the bilateral filtering algorithm is expressed as follows:

[0040]

[0041] Where w(i,j,k,l) ​​represents the weight function in the bilateral filtering algorithm, d(i,j,k,l) ​​represents the weighted combination of the values ​​of neighboring pixels, r(i,j,k,l) ​​represents the value range template coefficients, (k,l) represents the center coordinates of the template window, (i,j) represents the coordinates of the other coefficients of the template window, and σ d Let represent the standard deviation of the Gaussian function, f(i,j) represent the pixel value at point (i,j) in the image, f(k,l) represent the pixel value at point (k,l) in the image, and σ r This represents the standard deviation of the Gaussian function.

[0042] The beneficial effects of this invention are:

[0043] This application has the ability to suppress side lobes and clutter while preserving details of the ship target, thus avoiding any impact on image resolution.

[0044] This application achieves efficient denoising and sidelobe suppression of SAR images by combining a self-supervised deep learning framework with tone mapping techniques and clustering algorithms, without relying on large amounts of labeled data or pre-known noise distributions. This method can automatically learn and preserve the details of ship targets, while compressing high dynamic range and improving image contrast and detail visibility, significantly enhancing the quality of SAR images and the accuracy of subsequent processing tasks. Furthermore, this application reduces the demand for computational resources, improves processing efficiency, and makes it more suitable for practical applications. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the steps involved in implementing this application;

[0046] Figure 2 The result image is obtained through tone mapping, clustering, and weighting functions;

[0047] Figure 3 The figure shows a comparison of the experimental results of the proposed self-supervised deep learning method. Detailed Implementation

[0048] It should be noted that, where there is no conflict, the various embodiments disclosed in this application can be combined with each other.

[0049] Specific implementation method one: Refer to Figure 1 This embodiment specifically describes a self-supervised deep learning sidelobe and clutter suppression method for ship targets in SAR images, comprising:

[0050] Step 1: Acquire SAR images containing ship targets;

[0051] Step 2: Process the SAR image using tone mapping technology;

[0052] Step 3: Use a clustering algorithm to extract the target rectangle contour from the image processed by tone mapping technology;

[0053] Step 4: Use a weighted function to filter and cluster the extracted target rectangle contours to obtain the contours of the ship targets, and map the contours of the ship targets onto the SAR image from Step 1 to obtain the mapped image.

[0054] Step 5: Using the SAR image and the mapped image from Step 1, obtain the loss function through deep learning, and update the supervised deep learning model through the loss function until the loss function converges, thus obtaining the image after denoising and sidelobe suppression.

[0055] This application aims to enable a model to achieve high-quality denoising and sidelobe suppression of SAR images at a low cost without any training data or prior knowledge of noise distribution, through a self-supervised learning framework. The model automatically learns the ability to suppress sidelobes and clutter while preserving details of ship targets. This method leverages the inherent characteristics of SAR images and achieves effective suppression of sidelobes and clutter through a defined loss function and network structure.

[0056] To achieve the above objectives, the technical solution adopted in this application includes the following steps:

[0057] 1. Acquire SAR images containing ship targets;

[0058] Step S1: Extract map data and image data containing ship targets from existing open-source data, and perform preprocessing on the map data and image data, including image registration and filtering. Then, extract a 500×500 pixel SAR image containing ship targets from the preprocessed image as the input to the network.

[0059] 2. Preprocessing SAR images using tone mapping techniques to adapt to their high dynamic range characteristics, including:

[0060] Logarithmic tone mapping is used to preprocess SAR images to effectively adapt to their high dynamic range characteristics, while improving image contrast and detail visibility.

[0061] Step S2 involves processing the SAR image slices using an adaptive logarithmic tone mapping method. The brightness values ​​of the image are adjusted to effectively adapt to its high dynamic range characteristics, while simultaneously improving image contrast and detail visibility. Specifically:

[0062] Step S21, calculate the logarithmic mean of the SAR image:

[0063]

[0064] Where: Y(x) represents the brightness channel of the image converted to XYZ space; W is the total number of pixels in the image; δ is a small positive number, which is 10 in this example. -4 ;

[0065] Step S22: Use the average value of the logarithm of the image brightness value Y as the key value.

[0066]

[0067] Among them, Y max Y represents the maximum brightness value. min This is the minimum brightness value.

[0068] Step S23: Divide the SAR image into partitions based on the key value as a threshold.

[0069] L T =L max -(0.9+0.1key)(L max -L min )

[0070] L H =L min +[0.6+0.4×(1-key)](L max -L min )

[0071] Where: L T The threshold for low-light regions in a brightness image; L H The threshold for the high-illuminance region of the brightness image; L max and L min These represent the maximum and minimum values ​​of the image normalized to between 0 and 1, respectively. Image brightness is less than L. T At that time, it is a low-light area; greater than L H At that time, it is a high-illuminance area; between L T and L H The brightness range between these two points is the medium illumination range.

[0072] Step S24: Adaptive logarithmic method is used to achieve high dynamic range compression processing.

[0073]

[0074] Where: L i (x,y) represents the display brightness after tone mapping; L w (x,y) and Lw,max These represent the original brightness and the maximum value of the original brightness, respectively; q i ∈[1,∞],k i ∈[1,∞]; when i=T,W,H=1,2,j=1,S1=10,j=2,S2=2,S is the base of the logarithmic mapping;q andk are constants.

[0075] 3. Clustering algorithms are used to extract the target rectangle contour from the preprocessed image;

[0076] The image data is processed using the DBSCAN clustering algorithm (followed by binarization). Then, using the center point coordinates of the rectangle, the width and height of the rectangle, and the rotation angle of the rectangle, the minimum area rectangle that can enclose the contour is calculated to obtain the target area and background area in the SAR image.

[0077] Step S3: Use the K-means clustering algorithm to extract the target rectangle contour of the preprocessed image. Specifically, determine an initial cluster center for each cluster, calculate the objective function of the cluster center, assign the samples to the nearest center vector, and reduce the value of the objective function.

[0078]

[0079] In the formula, x i Let x represent the i-th sample, and j represent x. i clusters, p j Let represent the center point of the cluster, and n represent the total number of samples. Samples are assigned to the nearest cluster according to the minimum distance principle. The mean of the samples in each cluster is used as the new cluster center. This process of updating the cluster centers is repeated until convergence, resulting in K clusters. In this example, K is set to 2.

[0080] 4. A weighted function is used to perform precise filtering and clustering on the extracted contours to distinguish between the background region and the target region, thereby obtaining the contours of the ship target. Furthermore, the extracted contours of similar ship targets are mapped onto the original SAR image to obtain the mapped image.

[0081] Step S4 involves using a weighted function to perform precise filtering and clustering on the extracted contours to distinguish between background and target regions, obtaining contours similar to the ship target. These extracted contours are then mapped onto the original SAR image to obtain... Figure 2 The results shown require the following specific steps:

[0082] Step S41: For each contour obtained from clustering, apply a bilateral filtering algorithm for edge-preserving denoising to maintain the clarity of the target contour while removing speckle noise and subtle clutter in the image. The bilateral filtering algorithm is defined as follows:

[0083]

[0084] Where: d(i,j,k,l) ​​is the weighted combination of the values ​​of neighboring pixels; r(i,j,k,l) ​​is the value range template coefficient; where (k,l) is the center coordinate of the template window; (i,j) is the coordinate of the other coefficients of the template window; σ d σ is the standard deviation of the Gaussian function; the function f(x,y) represents the pixel value of the image at point (x,y); σ r is the standard deviation of the Gaussian function.

[0085] Step S42: Apply a second clustering process to the bilaterally filtered image. This clustering aims to further refine the separation between the target and the background, especially in the target edge region. This step allows for a more accurate distinction between the ship target and its surrounding sidelobes and clutter.

[0086] Step S43: Map the refined target contour back to the original SAR image. This step ensures accurate positioning of the target contour by comparing the contour positions in the original image and the preprocessed image, providing accurate target information for the suppression of sidelobes and clutter.

[0087] 5. Utilizing a self-supervised deep learning model, the network is updated by comparing the processing results with the error of the high dynamic range input image to optimize denoising and sidelobe suppression effects, including;

[0088] The original image is fed into a convolutional neural network, and after the aforementioned preprocessing steps, the network outputs a mapped image. This image is compared with the original SAR image, and the resulting error is used to guide the optimizer in updating the parameters of each layer of the network through a loss function. This process is repeated until the error between the network output and the original noise-free image is minimized. At this point, the network training is complete, and the final output is an image that has undergone denoising and sidelobe suppression.

[0089] Step S5: Utilize the UNet network as a self-supervised deep learning model to suppress sidelobes and clutter. The UNet model, with its powerful feature extraction capabilities and effective contextual information transfer mechanism, can effectively suppress sidelobes and clutter without losing target details.

[0090] Step S51: The preprocessed SAR image and the image mapped with the target contour are used as input to the UNet model. The model automatically adjusts its parameters through self-supervised learning to minimize the error between the output image and the high dynamic range input image.

[0091] Step S52: Use mean squared error (MSE) as the loss function to quantify the difference between the model output and the target image. Optimize the parameters of each layer of the model using the backpropagation algorithm to improve the suppression of sidelobes and clutter.

[0092] Step S53: Repeat steps S51 and S52 until the model's loss on the validation set no longer decreases significantly, indicating that the model has converged. At this point, the model can effectively denoise and suppress sidelobes in SAR images while preserving important details of ship targets. The results are as follows: Figure 3 As shown.

[0093] The self-supervised training model has an overall structure of UNet. The feature extraction part of UNet employs multiple convolutional and pooling layers to progressively extract high-level features from the image, while preserving low-level feature information through skip connections. In self-supervised training, the original image and the preprocessed image are used as inputs to the UNet network model, while the denoised and sidelobe-suppressed image is used as the model's output. The mean squared error (MSE) loss function is used to quantify the difference between the output image and the target image. The loss function value is obtained through forward propagation, and then the parameters in each convolutional layer of the model are optimized using the backpropagation algorithm. These steps are repeated until the loss function value no longer decreases, indicating that the model has converged. In this way, the UNet model can effectively perform denoising and sidelobe suppression of SAR images through self-supervised learning without external labeled data.

[0094] It should be noted that the specific embodiments are merely explanations and illustrations of the technical solution of the present invention and should not be used to limit the scope of protection. Any modifications made in accordance with the claims and specification of the present invention that are only partial should still fall within the protection scope of the present invention.

Claims

1. A self-supervised deep learning method for suppressing sidelobes and clutter in SAR images of ship targets, characterized in that... include: Step 1: Acquire SAR images containing ship targets; Step 2: Process the SAR image using tone mapping technology; Step 3: Use a clustering algorithm to extract the target rectangle contour from the image processed by tone mapping technology; Step 4: Use a weighted function to filter and cluster the extracted target rectangle contours to obtain the contours of the ship targets, and map the contours of the ship targets onto the SAR image from Step 1 to obtain the mapped image. Step 5: Using the SAR image and the mapped image from Step 1, obtain the loss function, and update the supervised deep learning model through the loss function until the loss function converges, thus obtaining the image after denoising and sidelobe suppression. The specific steps of step two are as follows: Step 21: Calculate the logarithmic mean of the SAR image containing the ship target. : Step 22: Obtain the key value using the brightness value and logarithmic mean of the pixels in the image; Steps two and three: Partition the SAR image using the key value. The specific steps are as follows: Using the key value to obtain the threshold of the low-light region of the brightness image Threshold for high-illuminance areas in brightness images and the pixel brightness in the SAR image is less than Yes, it is classified as a low-light area, and the pixel brightness is greater than Yes, it falls into the high-illuminance area, and the pixel brightness is between and Those areas between these levels are classified as medium illumination zones. Step 24: Compress the partitioned SAR image using an adaptive logarithmic method to obtain the display brightness after tone mapping; The and Represented as: in, and These represent the maximum and minimum values ​​of the image after normalization to a value between 0 and 1; The display brightness after tone mapping is expressed as: in, This indicates the display brightness after tone mapping. and These represent the original brightness and the maximum value of the original brightness, respectively. , It is a constant. , ; ; hour, , , , It represents the base of the logarithmic mapping.

2. The self-supervised deep learning sidelobe and clutter suppression method for ship targets in SAR images according to claim 1, characterized in that... The SAR image containing the ship target in step one is a preprocessed image. The preprocessing includes complex data conversion, multi-view processing, filtering, and geocoding.

3. The self-supervised deep learning sidelobe and clutter suppression method for ship targets in SAR images according to claim 2, characterized in that... The logarithmic mean of the SAR image containing the ship target Represented as: in, This indicates that the image has been converted to a luminance channel in the XYZ space. The total number of pixels in the image. Represents positive numbers. Pick , This represents the pixel position index in the image.

4. The self-supervised deep learning sidelobe and clutter suppression method for ship targets in SAR images according to claim 3, characterized in that... The key value is represented as: in, This represents the maximum brightness value of a pixel. This represents the minimum brightness value of a pixel.

5. The self-supervised deep learning sidelobe and clutter suppression method for ship targets in SAR images according to claim 1, characterized in that... The clustering algorithm in step three is the K-means clustering algorithm, specifically expressed as follows: in, Indicates the first The nth sample, that is, the nth image. 1 pixel express clusters, Indicates the center point corresponding to the cluster. This represents the total number of samples.

6. The self-supervised deep learning sidelobe and clutter suppression method for ship targets in SAR images according to claim 1, characterized in that... The specific steps of step four are as follows: Step 41: Process the extracted target rectangular contour using a bilateral filtering algorithm; Step 42: Perform clustering processing on the bilaterally filtered image to obtain the refined target contour; Step 43: Map the refined target contour back to the original SAR image to obtain the mapped image.

7. The self-supervised deep learning sidelobe and clutter suppression method for ship targets in SAR images according to claim 6, characterized in that... The bilateral filtering algorithm is expressed as follows: in, This represents the weight function in the bilateral filtering algorithm. This represents a weighted combination of the values ​​of neighboring pixels. Represents the range template coefficients. Indicates the center coordinates of the template window. The coordinates of other coefficients in the template window. The standard deviation of the Gaussian function is represented by... Indicates the image at point Pixel value at that location, Indicates the image at point Pixel value at that location, This represents the standard deviation of the Gaussian function.

Images