A radar target image mixed noise elimination method based on wavelet transform

Abstract

This invention relates to a method for eliminating mixed noise in radar target images based on wavelet transform, belonging to the field of radar target image noise reduction. A wavelet coefficient correlation denoising algorithm is used to perform a coarse filtering on the high-frequency information and the high-frequency components of low-frequency information in the horizontal, vertical, and diagonal directions of the R, G, and B channels of the image, eliminating most of the noise with large amplitudes. Simultaneously, an improved adaptive median filtering algorithm is designed to perform a secondary fine-grained filtering on the reconstructed R, G, and B channels. The proposed method can recover image information well under mixed noise conditions of varying intensities, exhibiting strong stability and enabling noise reduction of radar target images under different mixed noise conditions.

Description

Technical Field

[0001] This invention relates to the field of noise reduction in radar target images, and more specifically, to a method for eliminating hybrid noise in radar target images based on wavelet transform. Background Technology

[0002] Radar target images are primarily affected by speckle noise caused by the coherence of scattering phenomena and Gaussian noise caused by complex environments. Therefore, in complex electromagnetic environments, the effectiveness of radar imaging can directly impact the resolution of radar target identification, thus delaying commanders' ability to formulate effective strategies and potentially altering the course of the war. Consequently, effective radar target image processing algorithms are essential. A key challenge in radar target image processing is that noise, edges, and textures are all high-frequency components, making them difficult to distinguish during noise reduction. Furthermore, the denoised image inevitably loses some edge and detail information. This remains a significant problem in the field of image processing today.

[0003] Currently, radar image processing technology is mainly divided into spatial domain-based denoising methods and frequency domain-based denoising methods. Spatial domain-based filtering methods can eliminate noise to a certain extent, but may lead to image blurring and loss of image edge and detail information. Frequency domain-based image denoising algorithms include methods based on Fast Fourier Transform (FFT), Discrete Cosine Transform (DCT), and Wavelet Transform (WT). WT is developed based on FFT. Due to its ability to perform time-frequency analysis simultaneously and quickly detect data abrupt changes, wavelet transform has become an important research topic in image and signal processing. However, because wavelet transform lacks directionality, possessing only horizontal, vertical, and diagonal information, it cannot optimally represent two-dimensional images containing line or surface singularities. This causes distortion when processing two-dimensional images, preventing the best reconstruction of the original image.

[0004] In recent years, many researchers and scholars have studied the problem of image noise reduction based on spatial and frequency domains. For example, You et al. proposed an algorithm for Gaussian noise reduction of color images before edge detection based on frequency domain wavelet transform (hereinafter referred to as the algorithm M1 in this invention) (You N, Han L, Zhu D, et al. Research on imagedenoising in edge detection based on wavelet transform[J]. Applied Sciences,2023,13(3):1837.). This algorithm can improve the signal-to-noise ratio of the image and retain as much edge information as possible. The peak signal-to-noise ratio (PSNR) and root mean square error (MSE) of the denoised image can reach 23.48dB and 299.49, respectively. Deng et al. applied different wavelet thresholds at different wavelet scales to perform image denoising based on Donoho's VisuShrink image denoising method (referred to as M2 in this invention) in the spatial domain (Deng G, Liu ZA wavelet image denoising based on the new threshold function [C] / / 2015 11th International Conference on Computational Intelligence and Security (CIS). IEEE, 2015:158-161.). This algorithm can eliminate Gaussian noise in the image while preserving the edge information of the image relatively well. Faragallah OS et al. proposed an effective method for suppressing salt-and-pepper (S&P) noise in images based on the spatial domain within the framework of the Adaptive Switching Weighted Median Filter (ASWMF) (hereinafter referred to as the algorithm M3 in this invention) (Faragallah OS, Ibrahem H M. Adaptive switching weighted median filter framework for suppressing salt-and-pepper noise[J]. AEU-International Journal of Electronics and Communications, 2016, 70(8):1034-1040.).This algorithm examines noise candidates by using the local average value during the noise detection phase, and then uses an adaptive weighted median filter to replace the detected noise pixels with their weighted median values ​​within a set window size. Test results show that this method achieves superiority and efficiency of up to 90% in eliminating S&P noise.

[0005] Most existing image denoising methods are designed for single image noise conditions, but real-world environments are more complex, and in most cases, there is interference from mixed noise. In addition, these algorithms are designed for denoising grayscale images. Even for color images, the image needs to be converted to grayscale space before denoising, which greatly increases the computational and time overhead and is not conducive to applications with high real-time requirements. Summary of the Invention

[0006] The technical problem to be solved by this invention is:

[0007] To address the problem of test images being contaminated by mixed noise, which leads to a decrease in the resolution of radar target images, a method for eliminating mixed noise in radar target images based on wavelet transform is proposed. This method reduces the interference of mixed noise on radar target images and reconstructs the details, edges, and color information of the images.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0009] A method for eliminating hybrid noise in radar target images based on wavelet transform, characterized by comprising:

[0010] The radar target image affected by mixed noise is decomposed into three channels: R, G, and B. Wavelet transform is performed on each of the R, G, and B channels to extract high-frequency information in the horizontal, vertical, and diagonal directions, as well as the high-frequency part of the low-frequency information in each channel.

[0011] Wavelet coefficient correlation denoising is performed on the high-frequency information and the high-frequency part of the low-frequency information in the horizontal, vertical and diagonal directions of each channel to obtain new high-frequency information and low-frequency information of each channel in each direction.

[0012] The image's R, G, and B channels are reconstructed using the new high-frequency and low-frequency information of each channel in each direction, thus completing the first coarse filtering of the image.

[0013] An improved adaptive median filtering algorithm is used to perform secondary fine-grained filtering on the reconstructed R, G, and B channels;

[0014] The three new R, G, and B channels obtained from the secondary fine-grained filtering are integrated to reconstruct the radar target image.

[0015] A further technical solution of the present invention: Wavelet transform is performed on each of the R, G, and B channels to extract high-frequency information in the horizontal, vertical, and diagonal directions, as well as the high-frequency components of the low-frequency information in each channel. Specifically:

[0016]

[0017]

[0018]

[0019]

[0020] Where j is the wavelet decomposition level, initially set to 0; {h k} represents a low-pass filter, {g k} is a high-pass filter, cA j+1 For low-frequency coefficients, cH j+1 For horizontal high-frequency coefficients, cV j+1 For vertical high-frequency coefficients, cD j+1 For diagonal high-frequency coefficients, cA j (k,l) represents the original radar target image, m and n represent the number of rows and columns of the matrix, respectively, and k and l have no special meaning, they just represent the coordinates of a point in the matrix.

[0021] A further technical solution of the present invention: wavelet coefficient correlation denoising is performed on the high-frequency information and the high-frequency component of the low-frequency information in the horizontal, vertical, and diagonal directions of each channel, using the following formula:

[0022] CW j,k =W j,k W j+1,k

[0023]

[0024] Among them, W j,k These are high-frequency wavelet coefficients, CW j,k W is the wavelet correlation coefficient at the k-th point on scale j. j,k These are the normalized wavelet coefficients, PW j PCW represents the energy of wavelet coefficients at scale j. j The energy of the correlation coefficient at scale j.

[0025] A further technical solution of the present invention: The image's R, G, and B channels are reconstructed using the new high-frequency and low-frequency information of each channel in each direction, specifically as follows:

[0026] cA j (m,n)=A j+1 +Hj+1 +V j+1 +D j+1

[0027] Among them, A j+1 H j+1 V j+1 and D j+1 It is a sub-image of the j-scale image reconstructed from low-frequency coefficients and high-frequency coefficients in three directions, cA j (m,n) represents the reconstructed image.

[0028] A further technical solution of the present invention: A modified adaptive median filtering algorithm is used to perform secondary fine-grained filtering on the reconstructed R, G, and B channels, specifically:

[0029] Expand the original image pixel matrix:

[0030] The original image matrix is ​​expanded in four directions (top, right, bottom, left) with expansion ranges of N, N+1, N+1, and N respectively. The dimension of the expanded image matrix will reach X1 = (m+2*N+1, n+2*N+1), where N is the expansion radius.

[0031] Determine the filter window size:

[0032] Starting from the smallest filtering radius r=1, calculate the neighborhood I of each pixel in the original image along the matrix row direction, using the following formula:

[0033] I = X1(ir:i+r,jr:j+r)

[0034] In the formula, i and j represent the coordinates of the pixel.

[0035] Reorder the pixels within the neighborhood I to obtain I', and arrange the pixels in each column in ascending order. Find the maximum value I' within the neighborhood. max Minimum value I' min The center pixel in the neighborhood is taken as the median I' med If the value of the center pixel is 0, then set I' med =1, determine I' med Does it belong to (I' min ,I' max If the above conditions are met, the radius of this neighborhood is considered to meet the actual filtering requirements, and the size of the neighborhood is used as the size of the filtering window; otherwise, the filtering radius is further expanded to r = r + 1, r ≤ N + 1, and the calculation is repeated until a suitable filtering window size is found.

[0036] Noise determination: Each pixel in the image is determined to be a real pixel or noise based on the filtering window.

[0037] A further technical solution of the present invention: the step of determining whether each pixel of the image is a real pixel or noise based on the filtering window specifically involves:

[0038] Determine if a pixel in the original image lies between the maximum and minimum points of the filtering window. If this condition is met, the pixel is considered a true pixel and remains unchanged. If the condition is not met, the pixel has been noise-controlled, and the median point I' within the filtering window is used. med Replace it.

[0039] A computer system is characterized by comprising: one or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method described above.

[0040] A computer-readable storage medium is characterized by storing computer-executable instructions, which, when executed, are used to implement the above-described method.

[0041] The beneficial effects of this invention are as follows:

[0042] This invention provides a wavelet transform-based method for eliminating mixed noise in radar target images (referred to as algorithm M4 in this invention). It overcomes some shortcomings of spatial and frequency domain image denoising algorithms. This invention combines a spatial domain-based median filtering algorithm and a frequency domain-based wavelet coefficient correlation algorithm. Through analysis of the characteristics of mixed noise in radar target images, it can eliminate mixed noise with high precision while solving problems such as blurred image edges, blurred details, and color distortion. It achieves excellent radar target image restoration.

[0043] Compared with the prior art, the present invention has the following advantages:

[0044] First, it exhibits strong robustness. Targeting the characteristics of mixed noise in radar target images, it uses a wavelet transform coefficient correlation algorithm to mine deep features of mixed noise data, greatly reducing noise interference with the image. It can still recover the essential features of the image well under different mixed noise intensities, and completely preserve the image's edges, details, and color information.

[0045] Second, it has high reliability. Combining wavelet transform, an improved adaptive median filtering algorithm is proposed. Based on the image data structure, it adaptively selects an appropriate filtering window size for each pixel, enhances the recognition of noisy pixels, eliminates noise points to the maximum extent, and retains real pixels.

[0046] Third, it exhibits wide transferability. Wavelet transform can select different wavelet functions based on different noise characteristics, and the proposed improved adaptive filtering algorithm can change the filtering radius to expand the filtering range according to the characteristics of the image itself and the noise. Therefore, this method can adapt to different image and noise types and has good transferability.

[0047] Fourth, it is easy to implement. The algorithm architecture is relatively clear. It is an improvement and combination based on wavelet transform and median filtering, which is easy to understand and has good feasibility. Attached Figure Description

[0048] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0049] Figure 1 This is a flowchart illustrating the implementation steps of the present invention;

[0050] Figure 2 This is an expanded pixel matrix map of a radar target image.

[0051] Figure 3 A schematic diagram showing how to set the filter window size;

[0052] Figure 4 The denoising effects of different methods when the mixed noise intensity is σ1=0.01 and d1=10% are shown in the following figures: (a) Original image; (b) Noisy image; (c) Denoising effect of M1; (d) Denoising effect of M2; (e) Denoising effect of M3; (f) Denoising effect of M4.

[0053] Figure 5 The denoising effects of different methods at mixed noise intensity of σ2=0.02 and d2=20% are shown in the following figures: (a) Original image; (b) Noisy image; (c) Denoising effect of M1; (d) Denoising effect of M2; (e) Denoising effect of M3; (f) Denoising effect of M4.

[0054] Figure 6 The denoising effects of different methods at mixed noise intensity σ3=0.03 ​​and d3=30% are shown in the following figures: (a) Original image; (b) Noisy image; (c) Denoising effect of M1; (d) Denoising effect of M2; (e) Denoising effect of M3; (f) Denoising effect of M4.

[0055] Figure 7 Curves showing the changes in PSNR and MSE values ​​of radar target images with noise intensity after denoising using different methods: (a) PSNR; (b) MSE. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0057] This invention provides a method for eliminating hybrid noise in radar target images based on wavelet transform, such as... Figure 1 As shown, it includes the following steps:

[0058] Step 1: Based on the characteristics of color images, the radar target image affected by mixed noise is decomposed into three channels: R, G, and B. Wavelet transform is performed on each channel to extract the high-frequency information in the horizontal, vertical, and diagonal directions of each channel, as well as the high-frequency part of the low-frequency information.

[0059] Step 2: Perform wavelet coefficient correlation denoising on the high-frequency components of the low-frequency information in the vertical, horizontal, and diagonal channels, as well as the high-frequency information extracted from the horizontal, vertical, and diagonal channels, to obtain new high-frequency and low-frequency information in each direction for each channel.

[0060] Step 3: Reconstruct the R, G, and B channels of the image using the new high-frequency and low-frequency information, completing the first coarse filtering of the image. Through this coarse filtering, the system can effectively eliminate most of the noise with large amplitude values ​​in the high and low frequency information of the radar target image;

[0061] Step 4: Based on the characteristics of mixed noise, an improved adaptive median filtering algorithm is proposed to perform secondary fine-grained filtering on the reconstructed R, G, and B channels. This enables deeper noise elimination for each channel of the image and serves to calibrate the pixels of each channel after reconstruction, thus better preserving the edge, details, and color information of the image.

[0062] Step 5: Integrate the noise-reduced R, G, and B channels to reconstruct the radar target image.

[0063] The specific steps described above are as follows:

[0064] Step 1: Perform two-dimensional decomposition of the radar noise image using wavelet transform, dividing it into three channels: R, G, and B. Then, perform multi-directional wavelet transform on each channel to extract high-frequency information. Assuming the radar target image is represented as f(x,y), the two-dimensional discrete wavelet transform of each channel can be calculated using the following formula:

[0065] cA0=f(m,n)

[0066]

[0067]

[0068]

[0069]

[0070] In the formula, j is the wavelet decomposition level, {h k} and {g k} represent low-pass and high-pass filters, respectively. Image cA at scale j. j It is decomposed into: low-frequency coefficients cA j+1 Horizontal high frequency coefficient cH j+1 Vertical high frequency coefficient cV j+1 and diagonal high-frequency coefficients cD j+1 The system then decomposes the low-frequency information again, extracting the high-frequency components. Finally, all the extracted high-frequency information is processed accordingly.

[0071] Step 2: A wavelet coefficient correlation denoising algorithm is used to perform a first coarse filtering on all extracted high-frequency information. This utilizes the strong correlation between wavelet coefficients in different layers of the image, a characteristic that noise lacks. The essence of the wavelet coefficient correlation denoising algorithm is to compare the normalized correlation coefficients at each position in each layer, and determine whether each data point is a real pixel or a noise point based on the correlation between the coefficients. The wavelet coefficient correlation algorithm is used to denoise the high-frequency information extracted in three directions for each channel, and also to denoise the high-frequency components of the low-frequency information. This approach not only better eliminates residual large-amplitude noise, but also ensures that the system does not damage the data structure of other parts when processing specific parts of the data, thus better preserving the characteristics of each part of the data.

[0072] The specific implementation process is as follows:

[0073] CW j,k =W j,k W j+1,k

[0074]

[0075] In the formula, W j,k These are high-frequency wavelet coefficients, CW j,k W is the wavelet correlation coefficient at the k-th point on scale j. j,k These are the normalized wavelet coefficients, PW j PCW represents the energy of the wavelet coefficients at scale j. j This represents the correlation coefficient energy at scale j.

[0076]

[0077]

[0078] Based on the characteristics of wavelet functions and the number of wavelet decomposition levels M, the high-frequency wavelet coefficients W of each level are first calculated. j,k And with the normalization coefficient W j,k Compare, if W j,k ≥W j,k The system then considers that point to be a real pixel, and sets W... j,k =W j,k And set W j,k =0; if W j,k <W j,k The system then considers the pixel to be controlled by noise and retains W. j,k W j,k =0. Then recalculate W at each scale. j,k The actual pixels are preserved in W. j,k In the middle, the noise points are preserved in W. j,k .

[0079] Step 3: Using the new high-frequency information obtained from each channel after noise reduction, reconstruct the low-frequency information and the R, G, and B channels of the image, completing the first coarse filtering of the image. The image reconstruction calculation is shown in the following formula:

[0080]

[0081] Where A j+1 H j+1 V j+1 and D j+1 It is a sub-image of the j-scale image reconstructed from low-frequency coefficients and high-frequency coefficients in three directions. Through coarse filtering, the system can effectively eliminate most of the noise in the high and low frequency information of the radar target image.

[0082] Step 4: The proposed Improved Adaptive Median Filtering (LAMF) algorithm is used to perform secondary fine-grained filtering on the reconstructed R, G, and B channels. This further reduces noise and calibrates the reconstructed image, better preserving edge and detail information. The LAMF algorithm mainly consists of the following three steps:

[0083] 1) Expand the original image pixel matrix

[0084] To ensure the filtering window effectively performs filtering, meaning that the data within the filtering window is not empty and is valid, the system first needs to set an image matrix expansion radius N for the algorithm. To ensure the algorithm's timeliness and reduce the system's computational overhead, after testing and experimentation, the system's overall performance is optimal when the expansion radius does not exceed 10. Therefore, the pixels of the original image need to be expanded. Assuming the original image is a matrix X = m × n, the system then expands the original image matrix in four directions—top, right, bottom, and left—with expansion increments of N, N+1, N+1, and N respectively. The expanded image matrix will then have dimensions X1 = (m + 2 * N + 1, n + 2 * N + 1), where m and n represent the number of rows and columns of the matrix, respectively. For example, assuming N = 2 and m = n = 5, after four expansions, the pixel matrix becomes X1 = 10 × 10. The specific expansion process is as follows... Figure 2 As shown.

[0085] The system calculates the filter window size. After the matrix is ​​expanded, it finds the position of the original image pixel matrix in the expanded matrix and selects a suitable filter window for each pixel in the matrix. For example, the algorithm's calculation process is described in detail using pixel (7,7) in the expanded matrix as an example. The window selection process is as follows: Figure 3 As shown. The system first calculates the neighborhood I of each pixel in the original image along the matrix rows, starting from the smallest filtering radius r=1, using the following formula:

[0086] I = X1(ir:i+r,jr:j+r)

[0087] In the formula, i and j represent the coordinates of the pixel. Then, the pixels in the neighborhood I are reordered to obtain I', with each column of pixels arranged in ascending order. Finally, the maximum value I' in the neighborhood is found. max Minimum value I' min The center pixel in the neighborhood is taken as the median I' med If the value of the center pixel is 0, then set I'. med =1, finally judge I' med Does it belong to (I' min ,I' max If the above conditions are met, the radius of this neighborhood is considered to satisfy the actual filtering requirements. The system then uses the size of the neighborhood as the size of the filtering window. Otherwise, the system continues to expand the filtering radius r = r + 1 (r ≤ N + 1) and recalculates until a suitable filtering window size is found. Through the above calculations, the system can set a suitable filtering window for each pixel.

[0088] 2) Noise determination

[0089] After determining the filter window size, the system needs to determine whether each pixel in the image is a real pixel or noise. The specific process is as follows: First, it is determined whether the pixel in the original image is between the maximum and minimum points of the filter window. If this condition is met, the pixel is considered a real pixel and remains unchanged. If this condition is not met, it means that the pixel has been controlled by noise, and the median point I' within the filter window is used. med The replacement is then performed. This completes the image filtering process.

[0090] Step 5: Integrate the noise-reduced R, G, and B channels to reconstruct the radar target image.

[0091] The effectiveness of this invention is further illustrated by the following experiments using measured data:

[0092] 1. Experimental Scenario:

[0093] The experimental platform consisted of a 64-bit Windows 10 system with a 2.8GHz CPU and 8GB of memory. The original image was a radar target search image with a resolution of 639×632 (width×height) and was a color radar target image. Mixed noise containing Gaussian white noise and multiplicative speckle noise was added to the image. The noise intensities were as follows: the variances of the Gaussian white noise were σ1 = 0.01, σ2 = 0.02, and σ3 = 0.03; the noise densities of the multiplicative speckle noise were d1 = 10%, d2 = 20%, and d3 = 30%. To verify the algorithm's ability to eliminate mixed noise in radar target images, different intensities of mixed noise were added to the radar target search image for testing. The PSNR and MSE values ​​after denoising by different algorithms were calculated and compared.

[0094] Furthermore, the wavelet function, as a crucial parameter in this invention, directly impacts system performance. Since the wavelet coefficient correlation algorithm primarily utilizes the correlation between wavelet coefficients in the image to eliminate noise, a suitable wavelet function is essential for optimal algorithm performance. In the experiments, different wavelet functions were used in the wavelet correlation denoising algorithm. Gaussian white noise with an intensity of σ² = 0.02 was then added to the radar target image, followed by processing using the algorithm proposed in this invention. The correlation coefficient between the high-frequency wavelet coefficients of the denoised image and the high-frequency wavelet coefficients of the original image was calculated; a higher correlation coefficient indicates better denoising performance.

[0095] 2. Experiment Content:

[0096] 2.1) Different wavelet functions were added to the algorithm proposed in this invention, and denoising tests were performed on radar target images respectively. The correlation coefficient results are shown in Table 1.

[0097] 2.2) The proposed algorithm was tested under different mixed noise intensities, and the noise reduction effects of each algorithm were compared as follows: Figure 4 , Figure 5 , Figure 6 As shown in Table 2, the comparison results of the trends of PSNR and MSE values ​​of various algorithms in terms of noise reduction performance evaluation with noise intensity are shown in Table 2 and 3. Figure 7 As shown.

[0098] Table 1 Comparison of results using different wavelet functions under Gaussian white noise.

[0099]

[0100] Table 2 Comparison of PSNR and MSE results of different algorithms after image processing at different noise intensities.

[0101]

[0102] Experimental results show that, in terms of wavelet function selection, the 'Haar' wavelet function yields the highest high-frequency correlation coefficient among the three wavelet functions. This demonstrates that the algorithm proposed in this invention achieves the best image denoising effect when using the 'Haar' wavelet function. Furthermore, the denoising effects of different algorithms on mixed noise reveal that as noise intensity increases, the performance of other algorithms deteriorates sharply, resulting in significant loss of image edge and detail information, and color distortion. The algorithm proposed in this invention, facing mixed noise of varying intensities, can preserve complete image edge, detail, and color information, improving target recognition capabilities under low signal-to-noise ratio conditions. The denoising performance differences between various signal-to-noise ratios are small, and it performs well under mixed noise conditions of varying intensities. Therefore, the wavelet transform-based radar target image mixed noise elimination method of this invention can effectively reduce noise in radar target images under a wide range of mixed noise conditions.

[0103] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in the present invention, and such modifications or substitutions should all be covered within the scope of protection of the present invention.

Claims

1. A method for eliminating hybrid noise in radar target images based on wavelet transform, characterized in that, include: The radar target image affected by mixed noise is decomposed into three channels: R, G, and B. Wavelet transform is performed on each of the R, G, and B channels to extract high-frequency information in the horizontal, vertical, and diagonal directions, as well as the high-frequency part of the low-frequency information in each channel. Wavelet coefficient correlation denoising is performed on the high-frequency components of the high-frequency and low-frequency information in the horizontal, vertical, and diagonal directions of each channel to obtain new high-frequency and low-frequency information for each channel in each direction; the formula used is: in, These are high-frequency wavelet coefficients. yes Scale Wavelet correlation coefficient of points These are normalized wavelet coefficients. represent Wavelet coefficient energy at scale represent Correlation coefficient energy of scale The image's R, G, and B channels are reconstructed using the new high-frequency and low-frequency information of each channel in each direction, thus completing the first coarse filtering of the image. An improved adaptive median filtering algorithm is used to perform secondary fine-grained filtering on the reconstructed R, G, and B channels; specifically: Expand the original image pixel matrix: respectively , , , The expansion is performed by extending the original image matrix in four directions: top, right, bottom, and left. The dimension of the expanded image matrix will reach [missing value]. ,in, To expand the radius; Determine the filter window size: From the smallest filter radius Begin by calculating the neighborhood of each pixel in the original image along the row direction of the matrix. Calculate using the following formula: In the formula, and These represent the coordinates of the pixels; For the neighboring area The pixels within are reordered to obtain Each column of pixels is sorted in ascending order, and the maximum value within its neighborhood is found. Minimum value The center pixel in the neighborhood is used as the median. If the value of the center pixel is 0, then set ,judge Does it belong to If the above conditions are met, the radius of this neighborhood is considered to meet the actual filtering requirements, and the size of the neighborhood is used as the size of the filtering window; otherwise, the filtering radius is further expanded. Recalculate until a suitable filter window size is found; Noise determination: Determine whether each pixel in the image is a real pixel or noise based on the filtering window; The three new R, G, and B channels obtained from the secondary fine-grained filtering are integrated to reconstruct the radar target image.

2. The method for eliminating hybrid noise in radar target images based on wavelet transform according to claim 1, characterized in that, Wavelet transform is performed on each of the R, G, and B channels to extract high-frequency information in the horizontal, vertical, and diagonal directions, as well as the high-frequency components of the low-frequency information in each channel. Specifically: in, This represents the wavelet decomposition level, initially set to 0. It is a low-pass filter. For high-pass filters, Low-frequency coefficients For horizontal high-frequency coefficients, For vertical high frequency coefficients, For diagonal high-frequency coefficients, The original radar target image, and These represent the number of rows and columns of the matrix, respectively. , It has no special meaning; it simply represents the coordinates of a point in the matrix.

3. The method for eliminating hybrid noise in radar target images based on wavelet transform according to claim 1, characterized in that, The image's R, G, and B channels are reconstructed using the new high-frequency and low-frequency information of each channel in each direction, specifically as follows: in, , , and It is reconstructed from low-frequency coefficients and high-frequency coefficients in three directions. Sub-images of a scaled image This is the reconstructed image.

4. The method for eliminating hybrid noise in radar target images based on wavelet transform according to claim 1, characterized in that, The step of determining whether each pixel in the image is a real pixel or noise based on the filtering window is as follows: The algorithm determines whether a pixel in the original image lies between the maximum and minimum values ​​of the filtering window. If this condition is met, the pixel is considered a true pixel and remains unchanged. If the condition is not met, the pixel has already been noise-controlled, and the median value within the filtering window is used. Replace it.

5. A computer system, characterized in that... include: One or more processors, a computer-readable storage medium for storing one or more programs, wherein, when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the method of any one of claims 1-4.

6. A computer-readable storage medium, characterized in that... The device stores computer-executable instructions, which, when executed, are used to implement the method described in any one of claims 1-4.

Images