Distributed target image non-uniform structure processing method, system, equipment and medium for Mars ionosphere detection SAR (Synthetic Aperture Radar)

Abstract

The invention discloses a distributed target image non-uniform structure processing method, system and device for Mars ionosphere detection SAR and a medium, and the method comprises the steps: carrying out the preprocessing of an original distributed target SAR image, calculating a gray-level co-occurrence matrix according to the pixel value of the image, calculating the texture feature value of the center position of a window based on the gray-level co-occurrence matrix, and carrying out the calculation of the texture feature value of the center position of the window; traversing the preprocessed distributed target SAR image to generate a texture feature image; segmenting the texture feature image by adopting an Otsu threshold algorithm to obtain an initial mask; then generating an image mask by adopting a region growing algorithm to obtain a mask image; calculating the Hadamard product of the mask image matrix and the preprocessed distributed target SAR image matrix to obtain a homogenized distributed target SAR image; the system, the equipment and the medium are used for implementing the method. According to the method, the non-uniform region can be removed in a targeted manner, the phase of the uniform region can still be kept, and the method has the advantage of being high in non-uniform structure removal precision.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a distributed target image non-uniform structure processing method, system, equipment and medium for Mars ionosphere detection SAR. Background Art

[0002] Spaceborne synthetic aperture radar (SAR) has been widely used in deep space exploration. Researchers can detect the ionosphere through the effect of radar echoes on the ionosphere, which is one of the important means of detecting the ionosphere on Mars. Since the accuracy of radar echoes will affect the accuracy of ionosphere detection, it is necessary to perform radiation calibration on radar echoes. In view of the difficulty of deploying artificial calibrators on Mars, it is necessary to propose a radiation calibration method for distributed target scenes that is suitable for Mars-specific scenes and has strong applicability. The homogenized target image is very important for accurately obtaining the antenna radiation pattern and further radiation calibration processing using spaceborne synthetic aperture radars in flat and uniform areas. Therefore, for non-uniform areas in SAR images, it is necessary to specifically eliminate non-uniform structures.

[0003] At present, there are some studies on the processing methods of non-uniform areas in SAR radar images at home and abroad. Masanobu Shimada et al. (M. Shimada, A. Freeman, A technique for measurement of spaceborne SAR antenna patterns using distributed targets [J]. IEEE Transactions on Geoscience and Remote Sensing, 1995, 33 (1): 100-114) proposed a technology for measuring spaceborne SAR antenna patterns using distributed targets. The homogenization method based on the chi-square similarity test uses the chi-square similarity test to test the uniformity within a certain area size and remove the non-uniform part. However, if the area size is too small, the non-uniform part will not be removed, and if it is too large, the uniform part will be removed. Zhuang Li (Zhuang Li. Research on cross-calibration method of spaceborne SAR radiation based on natural interface targets [D]. Beijing University of Chemical Technology, 2022.) proposed a study on cross-calibration method of spaceborne SAR radiation based on natural interface targets. The filter-based homogenization method uses the pixels around the pixel point for mean filtering, median filtering and other filtering methods, and uses filtering processing to smooth the non-uniform parts to increase the uniformity of the SAR radar image. However, it does not essentially remove the non-uniform part, and the non-uniform part still exists in the SAR radar image. Summary of the invention

[0004] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a distributed target image non-uniform structure processing method, system, equipment and medium for Mars ionosphere detection SAR, which adopts the algorithm of grayscale co-occurrence matrix and texture eigenvalue. When there is a difference between the texture of the non-uniform structure and the texture of the uniform structure, the non-uniform structure can be removed, and the method has the characteristic of high precision in removing the non-uniform structure.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] A distributed target image non-uniform structure processing method for Mars ionosphere detection SAR includes the following steps:

[0007] Step 1, distributed target SAR image preprocessing: input the original distributed target SAR image, perform Frost filtering, suppress the coherent speckle of the distributed target SAR image, and obtain the distributed target SAR image after the coherent speckle suppression;

[0008] Step 2, texture feature image generation: according to the pixel values ​​of the preprocessed distributed target SAR image, the gray level co-occurrence matrix is ​​calculated, and based on the gray level co-occurrence matrix, the texture feature value at the center of the window is calculated, and the preprocessed distributed target SAR image is traversed to generate a texture feature image;

[0009] Step 3, mask generation: according to the grayscale histogram of the texture feature image, the Otsu threshold algorithm is used to segment the foreground and background of the texture feature image to obtain an initial mask; then the region growing algorithm is used to generate an image mask to obtain a mask image;

[0010] Step 4, mask processing: Calculate the Hadamard product of the mask image matrix and the pre-processed distributed target SAR image matrix to obtain a homogenized distributed target SAR image.

[0011] The process of speckle suppression in step 1 is as follows:

[0012] The Frost filtering algorithm is used to suppress the coherent speckle of the input original distributed target SAR image. That is, according to equations (1) and (2), the four neighborhoods are used to traverse each pixel value in the image in turn to obtain the pixel gray value I after filtering. i ' j , the expression is as follows:

[0013] I′ ij =∑ p∈η m p I ij (1)

[0014]

[0015] Where η represents the four neighborhoods, i.e., the four pixels above, below, left, and right adjacent to the central pixel, I ij Represents the original distributed target SAR image pixel value, m p represents the weight value of the central pixel of the four neighborhoods, K represents the coordination factor, C s represents the pixel variance in the four neighborhoods, d s,p Represents the Euclidean distance between a single pixel in the four neighborhoods and the central pixel.

[0016] The process of generating the texture feature image in step 2 includes:

[0017] Step 2.1, set the window size:

[0018] Assume the window size is L*L, usually L is much smaller than n and is an odd number, the expression is as follows:

[0019]

[0020] In the formula, w represents the window, and L represents the number of rows and columns of the window;

[0021] Step 2.2, mirror-extend the image edge of the distributed target SAR image after speckle suppression, specifically:

[0022] Assume that the distributed target SAR image matrix I after speckle suppression is:

[0023]

[0024] In the formula, S 11 represents the value at the (1, 1)th position in the distributed target SAR image data, S nm Represents the value at the (n, m)th position in the image data;

[0025] The distributed target SAR image data after speckle suppression is mirror-extended to the image edge:

[0026]

[0027] In the formula, S ij (1≤i≤n,1≤j≤m) represents the data of the distributed target SAR image data after the image edge is extended by mirror, and I1 represents the distributed target SAR image matrix after the image edge is extended by mirror;

[0028] Step 2.3, read S ij Data with a window size of L*L in (1≤i≤n,1≤j≤m):

[0029]

[0030] In the formula, It represents the value at the (1, 1) position in the data of the distributed target SAR image after the image edge is extended by the mirror. Represents the value at the (L, L)th position in the data of the distributed target SAR image after the image edge is extended by the mirror;

[0031] Step 2.4, call the gray-level co-occurrence matrix function, calculate the gray-level co-occurrence matrix value corresponding to the center pixel of window w1 when the pixel pair spacing is 1 and the pixel pair position is 0°, 45°, 90°, and 135°, and normalize it to obtain the normalized gray-level co-occurrence matrix P' as the gray-level co-occurrence matrix calculation value for texture feature extraction. The expression is as follows:

[0032]

[0033] Where P 0° Represents the gray level co-occurrence matrix when the pixel pair position is 0°, P 45° Represents the gray level co-occurrence matrix when the pixel pair position is 45°, P 90° Represents the gray level co-occurrence matrix when the pixel pair position is 90°, P 135° Represents the gray level co-occurrence matrix when the pixel pair position is 135°, P′ 0° Represents the normalized gray-level co-occurrence matrix of the pixel pair at an angle of 0°, P′ 45° Represents the normalized gray-level co-occurrence matrix of the pixel pair at a 45° angle, P′ 90° Represents the normalized gray-level co-occurrence matrix of the pixel pair at a 90° angle, P′ 135° Represents the normalized gray-level co-occurrence matrix of pixel pairs at an angle of 135°;

[0034] Step 2.5, use the normalized gray-level co-occurrence matrix P' to calculate the angular second-order moment of the center pixel of the window as the eigenvalue of the center pixel of the L*L window. The expression is as follows:

[0035]

[0036] Where E represents the angular second-order moment value of the center pixel of the window;

[0037] Step 2.6, traverse the preprocessed distributed target SAR image and calculate the eigenvalue of each pixel:

[0038] Slide the L*L window, and traverse each pixel in the distributed target SAR image after preprocessing with the center pixel of the window. Calculate the eigenvalue of each pixel to obtain the texture feature image after edge mirror expansion. The expression is as follows:

[0039]

[0040] Where I2 represents the texture feature image matrix after edge mirror expansion;

[0041] Step 2.7, crop the texture feature image after edge mirror expansion to obtain the texture feature image of the distributed target SAR image after coherent speckle suppression, which is expressed as follows:

[0042]

[0043] Where I3 represents the texture feature image matrix of the distributed target SAR image after speckle suppression.

[0044] The process of generating the image mask in step 3 includes:

[0045] Step 3.1, segment the grayscale histogram of the texture feature image in step 2.7 using the Otsu threshold algorithm:

[0046] According to formula (11), the image grayscale value is traversed so that the inter-class variance σ of the foreground and background 2 Maximum, the image gray value at this time is used as the image threshold t for segmentation, the expression is as follows;

[0047] σ 2 =w0(t)w1(t)[μ0(t)-μ1(t)] 2 (11)

[0048] Wherein, w0 represents the ratio of the number of foreground pixels to the total number of pixels after threshold segmentation, w1 represents the ratio of the number of background pixels to the total number of pixels after threshold segmentation, μ0 represents the average value of foreground pixels after threshold segmentation, and μ1 represents the average value of background pixels after threshold segmentation;

[0049] Step 3.2, performing region growing on the segmented texture feature image, selecting points with zero pixel values ​​in different regions of the segmented texture feature image as initial seed points I4(i,j);

[0050] Step 3.3, compare the growth conditions of the initial seed point I4(i,j) with its eight adjacent points. If they meet the requirements, save the point to the growth region matrix. The growth conditions are as follows:

[0051] I4(i+u,j+v)-I4(i,j)≤0.15 (12)

[0052] In the formula, I4(i,j) represents the pixel value of the seed point, and I4(i+u,j+v) represents the eight points adjacent to the seed point;

[0053] Step 3.4, repeat step 3.3 for the remaining initial seed points until all initial seeds are traversed to obtain the image mask.

[0054] The process of mask processing in step 4 is as follows:

[0055] Assume that the size of the distributed target SAR image matrix after preprocessing is m*n. After steps 2 and 3, the size of the image mask matrix is ​​also m*n. According to formula (13), the Hadamard product of the distributed target SAR image matrix after preprocessing and the image mask matrix is ​​calculated to obtain the uniform distributed target SAR image S', which is expressed as follows:

[0056] S'=(I0⊙H) m*n (13)

[0057] Where I represents the original distributed target SAR image matrix, and H represents the image mask matrix.

[0058] The present invention also provides a distributed target image non-uniform structure processing system for Mars ionosphere detection SAR, comprising:

[0059] Image preprocessing module: used to suppress the coherent speckle of the original distributed target SAR radar image to obtain the preprocessed distributed target SAR image;

[0060] Image generation module: According to the pixel values ​​of the preprocessed distributed target SAR image, the gray level co-occurrence matrix is ​​calculated. Based on the gray level co-occurrence matrix, the texture feature value at the center of the window is calculated, and the preprocessed distributed target SAR image is traversed to generate a texture feature image.

[0061] Mask generation module: used to segment the texture feature image using the Otsu threshold algorithm, and then generate the image mask using the region growing algorithm to obtain the mask image;

[0062] Mask processing module: used to calculate the Hadamard product between the mask image matrix and the preprocessed distributed target SAR image matrix to obtain a homogenized distributed target SAR image.

[0063] The present invention also provides a distributed target image non-uniform structure processing device for Mars ionosphere detection SAR, comprising:

[0064] Memory: a computer-readable device used to store the computer program of the above-mentioned distributed target image non-uniform structure processing method for Mars ionosphere detection SAR;

[0065] Processor: used to implement the above-mentioned distributed target image non-uniform structure processing method for Mars ionosphere detection SAR when executing the computer program.

[0066] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it can implement the above-mentioned distributed target image non-uniform structure processing method for Mars ionosphere detection SAR.

[0067] Compared with the prior art, the beneficial effects of the present invention are:

[0068] 1. The present invention adopts the amplitude value of the radar signal and only operates on the intensity of the distributed target SAR image, and does not involve the processing of the radar phase value, and has the characteristic of maintaining the radar data phase of a uniform scene.

[0069] 2. The present invention only uses the amplitude value of the radar signal and is independent of the specific characteristics of the target. It has the characteristics of strong adaptability and is suitable for SAR image homogenization in most scenes.

[0070] 3. Compared with the traditional method that requires the deployment of manual calibrators at the calibration site to obtain calibration data to assist in subsequent processing, the present invention only uses distributed target SAR image data and does not require external manual calibrator equipment data supplement or support, making the method simpler and easier to implement in actual engineering applications.

[0071] 4. The present invention adopts the algorithm of gray-level co-occurrence matrix and texture eigenvalue, and when there is a difference between the texture of the non-uniform structure and the texture of the uniform structure, the non-uniform structure can be removed, and has the characteristic of high precision in removing the non-uniform structure.

[0072] 5. The four-step process of the present invention uses only matrix operations, which has the characteristics of high computational efficiency and does not rely on equation solving, but can be achieved only by matrix processing calculations.

[0073] 6. The present invention adopts the Otsu threshold algorithm. The image will produce some noise points. The noise points are eliminated through the region growing algorithm, further improving the quality of the mask image.

[0074] In summary, compared with the prior art, the present invention adopts the algorithm of gray-level co-occurrence matrix and texture eigenvalue, and when there is a difference between the texture of the non-uniform structure and the texture of the uniform structure, the non-uniform structure can be removed, and the non-uniform structure removal accuracy is high; by adopting the Otsu threshold algorithm and the regional growing algorithm, the noise points are eliminated, and the quality of the mask image is further improved; in addition, the present invention only operates on the intensity of the distributed target SAR image, and has the characteristics of maintaining the radar data phase of the uniform scene and strong adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 It is a flow chart of the method of the present invention.

[0076] Figure 2is the original distributed target SAR image used in the present invention.

[0077] Figure 3 It is a distributed target SAR image after speckle suppression generated in the present invention.

[0078] Figure 4 It is the texture feature value image generated in the present invention.

[0079] Figure 5 It is the initial mask generated by the Otsu threshold algorithm segmentation in the present invention.

[0080] Figure 6 It is the mask image after region growing generated in the present invention.

[0081] Figure 7 It is the homogenized distributed target SAR image in the present invention.

[0082] Figure 8 It is the original distributed target SAR image phase map in the present invention.

[0083] Fig. 9 It is the phase diagram of the distributed target SAR image after homogenization in the present invention. DETAILED DESCRIPTION

[0084] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0085] See also Figure 1 , a distributed target image non-uniform structure processing method for Mars ionosphere detection SAR, comprising the following steps:

[0086] Step 1, distributed target SAR image preprocessing: Input the original distributed target SAR image ( Figure 2 ), perform Frost filtering to suppress the coherent speckle of the distributed target SAR image, and obtain the distributed target SAR image after the coherent speckle suppression, as shown in Figure 3 As shown, specifically:

[0087] The Frost filtering algorithm is used to suppress the coherent speckle of the input original distributed target SAR image. That is, according to equations (1) and (2), the four neighborhoods are used to traverse each pixel value in the image in turn to obtain the pixel gray value I′ after filtering. ij , the expression is as follows:

[0088] I′ ij =Σ p∈η m p I ij (1)

[0089]

[0090] Where η represents the four neighborhoods, i.e., the four pixels above, below, left, and right adjacent to the central pixel, I ij Represents the pixel value of the original distributed target SAR image, m p represents the weight value of the central pixel of the four neighborhoods, K represents the coordination factor, C s represents the pixel variance in the four neighborhoods, d s,p Represents the Euclidean distance between a single pixel in the four neighborhoods and the central pixel;

[0091] Step 2, texture feature image generation: calculate the gray level co-occurrence matrix of the preprocessed distributed target SAR image, calculate the texture eigenvalue, and generate the texture feature image, such as Figure 4 As shown, specifically:

[0092] Step 2.1, set the window size:

[0093] Assume the window size is L*L, usually L is much smaller than n and is an odd number, the expression is as follows:

[0094]

[0095] In the formula, w represents the window, and L represents the number of rows and columns of the window;

[0096] Step 2.2, mirror-extend the image edge of the distributed target SAR image after speckle suppression, specifically:

[0097] Assume that the distributed target SAR image data after speckle suppression is:

[0098]

[0099] In the formula, S 11 represents the value at the (1, 1)th position in the distributed target SAR image data, S nm represents the value at the (n, m)th position in the image data, and I represents the distributed target SAR image matrix after speckle suppression;

[0100] The distributed target SAR image data after speckle suppression is mirror-extended to the image edge:

[0101]

[0102] In the formula, S ij (1≤i≤n,1≤j≤m) represents the data of the distributed target SAR image data after the image edge is extended by mirror, and I1 represents the distributed target SAR image matrix after the image edge is extended by mirror;

[0103] Step 2.3, read S ijData with a window size of L*L in (1≤i≤n,1≤j≤m):

[0104]

[0105] In the formula, It represents the value at the (1, 1) position in the data of the distributed target SAR image after the image edge is extended by the mirror. Represents the value at the (L, L)th position in the data of the distributed target SAR image after the image edge is extended by the mirror;

[0106] Step 2.4, call the gray-level co-occurrence matrix function, calculate the gray-level co-occurrence matrix value corresponding to the center pixel of window w1 when the pixel pair spacing is 1 and the pixel pair position is 0°, 45°, 90°, and 135°, and normalize it to obtain the normalized gray-level co-occurrence matrix P' as the gray-level co-occurrence matrix calculation value for texture feature extraction. The expression is as follows:

[0107]

[0108] Where P 0° Represents the gray level co-occurrence matrix when the pixel pair position is 0°, P 45° Represents the gray level co-occurrence matrix when the pixel pair position is 45°, P 90° Represents the gray level co-occurrence matrix when the pixel pair position is 90°, P 135° Represents the gray level co-occurrence matrix when the pixel pair position is 135°, P′ 0° Represents the normalized gray-level co-occurrence matrix of the pixel pair at an angle of 0°, P4' 5° Represents the normalized gray-level co-occurrence matrix of the pixel pair at a 45° angle, P′ 90° Represents the normalized gray-level co-occurrence matrix of the pixel pair at a 90° angle, P′ 135° Represents the normalized gray-level co-occurrence matrix of pixel pairs at an angle of 135°;

[0109] Step 2.5, use the normalized gray-level co-occurrence matrix P' to calculate the angular second-order moment of the center pixel of the window as the eigenvalue of the center pixel of the L*L window. The expression is as follows:

[0110]

[0111] Where E represents the angular second-order moment value of the center pixel of the window;

[0112] Step 2.6, traverse the preprocessed distributed target SAR image and calculate the eigenvalue of each pixel:

[0113] Slide the L*L window, and traverse each pixel in the distributed target SAR image after preprocessing with the center pixel of the window. Calculate the eigenvalue of each pixel to obtain the texture feature image after edge mirror expansion. The expression is as follows:

[0114]

[0115] Where I2 represents the texture feature image matrix after edge mirror expansion;

[0116] Step 2.7, crop the texture feature image after edge mirror expansion to obtain the texture feature image of the distributed target SAR image after coherent speckle suppression, which is expressed as follows:

[0117]

[0118] Where I3 represents the texture feature image matrix of the distributed target SAR image after coherent speckle suppression;

[0119] Step 3, mask generation: According to the grayscale histogram of the texture feature image in step 2, the Otsu threshold algorithm is used to segment the image foreground and background to obtain the initial mask, such as Figure 5 As shown; the image mask is generated by using the region growing algorithm to obtain the mask image, as shown Figure 6 As shown, specifically:

[0120] Step 3.1, segment the grayscale histogram of the texture feature image in step 2.7 using the Otsu threshold algorithm:

[0121] According to formula (11), the image grayscale value is traversed so that the inter-class variance σ of the foreground and background 2 Maximum, the image gray value at this time is used as the image threshold t for segmentation, the expression is as follows;

[0122] σ 2 =w0(t)w1(t)[μ0(t)-μ1(t)] 2 (11)

[0123] Wherein, w0 represents the ratio of the number of foreground pixels to the total number of pixels after threshold segmentation, w1 represents the ratio of the number of background pixels to the total number of pixels after threshold segmentation, μ0 represents the average value of foreground pixels after threshold segmentation, and μ1 represents the average value of background pixels after threshold segmentation;

[0124] Step 3.2, performing region growing on the segmented texture feature image, selecting points with zero pixel values ​​in different regions of the segmented texture feature image as initial seed points I4(i,j);

[0125] Step 3.3, compare the growth conditions of the initial seed point I4(i,j) with its eight adjacent points. If they meet the requirements, save the point to the growth region matrix. The growth conditions are as follows:

[0126] I4(i+u,j+v)-I4(i,j)≤0.15 (12)

[0127] In the formula, I4(i,j) represents the pixel value of the seed point, and I4(i+u,j+v) represents the eight points adjacent to the seed point;

[0128] Step 3.4, repeat step 3.3 for the remaining initial seed points until all initial seeds are traversed to obtain the final region growing result, which is used as the generated image mask;

[0129] Step 4, mask processing: Calculate the Hadamard product of the generated image mask matrix and the pre-processed distributed target SAR image matrix to obtain a uniform distributed target SAR image, such as Figure 7 As shown, specifically:

[0130] Assume that the size of the distributed target SAR image matrix after preprocessing is m*n. After steps 2 and 3, the size of the image mask matrix is ​​also m*n. According to formula (13), the Hadamard product of the distributed target SAR image matrix after preprocessing and the image mask matrix is ​​calculated to obtain the uniform distributed target SAR image. The expression is as follows:

[0131] S'=(I0⊙H) m*n (13)

[0132] Where S' represents the homogenized distributed target SAR image, I represents the preprocessed distributed target SAR image matrix, and H represents the image mask matrix.

[0133] A distributed target image non-uniform structure processing system for Mars ionosphere detection SAR, comprising:

[0134] Image preprocessing module: used to suppress the coherent speckle of the original distributed target SAR radar image to obtain the preprocessed distributed target SAR image;

[0135] Image generation module: According to the pixel values ​​of the preprocessed distributed target SAR image, the gray level co-occurrence matrix is ​​calculated. Based on the gray level co-occurrence matrix, the texture feature value at the center of the window is calculated, and the preprocessed distributed target SAR image is traversed to generate a texture feature image.

[0136] Mask generation module: used to segment the texture feature image using the Otsu threshold algorithm, and then generate the image mask using the region growing algorithm to obtain the mask image;

[0137] Mask processing module: used to calculate the Hadamard product between the mask image matrix and the preprocessed distributed target SAR image matrix to obtain a homogenized distributed target SAR image.

[0138] A distributed target image non-uniform structure processing device for Mars ionosphere detection SAR, comprising:

[0139] Memory: a computer-readable device used to store the computer program of the above-mentioned distributed target image non-uniform structure processing method for Mars ionosphere detection SAR;

[0140] Processor: used to implement the above-mentioned distributed target image non-uniform structure processing method for Mars ionosphere detection SAR when executing the computer program.

[0141] A computer-readable storage medium stores a computer program, which, when executed by a processor, can implement the above-mentioned distributed target image non-uniform structure processing method for Mars ionosphere detection SAR.

[0142] Compared with the original distributed target SAR image ( Figure 2 ) compared with the uniform distributed target SAR image ( Figure 7 ) It can be clearly seen that the non-uniform part (yellow light) in the original distributed target SAR image is effectively removed, which improves the uniformity of the distributed target SAR image.

[0143] Compared with the phase image of the original distributed target SAR image ( Figure 8 ), the uniform distributed target SAR image obtained by the processing method of the present invention ( Fig. 9 ) can be clearly seen from the phase diagram that there is no change in the phase of the uniform area, indicating that through the processing method of the present invention, while the non-uniform area is targetedly eliminated, the phase of the uniform area can still be maintained, thereby achieving the uniformity of the distributed target SAR image and greatly improving the uniformity of the image.

[0144] The coefficient of variation CV reflects the degree of data dispersion through the ratio of the standard deviation to the mean. The smaller the value, the more concentrated the data distribution, that is, the higher the uniformity. In image analysis, the coefficient of variation is often used to describe the uniformity of image pixel values, thereby indirectly reflecting the quality or effect of the image. The smaller the coefficient of variation, the higher the uniformity of the image, indicating that the pixel value distribution is more concentrated; the larger the coefficient of variation, the worse the uniformity, indicating that the pixel value distribution in the image is more discrete. The expression is as follows:

[0145]

[0146] In the formula, σ represents the standard deviation of the sample, and u represents the mean of the sample;

[0147] The coefficient of variation CV of the SAR image after image preprocessing is 107.27, and the coefficient of variation CV of the image after processing by the present invention is 73.48, indicating that the uniformity of the image after processing by the present invention is improved.

[0148] The correlation coefficient r is often used to evaluate the similarity of data. Its value range is between -1 and 1: 1 means a perfect positive correlation, 0 means no correlation, and -1 means a perfect negative correlation. The correlation coefficient can be used to quantitatively describe the degree of similarity between data. The expression is as follows:

[0149]

[0150] Where, X i represents the radar phase matrix element value after preprocessing, represents the average value of the radar phase matrix elements after preprocessing, Y i represents the radar phase matrix element value after being processed by the present invention, represents the average value of the radar phase matrix elements after being processed by the present invention;

[0151] The correlation coefficient between the radar phase after the processing of the present invention and the radar phase after the preprocessing is 0.98, indicating that the radar phase is well maintained.

Claims

1. A distributed target image non-uniform structure processing method for Mars ionosphere detection SAR, characterized in that: The following steps are involved: Step 1, distributed target SAR image preprocessing: input the original distributed target SAR image, perform Frost filtering, suppress the coherent speckle of the distributed target SAR image, and obtain the distributed target SAR image after the coherent speckle suppression; Step 2, texture feature image generation: according to the pixel values ​​of the preprocessed distributed target SAR image, the gray level co-occurrence matrix is ​​calculated, and based on the gray level co-occurrence matrix, the texture feature value at the center of the window is calculated, and the preprocessed distributed target SAR image is traversed to generate a texture feature image; Step 3, mask generation: according to the grayscale histogram of the texture feature image, the Otsu threshold algorithm is used to segment the foreground and background of the texture feature image to obtain an initial mask; then the region growing algorithm is used to generate an image mask to obtain a mask image; Step 4, mask processing: Calculate the Hadamard product of the mask image matrix and the pre-processed distributed target SAR image matrix to obtain a homogenized distributed target SAR image.

2. The method for processing non-uniform structure of distributed target images for Mars ionosphere detection SAR according to claim 1, characterized in that: The process of speckle suppression in step 1 is as follows: The Frost filtering algorithm is used to suppress the coherent speckle of the input original distributed target SAR image. That is, according to equations (1) and (2), the four neighborhoods are used to traverse each pixel value in the image in turn to obtain the pixel gray value I after filtering. i ' j , the expression is as follows: I′ ij =∑ p∈η m p I ij (1) Where η represents the four neighborhoods, i.e., the four pixels above, below, left, and right adjacent to the central pixel, I ij Represents the pixel value of the original distributed target SAR image, m p represents the weight value of the central pixel of the four neighborhoods, K represents the coordination factor, C s represents the pixel variance in the four neighborhoods, d s,p Represents the Euclidean distance between a single pixel in the four neighborhoods and the central pixel.

3. The method for processing non-uniform structure of distributed target images for Mars ionosphere detection SAR according to claim 1, characterized in that: The process of generating the texture feature image in step 2 includes: Step 2.1, set the window size: Assume the window size is L*L, usually L is much smaller than n and is an odd number, the expression is as follows: In the formula, w represents the window, and L represents the number of rows and columns of the window; Step 2.2, mirror-extend the image edge of the distributed target SAR image after speckle suppression, specifically: Assume that the distributed target SAR image matrix I after speckle suppression is: In the formula, S 11 represents the value at the (1, 1)th position in the distributed target SAR image data, S nm Represents the value at the (n, m)th position in the image data; The distributed target SAR image data after speckle suppression is mirror-extended to the image edge: In the formula, S ij (1≤i≤n,1≤j≤m) represents the data of the distributed target SAR image data after the image edge is extended by mirror, and I1 represents the distributed target SAR image matrix after the image edge is extended by mirror; Step 2.3, read S ij Data with a window size of L*L in (1≤i≤n,1≤j≤m): In the formula, It represents the value at the (1, 1) position in the data of the distributed target SAR image after the image edge is extended by the mirror. Represents the value at the (L, L)th position in the data of the distributed target SAR image after the image edge is extended by the mirror; Step 2.4, call the gray-level co-occurrence matrix function, calculate the gray-level co-occurrence matrix value corresponding to the center pixel of window w1 when the pixel pair spacing is 1 and the pixel pair position is 0°, 45°, 90°, and 135°, and normalize it to obtain the normalized gray-level co-occurrence matrix P' as the gray-level co-occurrence matrix calculation value for texture feature extraction. The expression is as follows: Where P 0° Represents the gray level co-occurrence matrix when the pixel pair position is 0°, P 45° Represents the gray level co-occurrence matrix when the pixel pair position is 45°, P 90° Represents the gray level co-occurrence matrix when the pixel pair position is 90°, P 135° Represents the gray level co-occurrence matrix when the pixel pair is at an angle of 135°, P0' ° Represents the normalized gray-level co-occurrence matrix of the pixel pair at an angle of 0°, P4' 5° Represents the normalized gray-level co-occurrence matrix of the pixel pair at a 45° angle, P9' 0° Represents the normalized gray-level co-occurrence matrix of the pixel pair at a 90° angle, P1' 35° Represents the normalized gray-level co-occurrence matrix of pixel pairs at an angle of 135°; Step 2.5, use the normalized gray-level co-occurrence matrix P' to calculate the angular second-order moment of the center pixel of the window as the eigenvalue of the center pixel of the L*L window. The expression is as follows: Where E represents the angular second-order moment value of the center pixel of the window; Step 2.6, traverse the preprocessed distributed target SAR image and calculate the eigenvalue of each pixel: Slide the L*L window, and traverse each pixel in the distributed target SAR image after preprocessing with the center pixel of the window. Calculate the eigenvalue of each pixel to obtain the texture feature image after edge mirror expansion. The expression is as follows: Where I2 represents the texture feature image matrix after edge mirror expansion; Step 2.7, crop the texture feature image after edge mirror expansion to obtain the texture feature image of the distributed target SAR image after coherent speckle suppression, which is expressed as follows: Where I3 represents the texture feature image matrix of the distributed target SAR image after speckle suppression.

4. The method for processing non-uniform structure of distributed target images for Mars ionosphere detection SAR according to claim 1, characterized in that: The process of generating the image mask in step 3 includes: Step 3.1, segment the grayscale histogram of the texture feature image in step 2.7 using the Otsu threshold algorithm: According to formula (11), the image grayscale value is traversed so that the inter-class variance σ of the foreground and background 2 Maximum, the image gray value at this time is used as the image threshold t for segmentation, the expression is as follows; σ 2 =w0(t)w1(t)[μ0(t)-μ1(t)] 2 (11) Wherein, w0 represents the ratio of the number of foreground pixels to the total number of pixels after threshold segmentation, w1 represents the ratio of the number of background pixels to the total number of pixels after threshold segmentation, μ0 represents the average value of foreground pixels after threshold segmentation, and μ1 represents the average value of background pixels after threshold segmentation; Step 3.2, performing region growing on the segmented texture feature image, selecting points with zero pixel values ​​in different regions of the segmented texture feature image as initial seed points I4(i,j); Step 3.3, compare the growth conditions of the initial seed point I4(i,j) with its eight adjacent points. If they meet the requirements, save the point to the growth region matrix. The growth conditions are as follows: I4(i+u,j+v)-I4(i,j)≤0.15 (12) In the formula, I4(i,j) represents the pixel value of the seed point, and I4(i+u,j+v) represents the eight points adjacent to the seed point; Step 3.4, repeat step 3.3 for the remaining initial seed points until all initial seeds are traversed to obtain the image mask.

5. The method for processing non-uniform structure of distributed target images for Mars ionosphere detection SAR according to claim 1, characterized in that: The process of mask processing in step 4 is as follows: Assume that the size of the distributed target SAR image matrix after preprocessing is m*n. After steps 2 and 3, the size of the image mask matrix is ​​also m*n. According to formula (13), the Hadamard product of the distributed target SAR image matrix after preprocessing and the image mask matrix is ​​calculated to obtain the uniform distributed target SAR image S', which is expressed as follows: S'=(I0⊙H) m*n (13) Where I represents the original distributed target SAR image matrix, and H represents the image mask matrix.

6. A distributed target image non-uniform structure processing system for Mars ionosphere detection SAR, characterized in that: include: Image preprocessing module: used to suppress the coherent speckle of the original distributed target SAR radar image to obtain the preprocessed distributed target SAR image; Image generation module: According to the pixel values ​​of the preprocessed distributed target SAR image, the gray level co-occurrence matrix is ​​calculated. Based on the gray level co-occurrence matrix, the texture feature value at the center of the window is calculated, and the preprocessed distributed target SAR image is traversed to generate a texture feature image. Mask generation module: used to segment the texture feature image using the Otsu threshold algorithm, and then generate the image mask using the region growing algorithm to obtain the mask image; Mask processing module: used to calculate the Hadamard product between the mask image matrix and the preprocessed distributed target SAR image matrix to obtain a homogenized distributed target SAR image.

7. A distributed target image non-uniform structure processing device for Mars ionosphere detection SAR, characterized in that: include: Memory: a computer program for storing a method for processing non-uniform structures of distributed target images for Mars ionosphere detection SAR according to any one of claims 1 to 5, which is a computer-readable device; Processor: used to implement the distributed target image non-uniform structure processing method for Mars ionosphere detection SAR as described in any one of claims 1-5 when executing the computer program.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it can implement the distributed target image non-uniform structure processing method for Mars ionosphere detection SAR as described in any one of claims 1-5.

Images