SAR image dark region identification method based on superpixel statistical characteristics

Abstract

The invention provides an SAR image dark region recognition method based on superpixel statistical characteristics, and the method comprises the steps: collecting an SAR image of a target region through an SAR system platform, carrying out the homogeneous region segmentation of the SAR image, and carrying out the filtering and recombination of the SAR image, and obtaining an SAR image after superpixel filtering; then, searching a boundary range of an optimal double-peak region in the SAR image, and intercepting block images in the boundary range; and dividing the block image into a dark region and a signal region, solving a gray value which enables the between-class variance of the dark region and the signal region to be maximum by using a maximum between-class variance principle, and taking the gray value as an optimal threshold value, thereby determining a region formed by pixels smaller than the optimal threshold value in the SAR image after super-pixel filtering as the dark region. According to the method, the double-peak region is constructed through superpixel statistical characteristics, and the optimal threshold of the retrieval dark region is determined by using the threshold segmentation algorithm, so that the detection precision of the SAR image dark region is further improved.

Description

Technical Field

[0001] The invention belongs to the technical field of radar signal processing, and in particular relates to a SAR image dark area recognition method based on superpixel statistical characteristics. Background Art

[0002] Synthetic Aperture Radar (SAR) technology is widely used in terrain mapping, biomass estimation, ocean monitoring, disaster assessment and military reconnaissance due to its high-resolution and wide-area imaging capabilities.

[0003] Synthetic aperture radar records the echo of ground objects by actively transmitting electromagnetic waves. Its slant range imaging characteristics make the SAR images contain targets that cannot be illuminated by electromagnetic waves and shadow areas where electromagnetic wave signals cannot return to the receiver, as well as targets with low backscattering coefficients such as asphalt roads and water areas. These are noise-dominated radiation areas in SAR images and are collectively referred to as dark areas. Dark areas in SAR images have important application value. In mountainous areas and building areas, dark areas are conducive to estimating the elevation difference information of terrain objects. The texture information at the junction of dark areas and signal areas is conducive to target detection and classification. Analyzing the distribution characteristics of noise in dark areas is conducive to improving the accuracy of constellation SAR cross-radiation calibration and the effectiveness of SAR image filtering. Similarly, when using interferometric synthetic aperture radar technology to produce digital elevation models, the phase unwrapping error in the dark area will cause errors in elevation inversion. Therefore, it is necessary to compensate for the elevation error caused by the dark area in the interferometric synthetic aperture radar application. The ability to automatically and accurately detect dark areas in SAR images is of great value to the application of SAR in various fields. At present, how to quickly and accurately obtain dark areas in SAR images has become an urgent problem to be solved. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a method for identifying dark areas in SAR images based on superpixel statistical characteristics. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0005] A SAR image dark area recognition method based on superpixel statistical characteristics includes:

[0006] Step 1, using the SAR system platform to collect SAR images of the target area;

[0007] Step 2, performing homogeneous region segmentation on the SAR image to obtain multiple homogeneous regions;

[0008] Step 3, filtering the multiple homogeneous regions to obtain filtered homogeneous regions, and recombining the filtered homogeneous regions into a SAR image to obtain a super-pixel filtered SAR image;

[0009] Step 4, searching for the boundary range of the optimal bimodal area in the superpixel filtered SAR image;

[0010] Step 5, intercepting a block image within the boundary range from the super-pixel filtered SAR image;

[0011] Step 6, dividing the block image into a dark area and a signal area, and using the maximum inter-class variance principle to find the gray value that maximizes the inter-class variance between the dark area and the signal area, and using the gray value as the optimal threshold;

[0012] Step 7, searching for pixels smaller than the optimal threshold in the superpixel filtered SAR image, and forming a dark area.

[0013] Beneficial effects:

[0014] (1) The present invention proposes a dark area recognition method for SAR images based on superpixel statistical characteristics. The method can construct a bimodal distribution characteristic curve through superpixel statistical characteristics and converge to the global optimal solution by using the OTSU threshold segmentation method. This makes the threshold for dark area detection in SAR images interpretable, which makes up for the problem that the threshold for dark area detection in SAR images by traditional algorithms is inaccurate and improves the detection accuracy of dark areas in SAR images.

[0015] (2) The present invention realizes accurate automatic calculation of dark areas in SAR images. It only needs to input SAR image data without other auxiliary data. It has a high degree of automation and can meet new industry needs.

[0016] (3) The present invention supports multiple SAR working modes, including strip mode, spotlight mode, sliding spotlight mode, scanning mode, mosaic mode and TOPS mode, etc. The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of a method for identifying dark areas in SAR images based on superpixel statistical characteristics provided by an embodiment of the present invention;

[0018] Figure 2 The SAR image of Weinan area in Shaanxi Province and the homogeneous region segmentation result diagram thereof provided by the embodiment of the present invention;

[0019] Figure 3 A diagram showing the dark area detection experiment results of the mountainous area using the Tianhui-2 data provided by an embodiment of the present invention;

[0020] Figure 4 A diagram showing the dark area detection experiment results of the plain area of ​​the Tianhui-2 data provided in an embodiment of the present invention;

[0021] Figure 5 A schematic diagram of the comparison results between the method of the present invention and other existing methods in mountainous areas provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0023] like Figure 1 As shown, the present invention provides a SAR image dark area recognition method based on superpixel statistical characteristics, including:

[0024] Step 1, using the SAR system platform to collect SAR images of the target area;

[0025] In a specific embodiment of the present invention, step 1 comprises:

[0026] Step 11, building a SAR system platform;

[0027] Step 12, using the radar beam of the SAR system platform to illuminate the ground target within the target area to obtain the original echo returned by the ground target;

[0028] Step 13: obtaining a SAR image containing ground targets based on the original echo.

[0029] The SAR system platform can be a satellite, an airplane, an unmanned aerial vehicle, or a vehicle-mounted platform. During the radar startup period, each system platform uses a radar beam to illuminate the ground objects in the target area to record the echoes, and obtains a SAR image after processing the SAR original echoes. The working mode of the SAR system platform supports common imaging modes such as strip mode, spotlight mode, sliding spotlight mode, scanning mode, mosaic mode, and TOPS.

[0030] Step 2, performing homogeneous region segmentation on the SAR image to obtain multiple homogeneous regions;

[0031] As an optional embodiment of the present invention, step 2 includes:

[0032] Step 21, normalizing the amplitude of the SAR image to obtain a standardized SAR image;

[0033] In order to facilitate the calculation and storage of data, the SAR image amplitude needs to be quantified. Maximum and minimum value quantization is the most commonly used quantization method. Therefore, the standardization process in this step is expressed as follows:

[0034] A′(x,y)=255×(A(x,y)-A min ) / (A max -A min );

[0035] Where A′(x, y) is the grayscale matrix of the SAR image quantized to the range of [0, 255] after standardization, A(x, y) represents the grayscale matrix of the SAR image, and A max and A min are the maximum grayscale value and the minimum grayscale value on the SAR image, respectively. (x, y) represents the one-dimensional index of the row and column in the grayscale matrix;

[0036] Maximum and minimum value normalization is to uniformly quantize the range of data, usually to remove the influence of strong points in SAR data on quantization. max The value of is usually quantified using the 99.99% quantile data.

[0037] Step 22, according to the number of views L of multi-view processing, calculate the number of superpixels N that need to be deployed in the standardized SAR image labels , and according to the number of seeds N labels Generate seeds;

[0038] According to the number of views L in multi-view processing, the number of superpixel seeds that need to be deployed in the image is calculated as N labels =L 2 , then the set of superpixels is represented as The pixel point set of the standardized SAR image is

[0039] Step 23, selecting seeds as cluster centers in the standardized SAR image according to a predetermined row staggering method; wherein the row staggering method defines the spacing between adjacent seeds, and the starting positions of odd rows and even rows;

[0040] This step uses a hexagonal structure instead of the traditional rectangular structure, and the spacing between adjacent seeds is The starting position of the odd-numbered rows is S L / 2, the starting position of the even-numbered rows is S L The row staggered mode, where N labels =L 2 is the number of seeds, N SAR is the number of pixels in the standardized SAR image; by distributing the rows of cluster centers in this way, a more compact hexagonal network structure superpixel is generated. L An integer multiple of N. Usually, after the cluster center assignment is completed, the number of superpixels N needs to be recalculated and updated. labels , the cluster center is c i =[a i ,x i ,y i ] T ,i∈[1,N labels]. Similarly, the grayscale of each pixel in the SAR image is represented by: A j =[a j ,x j ,y j ] T ,j∈[1,N SAR ].

[0041] Step 24, adjusting the position of the cluster center according to the gradient of the pixel point to obtain a standardized SAR image with the adjusted cluster center position;

[0042] Since the superpixel seed points with uniform staggered distribution may be scattered on the image texture boundary, in order to make the cluster center c i It can better represent the properties of the superpixel area and enhance the clustering effect. Traditional methods usually readjust each cluster center to the lowest gradient position in its 3×3 neighborhood. The gradient is expressed as:

[0043]

[0044] Although the traditional method effectively avoids the cluster center appearing on the boundary, it is not effective when the boundary of the image is a more complex contour. The present invention uses a method based on edge detection to improve the simple linear iterative clustering algorithm, and uses the boundary as a conditional constraint on the cluster center in the clustering process. Edge detection is performed on the SAR image amplitude, and the detection result is used as a mask file to limit the movement of the cluster center. In the clustering process, the equivalent position of each seed is determined. If it appears on the edge of the image, it is moved to a position without a surrounding boundary.

[0045] Step 25, clustering the standardized SAR image with the adjusted cluster center position, so that the pixel points are clustered toward the cluster center with the highest similarity to obtain multiple super pixel blocks, and labeling each super pixel block; wherein the pixel labels in the same super pixel block are the same, and the pixel labels in different pixel blocks are different;

[0046] Image clustering is a key step in superpixel segmentation. Its core idea is to classify the pixels in the standardized SAR image into their 2S L ×2S L In the clustering process, iteratively calculate each candidate pixel to 2S L ×2S L The similarity of all cluster centers within the window range is determined by spatial similarity and color similarity. Let candidate pixel j and cluster center c i The traditional method is to separate the candidate pixel j and the cluster center c i The similarity between them is expressed as:

[0047]

[0048] In the formula, d(c i ,j) is smaller, the closer the candidate pixel j is to the cluster center c i The more similar. S is the compactness factor. m S The smaller the value, the smaller the weight of color similarity and the more compact the superpixel. i ,j) describes the pixel j and the cluster center c i The similarity between .

[0049] Considering the influence of speckle noise in SAR images, the weight factor w is introduced 0 , and at the same time, a 3×3 neighborhood is taken for smoothing at each pixel to improve clustering accuracy. The similarity is expressed as:

[0050]

[0051] In the formula, d(c i ,j) represents the candidate pixel j and cluster center c i The similarity between i represents the cluster center of the ith superpixel block, c i =[a i ,x i ,y i ] T , a i Represents the cluster center c i The amplitude value, x i ,y i Represents the cluster center c i The location information, x j ,y j represents the position index of candidate pixel j, Represents the cluster center c i The 3*3 neighborhood of Represents x i ,y i The amplitude value of the pixel at represents the 3*3 neighborhood of candidate pixel j, Represents x j ,y j The amplitude value of the pixel at Represents C i The amplitude vector composed of the amplitude values ​​of the pixels in i represents the i-th superpixel block,

[0052] w 0 The value range is [0,1]. In the iterative process of solving formula (15), the weight factor w is changed.0 The value of is used to achieve the clustering accuracy of the current clustering block. At the same time, the global iteration is continuously performed until the superpixel similarity label remains basically unchanged. Experiments show that the improved algorithm generally converges to the required error after 5 iterations.

[0053] Step 26, for each super pixel block, if the label of the super pixel block is the same as that of other super pixel blocks and they are not adjacent, a new label is reassigned to the super pixel block;

[0054] Since similar regions after segmentation may have multiple connections, although the attributes of these connected regions are very similar and belong to the same superpixel, the spatial distance distortion makes the superpixel not compact. Therefore, in the superpixel segmentation process, it is necessary to re-assign new labels to the multi-connected regions of the same superpixel to make them a new independent superpixel. Usually, the adjacent regions of the image after superpixel segmentation are marked for connection, and N is continuously updated. labels Reshape superpixel patch labels.

[0055] Step 27, calculate the area of ​​each super pixel block, if the area of ​​the super pixel block is less than the threshold, merge it into the nearest super pixel block, and treat each super pixel block as a homogeneous area.

[0056] After step 26, the superpixel block may be too small. For some superpixels whose area is too small but cannot be eliminated by morphological processing, a method of setting an area threshold is adopted.

[0057] As an optional implementation of the present invention, step 27 includes:

[0058] Step 271, calculate the area N of each superpixel block SAR / N labels , if the area of ​​the superpixel block is smaller than the threshold, it is merged into the nearest superpixel block, and a mask Mask[Im x,y ] Na×Nr Matrix, Im of elements in Mask x,y 1 represents a foreground pixel, Im x,y 0 is the background pixel; Na represents the number of pixels in the azimuth direction of the SAR image, and Nr represents the number of pixels in the range direction of the SAR image.

[0059] The area N of each superpixel estimated by the preset number of superpixel seeds SAR / N labels As a benchmark, the superpixel C with an area that is too small to be compared with the benchmark i Separate and finally mark the morphologically processed pixels and superpixels that are too small to generate a mask Mask[Im x,y ]Na×Nr Matrix. Im of the elements in Mask x,y 1 represents the pixel whose label needs to be replaced, which is called the foreground pixel. i,j If it is 0, it is a normal pixel, called a background pixel.

[0060] Step 272, calculate the non-zero pixel distance between the gray value of the foreground pixel marked by Mask in the image and the background pixel, which is expressed as:

[0061] f(p)=min[f(p),D(p,q)+f(p)];

[0062] Where p is the foreground pixel, q is the background pixel, D(p,q) is the distance between points p and q; f(p) represents the non-zero pixel distance.

[0063] Step 273, obtaining the minimum non-zero pixel distance by searching row by row and column by column, and assigning the background pixel label of the minimum non-zero pixel distance to the foreground pixel label to obtain a new superpixel label;

[0064] Step 274, update the number of seeds to obtain a new number of seeds N' labels ;

[0065] Step 275, renumber the pixels in the superpixel block, and update the cluster center to generate the final superpixel block as a homogeneous region; wherein the pixels in the homogeneous region are represented as:

[0066] C' i =[x′ i ,y′ i ] T ,i∈[1,N′ labels ];

[0067] In the formula, x′ i ,y′ i Represents the position index of the pixel points in the homogeneous area in the grayscale matrix.

[0068] Step 3, filtering the multiple homogeneous regions to obtain filtered homogeneous regions, and recombining the filtered homogeneous regions into a SAR image to obtain a super-pixel filtered SAR image; the super-pixel filtered SAR image is expressed by the formula:

[0069]

[0070] In the formula, represents the grayscale matrix of the SAR image after superpixel filtering, is the super pixel block C i The number of pixels in .

[0071] This step removes the influence of coherent speckle noise on the backscatter amplitude of the ground object through the idea of ​​superpixel averaging, and restores the radar backscatter amplitude distribution characteristics of the ground object itself.

[0072] Step 4, searching for the boundary range of the optimal bimodal area in the superpixel filtered SAR image;

[0073] A large number of experiments have shown that SAR images with dark areas and characterized by superpixels all have "double peak" characteristics. With the continuous improvement of SAR resolution, the size of a SAR image with a resolution of 3 meters and a width of 20 km is usually about 22000*22000 pixels, and the size of a SAR image with a resolution of 1 meter and a width of 30 km is usually about 55000*55000 pixels. In order to better reflect the double peak characteristics and accelerate the convergence of the iterative OTSU algorithm, the present invention searches for the optimal double peak area in the global SAR image.

[0074] As an optional implementation of the present invention, step 4 includes:

[0075] Step 41, setting the search window of the bimodal characteristic region and the dark region threshold T low and high radiation area threshold T high , the search window size is M×M;

[0076] For the SAR image quantized by the method of the present invention, the value is generally 20, and the high radiation area threshold T is set. high , generally the value is 80.

[0077] Step 42, using the search window to slide in the SAR image after superpixel filtering, each sliding obtains a window area in turn, and uses the dark area threshold T low and high radiation area threshold T high Corresponding calculation of the dark area ratio coefficient k in the window area low and high radiation area ratio k high ;

[0078] The dark area ratio k low and high radiation area ratio k high The calculation formula is:

[0079]

[0080] Where k is the dark area ratio coefficient k low Or high radiation area ratio k high , T is the dark area threshold T corresponding to k low Or high radiation area threshold T high , B represents the SAR image area covered by the search window.

[0081] With H = 200 as the iteration step of the search window, the search window is used to traverse the entire SAR image to obtain the dark area ratio coefficient sequence K low And the high radiation area ratio coefficient sequence K high .

[0082] Step 43: Set all dark area proportion coefficients k low and high radiation area ratio k high They are composed into sequences respectively;

[0083] Step 44, searching the sequence for a sequence that satisfies maxK low &&K high >0.3 corresponds to the bimodal region and is taken as the optimal bimodal region;

[0084] In order to ensure that the bimodal characteristic region contains both dark areas and high radiation areas, the optimal bimodal region is defined as maxK low &&K high >0.3 area.

[0085] Step 45, determining the boundary range of the optimal bimodal region.

[0086] This step outputs the boundary range of the optimal bimodal region DP(x,y) for calculating the optimal threshold of the dark area.

[0087] Step 5, intercepting a block image within the boundary range from the super-pixel filtered SAR image;

[0088] Step 6, dividing the block image into a dark area and a signal area, and using the maximum inter-class variance principle to find the gray value that maximizes the inter-class variance between the dark area and the signal area, and using the gray value as the optimal threshold;

[0089] As an optional implementation of the present invention, step 6 includes:

[0090] Step 61, dividing the block image into a dark area and a signal area;

[0091] The pixel value range of the optimal bimodal region DP(x,y) is [0,1,...,255]. Let the optimal threshold of the dark region be T s ∈(0,255), DP(x,y) is divided into two categories: dark area and signal area.

[0092] Step 62, setting the dark area probability Sum signal area probability Among them, T S is the optimal threshold T s ;p l Represents the probability of occurrence of gray value l in [0,1,...,255];

[0093] Step 63, using the dark area probability and the signal area probability to calculate the average grayscale value of the dark area and the signal area, respectively expressed as and

[0094] Step 64, using the maximum inter-class variance principle, find the gray value that maximizes the inter-class variance between the dark area and the signal area, and use the gray value as the optimal threshold.

[0095] As an optional implementation of the present invention, step 64 includes:

[0096] Step 641, find the inter-class variance of the dark area and the signal area, expressed as:

[0097]

[0098] In the formula, Represents the zero-order cumulative moment of the gray histogram, which indicates the probability of grayscale appearing between 0 and t. Represents the first-order cumulative moment of the gray histogram, the average gray value from 0 to t, then the average gray value of the entire SAR image is

[0099] Step 642, according to the existence of P for any t O M O +M B P B =M T , transform the between-class variance to obtain:

[0100]

[0101] Step 643, based on Perform discriminant analysis to find out The maximum gray value is taken as the optimal threshold T s , expressed as:

[0102]

[0103] Step 7, searching for pixels smaller than the optimal threshold in the superpixel filtered SAR image, and forming a dark area.

[0104] The dark area is The shadow threshold of the original SAR image is T s A max / 255.

[0105] The present invention provides a SAR image dark area recognition method based on superpixel statistical characteristics. The method uses a SAR system platform to collect SAR images of a target area, and performs homogeneous area segmentation and filtering and reorganization on the SAR image after superpixel filtering; then, the boundary range of the optimal bimodal area is searched in the SAR image, and the block image within the boundary range is intercepted; the block image is divided into a dark area and a signal area, and the gray value that maximizes the inter-class variance of the dark area and the signal area is obtained by using the maximum inter-class variance principle, and the gray value is used as the optimal threshold, so that the area composed of pixels less than the optimal threshold in the SAR image after superpixel filtering is determined as a dark area. The present invention constructs a bimodal area through superpixel statistical characteristics, and uses a threshold segmentation algorithm to determine the optimal threshold for retrieving dark areas, further improving the detection accuracy of dark areas in SAR images. In addition, the present invention realizes accurate automatic calculation of dark areas in SAR images, only needs to input SAR image data without other auxiliary data, has a high degree of automation, and can meet new industry needs.

[0106] The present invention is described below through simulation experiments to illustrate the technical effects of the present invention.

[0107] Simulation experiment:

[0108] Figure 2 The superpixel experiment was conducted using SAR data from Weinan City, Shaanxi Province. 1000*1187 pixels were captured from the data in the area for display. The horizontal axis represents the distance direction, and the vertical axis represents the azimuth direction.

[0109] Figure 3 This is the experimental result of Tianhui-2 data in mountainous areas. Figure 3 Figure (a) in the middle shows the original oblique plane SAR image of a mountainous area taken by Tianhui-2. Figure 3 The results in Figure (b) compare the results of traditional pixel-level statistical characteristics and superpixel statistical characteristics. It can be seen that in mountainous areas, traditional pixel-level statistical characteristics cannot distinguish the boundary between signal areas and dark areas, while the SAR image dark area recognition algorithm based on superpixel statistical characteristics proposed in the present invention can accurately calculate the dark area threshold and separate the SAR image signal area and dark area. The accurate threshold value of the mountainous area obtained by the method of the present invention is 37.5 (SAR original data amplitude value). Figure 3 The middle (c) figure shows the dark area of ​​the SAR image calculated based on the dark area threshold. Figure 3 The middle figure (d) shows the effect of superimposing the dark area boundary on the SAR image. It can be clearly seen that the dark area boundary segmented by the present invention is very consistent with the SAR image signal and the boundary of the dark area.

[0110] Figure 4 This is the experimental result of Tianhui-2 data in plain area. Figure 4 Figure (a) in the middle shows the original oblique plane SAR image of a plain area taken by Tianhui-2. Figure 4 The results in Figure (b) compare the results of traditional pixel-level statistical characteristics and superpixel statistical characteristics. It can be seen that in the flat area, the traditional pixel-level statistical characteristics cannot distinguish the boundary between the signal area and the dark area, while the SAR image dark area recognition algorithm based on superpixel statistical characteristics proposed in the present invention can accurately calculate the dark area threshold and separate the SAR image signal area and dark area. The accurate threshold of the flat area obtained by the method of the present invention is 40 (SAR original data amplitude value). Figure 4 The middle (c) figure shows the dark area of ​​the SAR image calculated based on the dark area threshold. Figure 4 The middle figure (d) shows the effect of superimposing the dark area boundary on the SAR image. It can be clearly seen that the dark area boundary segmented by the present invention is very consistent with the SAR image signal and the boundary of the dark area.

[0111] Figure 5 This is a comparison result of the algorithm in the mountainous area. In order to further verify the effectiveness of the algorithm of the present invention, the present invention compares the detection results of the pixel-level optimal threshold segmentation method and the superpixel CFAR method in the mountainous area. Figure 5 Specific results are given. In order to better verify the effectiveness of the dark area recognition algorithm of SAR images based on superpixel statistical characteristics, the boundary fit and segmentation accuracy indicators are introduced.

[0112] ① Boundary Fit (BFT)

[0113] The boundary fit is defined as the superpixel block C of the dark area automatically identified by the algorithm proposed in this invention. i ,i∈[1,N′ labels The pixels within the two pixel range of the edge of ] and the superpixel C′ of the real dark area segmented manually o ,o∈[1,N″ labels ]The edge overlap is expressed as

[0114]

[0115] TP(C′ o ,C i ) is defined as the real dark area C′ j There is C within two pixels of the edge pixel i edge pixels in the o ,C i ) In contrast, a high boundary fit indicates the effectiveness of dark region extraction.

[0116] ②Dark area segmentation accuracy (DASA)

[0117] DASA compares the dark areas C automatically identified by the algorithm proposed in this invention. i Image C′ of the real dark area segmented by artificial segmentation o The maximum overlapping area is used to measure the segmentation accuracy. DASA is defined as

[0118]

[0119] In the formula, A(C′ o ) is the real dark area C′ o The area of ​​the dark area C detected by the algorithm i The more pixels there are, the closer it is to the real dark area C′ o The more intersections there are, the larger the DASA value is.

[0120] Based on the pixel-level optimal threshold method and super-pixel CFAR method and the method of the present invention, the Figure 3 The BFT indexes of the experimental results of the mountainous area of ​​Tianhui-2 data are 0.5132, 0.7137 and 0.9228, and the SASA indexes are 0.8352, 0.8039 and 0.9373 respectively; the optimal threshold method based on the pixel level and the superpixel CFAR method and the method of the present invention are calculated. Figure 4 The BFT indicators of the experimental results of the Tianhui-2 data in the plain area are 0.5697, 0.7566 and 0.8962, and the SASA indicators are 0.8102, 0.7529 and 0.9676, respectively. The experimental results show that the proposed method has better effect in the field of automatic identification of dark areas in SAR images, and is significantly better than the traditional pixel-level optimal threshold method and superpixel CFAR detection method.

[0121] According to calculation, the SAR image dark area recognition method based on superpixel statistical characteristics proposed by the present invention can improve the recognition accuracy of SAR image dark areas to more than 90%. Moreover, the entire algorithm program only takes 1 minute and 50 seconds to process a 22000×22000 pixel spaceborne SAR image, which is quite efficient.

[0122] It is worth noting that the terms "first" and "second" in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0123] Although the present application is described herein in conjunction with various embodiments, in the process of implementing the claimed application, those skilled in the art may understand and implement other variations of the disclosed embodiments by reviewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality of components or steps.

[0124] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A method for identifying dark areas in SAR images based on superpixel statistical characteristics, characterized in that: include: Step 1, using the SAR system platform to collect SAR images of the target area; Step 2, performing homogeneous region segmentation on the SAR image to obtain multiple homogeneous regions; Step 3, filtering the multiple homogeneous regions to obtain filtered homogeneous regions, and recombining the filtered homogeneous regions into a SAR image to obtain a super-pixel filtered SAR image; Step 4, searching for the boundary range of the optimal bimodal area in the superpixel filtered SAR image; Step 5, intercepting a block image within the boundary range from the super-pixel filtered SAR image; Step 6, dividing the block image into a dark area and a signal area, and using the maximum inter-class variance principle to find the gray value that maximizes the inter-class variance between the dark area and the signal area, and using the gray value as the optimal threshold; Step 7, searching for pixels smaller than the optimal threshold in the superpixel filtered SAR image, and forming a dark area.

2. The method for identifying dark areas in SAR images based on superpixel statistical characteristics according to claim 1, characterized in that: Step 1 includes: Step 11, building a SAR system platform; Step 12, using the radar beam of the SAR system platform to illuminate the ground target within the target area to obtain the original echo returned by the ground target; Step 13: obtaining a SAR image containing ground targets based on the original echo.

3. The method for identifying dark areas in SAR images based on superpixel statistical characteristics according to claim 1, characterized in that: Step 2 includes: Step 21, normalizing the amplitude of the SAR image to obtain a standardized SAR image; Step 22, calculating the number of seeds of superpixels to be arranged in the standardized SAR image according to the number of views L of the multi-view processing, and generating seeds according to the number of seeds; Step 23, selecting seeds as cluster centers in the standardized SAR image according to a predetermined row staggering method; wherein the row staggering method defines the spacing between adjacent seeds, and the starting positions of odd rows and even rows; Step 24, adjusting the position of the cluster center according to the gradient of the pixel point to obtain a standardized SAR image with the adjusted cluster center position; Step 25, clustering the standardized SAR image with the adjusted cluster center position, so that the pixel points are clustered toward the cluster center with the highest similarity to obtain multiple super pixel blocks, and labeling each super pixel block; wherein the pixel labels in the same super pixel block are the same, and the pixel labels in different pixel blocks are different; Step 26, for each super pixel block, if the label of the super pixel block is the same as that of other super pixel blocks and they are not adjacent, a new label is reassigned to the super pixel block; Step 27, calculate the area of ​​each super pixel block, if the area of ​​the super pixel block is less than the threshold, merge it into the nearest super pixel block, and treat each super pixel block as a homogeneous area.

4. The method for identifying dark areas in SAR images based on superpixel statistical characteristics according to claim 3, characterized in that: The standardization process in step 21 is expressed by the formula: A′(x,y)=255×(A(x,y)-A min ) / (TO max -TO min ); Where A′(x, y) is the grayscale matrix of the standardized SAR image quantized to the range of [0, 255], A(x, y) represents the grayscale matrix of the SAR image, and A max and A min are the maximum grayscale value and the minimum grayscale value on the SAR image, respectively. (x, y) represents the position index of the row and column in the grayscale matrix; The predetermined row misalignment mode in step 23 is: The spacing between adjacent seeds is The starting position of the odd-numbered rows is S L / 2, the starting position of the even-numbered rows is S L The row staggered mode, where N labels =L 2 is the number of seeds, N SAR is the number of pixels of the normalized SAR image; The gradient in step 24 is expressed as: The similarity in step 25 is expressed by the formula: In the formula, d(c i ,j) represents the candidate pixel j and cluster center c i The similarity between i represents the cluster center of the ith superpixel block, c i =[a i ,x i ,y i ] T , a i Represents the cluster center c i The amplitude value, x i ,y i Represents the cluster center c i The location information, x j ,y j represents the position index of candidate pixel j, Represents the cluster center c i The 3*3 neighborhood of Represents x i ,y i The amplitude value of the pixel at represents the 3*3 neighborhood of candidate pixel j, Represents x j ,y j The amplitude value of the pixel at Represents C i The amplitude vector composed of the amplitude values ​​of the pixels in i represents the i-th superpixel block, 5. The method for identifying dark areas in SAR images based on superpixel statistical characteristics according to claim 4, characterized in that: Step 27 includes: Step 271, calculate the area N of each superpixel block SAR / N labels , if the area of ​​the superpixel block is smaller than the threshold, it is merged into the nearest superpixel block, and a mask Mask[Im x,y ] Na×Nr Im x,y Matrix, Im of elements in Mask x,y 1 represents a foreground pixel, Im x,y 0 is the background pixel; Na represents the number of pixels in the azimuth direction of the SAR image, and Nr represents the number of pixels in the range direction of the SAR image; Step 272, calculate the non-zero pixel distance between the gray value of the foreground pixel marked by Mask in the image and the background pixel, which is expressed as: f(p)=min[f(p),D(p,q)+f(p)]; Where p is the foreground pixel, q is the background pixel, D(p,q) is the distance between points p and q; f(p) represents the non-zero pixel distance; Step 273, obtaining the minimum non-zero pixel distance by searching row by row and column by column, and assigning the background pixel label of the minimum non-zero pixel distance to the foreground pixel label to obtain a new superpixel label; Step 274, update the number of seeds to obtain a new number of seeds N' labels ; Step 275, renumber the pixels in the superpixel block, and update the cluster center to generate the final superpixel block as a homogeneous region; wherein the pixels in the homogeneous region are represented as: C’ i =[x′ i ,y′ ii ] T ,i∈[1,N′ labels ]: In the formula, x′ i ,y′ i Represents the position index of the pixel points in the homogeneous area in the grayscale matrix.

6. The method for identifying dark areas in SAR images based on superpixel statistical characteristics according to claim 4, characterized in that: The SAR image after superpixel filtering in step 3 is expressed as: In the formula, represents the grayscale matrix of the SAR image after superpixel filtering, is the superpixel block C i The number of pixels in .

7. The method for identifying dark areas in SAR images based on superpixel statistical characteristics according to claim 1, characterized in that: Step 4 includes: Step 41, setting the search window of the bimodal characteristic region and the dark region threshold T low and high radiation area threshold T high , the search window size is M×M; Step 42, using the search window to slide in the SAR image after superpixel filtering, each sliding obtains a window area in turn, and uses the dark area threshold T low and high radiation area threshold T high Corresponding calculation of the dark area ratio coefficient k in the window area low and high radiation area ratio k high ; Step 43: Set all dark area proportion coefficients k low and high radiation area ratio k high They are composed into sequences respectively; Step 44, searching the sequence for a sequence that satisfies maxK low &&K high >0.3 corresponds to the bimodal region and is taken as the optimal bimodal region; Step 45, determining the boundary range of the optimal bimodal region.

8. The method for identifying dark areas in SAR images based on superpixel statistical characteristics according to claim 7, characterized in that: The dark area ratio k low and high radiation area ratio k high The calculation formula is: Where k is the dark area ratio coefficient k low Or high radiation area ratio k high , T is the dark area threshold T corresponding to k low Or high radiation area threshold T high , B represents the SAR image area covered by the search window.

9. The method for identifying dark areas in SAR images based on superpixel statistical characteristics according to claim 8, characterized in that: Step 6 includes: Step 61, dividing the block image into a dark area and a signal area; Step 62, setting the dark area probability Sum signal area probability Among them, T S is the optimal threshold T s ;p l Represents the probability of occurrence of gray value l in [0,1,...,255]; Step 63, using the dark area probability and the signal area probability to calculate the average grayscale value of the dark area and the signal area, respectively expressed as and Step 64, using the maximum inter-class variance principle, find the gray value that maximizes the inter-class variance between the dark area and the signal area, and use the gray value as the optimal threshold.

10. The method for identifying dark areas in SAR images based on superpixel statistical characteristics according to claim 9, characterized in that: Step 64 includes: Step 641, find the inter-class variance of the dark area and the signal area, expressed as: In the formula, Represents the zero-order cumulative moment of the gray histogram, which indicates the probability of grayscale appearing between 0 and t. Represents the first-order cumulative moment of the gray histogram, the average gray value from 0 to t, then the average gray value of the entire SAR image is Step 642, according to the existence of P for any t O M O +M B P B =M T , transform the between-class variance to obtain: Step 643, based on Perform discriminant analysis to find out The maximum gray value is taken as the optimal threshold T s , expressed as:

Images