An Improved Method for Selecting High-Coherence Points in Ground-Based SAR

Abstract

This invention relates to the field of image processing technology, specifically to an improved method for selecting high coherence points in ground-based SAR. The method includes the following steps: extracting amplitude and phase frames, calculating statistical features, and fusing them to generate an index map; dividing regions, constructing grayscale histograms, and generating Boolean masks; establishing a co-occurrence matrix to remove abnormal regions; establishing pixel connectivity relationships to generate a graph structure; and mapping paths to output a high coherence point selection scheme. In this invention, a multi-source fusion layer is constructed by combining amplitude and phase features to enhance image discrimination dimensions. Region recognition accuracy is improved through grayscale frequency range extraction and Boolean mask generation. Local structural features are extracted using the grayscale co-occurrence matrix, and abnormal pixels are removed by combining boundary coherence and density features. A graph structure relationship based on adjacency paths is established to form a pixel connectivity network. From this network, a set of pixels with aggregation and connectivity is selected to achieve high coherence point clustering output, enhancing the completeness of structural expression within complex regions and the rationality of target point aggregation.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to an improved method for selecting high coherence points in ground-based SAR. Background Technology

[0002] Image processing technology primarily studies how to acquire, enhance, analyze, and understand image information. It involves processing image data acquired from image sensors using algorithms for noise reduction, enhancement, feature extraction, image segmentation, pattern recognition, and image understanding. It is widely applied in remote sensing monitoring, medical imaging, industrial inspection, intelligent transportation, and many other fields. Remote sensing image processing, as an important branch, is dedicated to improving the interpretation accuracy of Earth observation images. Especially in synthetic aperture radar (SAR) imagery, image processing technology is crucial for monitoring surface deformation, target identification, and information extraction. Traditional ground-based SAR high-coherence point selection methods involve analyzing the amplitude and phase stability of time-series SAR images and using fixed thresholds to select a set of highly stable points for interferometric processing. These methods typically use amplitude deviation methods, phase standard deviation methods, or amplitude consistency indices from stable scatterer interferometry as criteria. Candidate pixels are first extracted, and then selection is made based on the pixels' stable performance over time. Some methods introduce local neighborhood judgment to eliminate noise points.

[0003] Relying on a single statistical indicator for pixel stability evaluation makes it difficult to identify subtle structural differences in mixed ground features. In cases where gray levels are similar but spatial characteristics differ significantly, misidentification is likely to occur. The lack of joint consideration of gray level relationships and boundary structures within the region makes it difficult to eliminate misjudged areas. In scenarios where pixel boundaries are discontinuous or gray level textures are blurred, edge diffusion and false point selection are likely to occur, resulting in unclear identification of internal regional structures and insufficient accuracy in point set selection, which affects graph structure construction and connectivity analysis in subsequent interferometric modeling processes. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide an improved method for selecting high coherence points in ground-based SAR, comprising the following steps:

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an improved method for selecting high coherence points in ground-based SAR, comprising the following steps:

[0006] S1: Extract the original image frames from the ground-based SAR amplitude and phase sequences. Calculate the mean and deviation of the amplitude images according to the map sheet. Perform difference expansion between phase frames and calculate the variance map. After normalization, overlay according to the regional label fusion logic to output the joint index layer map.

[0007] S2: Divide the urban area and mixed vegetation area into blocks in the joint index layer map, construct a gray level histogram by statistically analyzing the gray level distribution frequency of pixels in the block, select the frequency gray level interval as the recognition range, logically judge the pixel value of each block to generate a Boolean mask map, and output the regional distribution mask map.

[0008] S3: Based on the selected patch in the regional distribution mask, extract the original image to construct a gray-level co-occurrence matrix, obtain the gray-level correlation features of the patch, detect the boundary closure and density ratio, remove discontinuous and sparse regions of the boundary, and output the structural texture screening map.

[0009] S4: Extract cell coordinates from the structure texture filtering map, establish distance association between neighboring cells, compare path grayscale differences to determine adjacency rationality, construct the relationship between nodes and edges in the structure graph, and output the spatial connection graph structure object.

[0010] As a further embodiment of the present invention, the joint index layer map includes an amplitude statistical feature map, a phase difference feature map, a normalization processing map, and a region fusion map; the region distribution mask map includes an urban area identification mask, a vegetation mixed area mask, a grayscale frequency filtering map, and a logical judgment result map; the structural texture filtering map includes a grayscale correlation map, a boundary integrity map, a density distribution map, and a filtering region map; and the spatial connection map structure object includes a pixel node set, an adjacency relationship set, a path grayscale difference map, and a structural edge map.

[0011] As a further aspect of the present invention, the definition of the frequency grayscale range refers to the continuous grayscale value range that appears frequently in the grayscale histogram of the image patch and represents the grayscale distribution of the pixel, and is the criterion for judging the pixel value of the identification area.

[0012] The definition of the discontinuous and sparse regions of the boundary refers to the parts of the structural texture map where the edge contour is discontinuous, the pixel density is lower than a set threshold, and the verification area does not have completeness and representativeness.

[0013] The definition of establishing the distance association between neighboring pixels refers to determining the rationality of the connection by calculating the spatial distance and path grayscale changes between pixels, and constructing a node-edge network that reflects the spatial topological relationship.

[0014] As a further aspect of the present invention, the specific steps of S1 are as follows:

[0015] S101: Extract the original image frames from the ground-based SAR amplitude sequence and phase sequence, overlay the gray values ​​of each pixel in the amplitude image by multiple frames according to the map sheet, calculate the mean and deviation based on the total number, cumulative value and sum of squares, and generate an amplitude gray value statistics layer.

[0016] S102: Call the amplitude grayscale statistics layer, calculate the inter-frame phase difference value of the phase image frame according to the time sequence, perform linear expansion processing, and calculate the variance based on the set of expanded values ​​of each pixel to generate a phase expansion variance layer.

[0017] S103: Normalize the amplitude grayscale statistics layer and the phase expansion variance layer, set the overlay rules according to the region label, perform weighted overlay on the normalized value of each pixel, and obtain the joint index layer map.

[0018] As a further aspect of the present invention, the specific steps of S2 are as follows:

[0019] S201: Based on the urban area and vegetation mixed area block division results in the joint index layer map, the gray values ​​of all pixels in the block are statistically analyzed, the correspondence between gray values ​​and occurrence frequency is established according to the pixel gray level, the gray frequency distribution of each block is obtained, and the gray frequency statistics of the block are generated.

[0020] S202: Based on the grayscale frequency statistics of the patch, extract the grayscale value range in the middle segment of the cumulative frequency distribution, determine whether the frequency span of consecutive grayscale levels meets the threshold condition according to the frequency, obtain the upper and lower limits of grayscale values ​​that meet the condition, and generate the patch grayscale recognition interval.

[0021] S203: Call the grayscale recognition range of the image block, make a logical judgment on whether the grayscale value of each pixel in the image block falls into the corresponding recognition range, mark the judgment result as a binary Boolean value, and construct a binary logic graph according to the image area to obtain the regional distribution mask map.

[0022] As a further aspect of the present invention, the specific steps of S3 are as follows:

[0023] S301: Call the selected pixel patch in the regional distribution mask image, obtain the pixel gray value combination for the corresponding patch in the original image, perform joint frequency statistics on all adjacent pixel gray value pairs in each patch, construct a co-occurrence matrix according to gray value pairs, statistically analyze the distribution relationship of gray value pairs, and obtain the gray value correlation distribution characteristics of the patch.

[0024] S302: Based on the gray-level correlation distribution characteristics of the image patch, perform boundary closure judgment on the gray-level adjacency relationship of edge pixels within the image patch, detect whether there is a continuous gray-level connection for each edge point, filter out the boundary pixel point set with missing connection, and generate a boundary closed pixel set.

[0025] S303: Call the boundary closed pixel set, calculate the ratio of the total number of pixels to the number of candidate pixels for the remaining pixels in the patch, determine whether the patch meets the density threshold condition, remove the patch in the region where the pixel distribution is lower than the density ratio requirement, and obtain the structure texture screening map.

[0026] As a further aspect of the present invention, the specific steps of S4 are as follows:

[0027] S401: Extract the coordinate positions of the remaining pixels in the structure texture screening image, divide the image into blocks according to the image area, extract the spatial position of each pixel in each block, calculate the spatial-grayscale composite distance value, establish a pairing relationship between pixels according to the distance threshold, and obtain pixel distance association pairs;

[0028] S402: Call the pixel distance association pair, compare the gray value difference of the paired pixels, make a judgment based on the gray value difference and the set gray value difference threshold, filter out pixel pairs with gray value abrupt changes, and retain only pixel combinations that meet the adjacency conditions to obtain effective gray value adjacency pairs.

[0029] S403: Based on the grayscale adjacency pairs, the pixels are used as nodes in the structural graph, and the adjacency relationships between pixels are constructed as edges in the graph. All nodes and edges in the entire graph are uniformly organized to establish a connectivity network structure and obtain a spatial connection graph structure object.

[0030] As a further aspect of the present invention, the method further includes:

[0031] S5: Based on the node positions and number of connections in the spatial connection graph structure object, set the initial pixel point set in the vegetation mixed boundary block, perform full-map aggregation path selection and distribution balance judgment, map the path results to the original image, and obtain the ground-based SAR high coherence point selection scheme.

[0032] As a further aspect of the present invention, the ground-based SAR high coherence point selection scheme includes a clustering path map, a node initial set, a distribution equilibrium map, and a clustering result map.

[0033] As a further aspect of the present invention, the specific steps of S5 are as follows:

[0034] S501: Call all graph node positions and connection counts in the spatial connection graph structure object, set the starting cell point set in the vegetation mixed boundary block, sort the nodes by connection count from small to large and select the starting point set as candidate nodes, and trace adjacent nodes in the graph structure according to the connection path length to generate a candidate aggregation path set.

[0035] S502: Based on the candidate aggregation path set, calculate the spatial distribution variance of all nodes in the path, and calculate the average spatial distance between nodes on the path and path nodes, determine whether the distribution density balance meets the set spatial distribution balance threshold condition, and obtain the spatially balanced path set.

[0036] S503: Call the spatially balanced path set, establish cluster numbers according to the pixel positions in the original image corresponding to the path nodes, classify the paths into categories, map the pixels in each path back to the corresponding positions in the original image, output the cluster number layer, and obtain the ground-based SAR high coherence point selection scheme.

[0037] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0038] In this invention, a multi-source fusion layer is constructed by combining amplitude and phase features to enhance the image discrimination dimension. The accuracy of region recognition is improved by extracting gray-level frequency ranges and generating Boolean masks. Local structural features are extracted using gray-level co-occurrence matrices, and abnormal pixels are removed by combining boundary coherence and density features. A graph structure relationship based on adjacency paths is established to form a pixel connection network. From this network, a set of pixels with aggregation and connectivity is selected to achieve high coherence point clustering output, thereby enhancing the completeness of structural expression in complex regions and the rationality of target point aggregation. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram of the steps of the present invention;

[0041] Figure 2 This is a detailed schematic diagram of S1 of the present invention;

[0042] Figure 3 This is a detailed schematic diagram of S2 of the present invention;

[0043] Figure 4 This is a detailed schematic diagram of S3 of the present invention;

[0044] Figure 5 This is a detailed schematic diagram of S4 of the present invention;

[0045] Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0046] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0047] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0048] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0049] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0050] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0051] Please see Figure 1 This invention provides an improved method for selecting high-coherence points in ground-based SAR, comprising the following steps:

[0052] S1: Extract the original image frames from the ground-based SAR amplitude sequence and phase sequence, calculate the mean and deviation values ​​for the amplitude images according to the map sheet, perform difference expansion and calculate the variance map between the phase frames, perform normalization processing on the amplitude and phase layers, and perform overlay operation according to the fusion logic rules set by the region label to output the joint index layer map.

[0053] S2: Based on the division of urban and vegetation mixed areas in the joint indicator layer map, the frequency of gray-level distribution of pixels in the area map is statistically analyzed and a gray-level histogram is constructed. The frequency gray-level interval is selected as the recognition range. The pixel value logic is performed according to the block to generate a Boolean mask map and output the area distribution mask map.

[0054] S3: Call the selected pixel patch in the regional distribution mask map, construct the gray-level co-occurrence matrix of the selected area in the original image, obtain the gray-level correlation distribution features of the patch, perform boundary closure detection and density ratio judgment on the same area patch, remove the discontinuous boundary and sparse candidate point areas, and output the structure texture screening map.

[0055] S4: Extract the coordinates of the remaining pixels in the structural texture filtering graph, establish distance relationships between the pixels and their neighboring pixels, compare the grayscale differences of pixels in the connection path and determine the adjacency rationality, combine the adjacency results into the relationships between nodes and edges in the structural graph, and output the spatial connection graph structural object.

[0056] S5: Call the positions and number of all graph nodes in the spatial connection graph structure object, set the starting pixel point set in the vegetation mixed boundary map block, perform clustering path selection and distribution balance judgment on the whole map structure, map the path results onto the original image and output clustering labels to obtain the ground-based SAR high coherence point selection scheme.

[0057] The joint indicator layer map includes amplitude statistical feature map, phase difference feature map, normalization processing map, and region fusion map; the regional distribution mask map includes urban area identification mask, vegetation mixed area mask, gray-level frequency screening map, and logical judgment result map; the structural texture screening map includes gray-level correlation map, boundary integrity map, density distribution map, and screening area map; the spatial connectivity map structure object includes pixel node set, adjacency relationship set, path gray-level difference map, and structural edge map; the ground-based SAR high coherence point selection scheme includes clustering path map, node starting set, distribution equilibrium map, and clustering result map.

[0058] Please see Figure 2 The specific steps of S1 are as follows:

[0059] S101: Extract the original image frames from the ground-based SAR amplitude sequence and phase sequence, overlay the gray values ​​of each pixel in the amplitude image by multiple frames according to the map sheet, calculate the mean and deviation based on the total number, cumulative value and sum of squares, and generate an amplitude gray value statistics layer.

[0060] To extract the original image frames from the ground-based SAR amplitude and phase sequences, SAR image data needs to be loaded frame by frame. The acquisition time and area corresponding to each frame must be confirmed. Map sheets covering the same geographical area are managed by unified numbering; for example, map sheet A001 contains 15 image frames. Then, a multi-frame overlay operation is performed on the grayscale value of each pixel in the amplitude image. Before the operation, pixel-level spatial alignment should be completed for all image frames within the map sheet to ensure that the (i, j) pixel position in each frame corresponds consistently across all frames. For a pixel position (200, 300) in map sheet A001, its grayscale values ​​in the 15 image frames are 65, 68, 70, 71, 72, 75, 73, 69, 68, 66, 67, 70, 71, 74, and 72, respectively. The number of frames for this pixel is 15, and the total grayscale value is 1051. The total value is 73753. The mean gray value of this pixel is calculated as 1051 divided by 15, which gives 70.07. Then, the gray value deviation is calculated. First, the difference between each gray value and the mean is calculated and squared to obtain a set of deviation squares. Then, these are summed and divided by the number of frames to obtain the average deviation square value. Finally, the square root is taken to obtain the deviation value. If some pixels in a certain image frame have missing data, for example, the corresponding pixels in the 9th and 11th frames have no value, the number of valid frames is corrected to 13. The total gray value and the sum of the squares need to be recalculated after removing invalid frame data. The gray value of each frame needs to be amplitude standardized to remove gray value shifts caused by non-target factors such as sensor gain. For example, the maximum gray value is set to 80, and all gray values ​​are scaled proportionally to between 0 and 1. Then, the above superposition and deviation calculation are performed to finally obtain an amplitude gray value statistics layer containing the gray value statistics of each pixel.

[0061] S102: Call the amplitude grayscale statistics layer, calculate the inter-frame phase difference value of the phase image frame according to the time sequence, perform linear expansion processing, and calculate the variance based on the set of expanded values ​​of each pixel to generate the phase expansion variance layer.

[0062] The amplitude grayscale statistics layer is invoked, and the phase image frames are processed frame by frame in the time series. Two consecutive frames are selected as frame pairs for calculation. Frame 1 and 2 constitute frame pair 1, frame 2 and 3 constitute frame pair 2, and so on. The phase value of the current frame and the next frame is extracted for each pixel position, and the phase difference is obtained by subtraction. If the phase value of a pixel in frame pair 1 is 3.1 and 3.3 respectively, the difference is 0.2. When there are periodic abrupt changes between adjacent frames, linear expansion processing is required to remap the phase value to a continuously changing curve. The conventional processing method is to accumulate the difference frame by frame starting from the first frame and perform trend fitting. If there is a jump exceeding π in frame pair 3, i.e., about 3.14, it can be determined that there is a phase entanglement phenomenon, and the jump threshold can be set to 2.5. All inter-frame phase... When the absolute value of the position difference exceeds the threshold, the difference is adjusted to be the interpolation prediction result of the phase change trend of the previous and next frames, so as to ensure that the phase trajectory of each frame after reconstruction is continuous and stable. Taking a certain pixel as an example, the phase unfolded value obtained after processing is 2.9, 3.1, 3.2, 3.2, 3.3, 3.4, 3.3, 3.5, 3.4, a total of 9 frames. The average value is calculated to be 3.26. The sum of the squares of the differences between each unfolded value and the average value and divided by 9 gives a variance of 0.036. A small variance indicates that the phase trajectory of the pixel is stable. Here, the reference threshold for judging whether the variance is stable is set to 0.05. If it is lower than this value, it is judged as a relatively stable pixel. If it is higher than this value, it is marked as an unstable pixel. Each pixel in the variance layer corresponds to a specific value, forming a phase unfolded variance layer.

[0063] S103: Normalize the amplitude grayscale statistics layer and the phase expansion variance layer, set the overlay rules according to the region label, perform weighted overlay on the normalized value of each pixel, and obtain the joint index layer map.

[0064] Normalization is performed on the amplitude grayscale statistics layer and the phase expansion variance layer. All pixel values ​​from both layers are read, and the global minimum and maximum values ​​are calculated for each region, serving as the lower and upper limits for normalization, respectively. For example, the maximum grayscale value of the amplitude layer is 78, and the minimum is 45. The maximum value of the phase variance layer is 0.12, and the minimum is 0.01. For each pixel, its grayscale value is subtracted from the minimum value and then divided by the difference between the maximum and minimum values. For example, if pixel A has a grayscale value of 63, the normalized value is (63-45) / (78-45) = 0.545. The corresponding phase variance value is 0.04, so the normalized value is (0.04-0.01) / (0.12-0.01) = 0.273. Then, based on the region labels, overlay values ​​are set... Weighting is applied as follows: If pixel A belongs to an urban area, the amplitude normalization value weight is set to 0.7, and the phase normalization value weight is set to 0.3. The weighted summation result is 0.7×0.545+0.3×0.273=0.4738. If pixel B belongs to a farmland area, the amplitude and phase normalization value weights are set to 0.4 and 0.6, respectively. Under the same grayscale and phase values, the weighted result is 0.4×0.545+0.6×0.273=0.3816. When setting weights, the influence of regional differences on the sensitivity of the index is taken into account. Generally, the weight range is determined to be between 0.3 and 0.7 through field sample testing. Finally, the weighted result of each pixel is output as a joint index layer map, and each pixel value reflects the degree of normalization attribute combination in the corresponding region.

[0065] Please see Figure 3 The specific steps of S2 are as follows:

[0066] S201: Based on the results of the urban area and mixed vegetation area map division in the joint index layer map, the gray values ​​of all pixels in the map are statistically analyzed, the correspondence between gray values ​​and occurrence frequency is established according to the pixel gray level, the gray frequency distribution of each map is obtained, and the gray frequency statistics of the map are generated.

[0067] Based on the urban and mixed vegetation area tile division results in the joint index layer map, it is necessary to read the tile boundary information in the layer map and establish a mapping table between tile numbers and layer pixel coordinates. For example, the urban area tile numbers are set as U01 and U02, and the mixed vegetation area tile numbers are V01 and V02. For each pixel in tile U01, grayscale values ​​are extracted row by row to construct a grayscale value set. The grayscale value range is an integer between 0 and 255. Each pixel is counted once, forming a grayscale frequency statistics table. For example, there are 3000 pixels in tile U01, 400 pixels have a grayscale value of 85, and 430 pixels have a grayscale value of 86. This method is used to progressively construct the correspondence between grayscale values ​​and their frequency of occurrence. Each grayscale value is used as a key, and its frequency in the map is used as the value, forming grayscale frequency histogram distribution data. During the statistical process, grayscale level units need to be set, such as 1 or 2 per level. That is, grayscale values ​​85 and 86 are set as a group or counted independently. Using 1 grayscale unit per level can obtain more refined distribution features. If the number of pixels in map V02 is 2500 and the grayscale values ​​are mainly distributed between 110 and 140, then the corresponding grayscale frequency is dense in this range. The frequency of each grayscale value in this range is counted, and finally a set of statistical pairs with grayscale value as key and frequency as value is formed. The statistics are performed independently for each map, without merging data across maps, and the grayscale frequency statistics results for each map are finally output.

[0068] S202: Based on the grayscale frequency statistics of the map, extract the grayscale value range in the middle segment of the cumulative frequency distribution, determine whether the frequency span of consecutive grayscale levels meets the threshold condition, obtain the upper and lower limits of grayscale values ​​that meet the condition, and generate the grayscale recognition interval of the map.

[0069] Based on the grayscale frequency statistics of the image blocks, the grayscale value and frequency pair sets constructed in each block are processed sequentially. The frequencies are accumulated in ascending order of grayscale values ​​to obtain a cumulative frequency sequence, and the total frequency value is calculated. For example, in image block U01, the total number of pixels is 3000. Sorting by cumulative frequency, the cumulative frequency of grayscale values ​​below 85 is 1000, accounting for approximately 33%; the cumulative frequency of grayscale values ​​between 86 and 120 is 1700, accounting for 56%; and the cumulative frequency of grayscale values ​​between 121 and 150 is 300, accounting for 10%. Therefore, the grayscale range with the cumulative frequency in the middle segment is determined to be 86 to 120. The frequency span of all consecutive grayscale levels within this range is then judged, i.e., whether the frequency change between consecutive grayscale levels meets a threshold condition. A frequency span threshold of 5% is set. If the frequency of a certain grayscale value is 200, and the frequency of the next grayscale value is 215, the frequency change is considered significant. The value is calculated as (215-200) / 3000, which is approximately 0.5%. This does not meet the 5% threshold requirement, so the pair of values ​​is excluded. If the grayscale value is 110, the corresponding frequency is 160, and the grayscale value is 111, the corresponding frequency is 220. The frequency change is 60 / 3000, which is 2%, and it still does not meet the threshold. This process continues until the frequency change between consecutive grayscale levels is greater than or equal to 150 / 3000=5%. For example, the frequency span between grayscale values ​​95 and 96 is 180 and 330, respectively, which corresponds to a change of 150 / 3000=5%, meeting the threshold requirement. Therefore, 95 and 96 are included in the target grayscale range. All grayscale value segments that meet the frequency span threshold requirement within the middle range are extracted. The minimum value is recorded as the lower limit of the grayscale value, and the maximum value is recorded as the upper limit of the grayscale value. Finally, the grayscale recognition range corresponding to the block is obtained. For example, the recognition range of block U01 is [95, 120].

[0070] S203: Call the grayscale recognition range of the map patch, make a logical judgment on whether the grayscale value of each pixel in the map patch falls into the corresponding recognition range, mark the judgment result as a binary Boolean value, and construct a binary logic map according to the map area to obtain the regional distribution mask map;

[0071] The grayscale recognition range of the image blocks is invoked, and a recognition operation is performed on each block. The upper and lower limits of the recognition range are read. For example, the recognition range of block U01 is [95, 120]. Then, all pixels in the block are traversed, and the grayscale value of each pixel is read and judged one by one. The judgment method is to check whether the grayscale value of the pixel is within the recognition range. If it falls within the range, it is judged as True and marked with a logical value of 1; otherwise, it is judged as False and marked with a logical value of 0. For example, the grayscale value of pixel A in block U01 is 102, which falls within the recognition range, so the judgment value is 1. The grayscale value of pixel B is 0. The value is 85, which is not in the recognition range, and the judgment value is 0. Each judgment operation involves only one numerical comparison and boundary judgment, without involving other statistical operations. All judgment results are recorded according to the corresponding position of the pixel in the map block. The recording result is a binary matrix with the same row and column dimensions as the original map block layer. The judgment result matrices of all map blocks are merged according to the map sheet number to generate the overall map sheet binary map. For each map block in the map sheet, the corresponding recognition range is called to perform logical judgment and marking. Finally, a complete binary logical map is constructed according to the map sheet number. After merging, the layer is identified as a regional distribution mask map.

[0072] Please see Figure 4 The specific steps of S3 are as follows:

[0073] S301: Call the selected pixel patch in the regional distribution mask map, obtain the pixel gray value combination for the corresponding patch in the original image, perform joint frequency statistics on all adjacent pixel gray value pairs in each patch, construct a co-occurrence matrix according to gray value pairs, statistically analyze the distribution relationship of gray value pairs, and obtain the gray value correlation distribution characteristics of the patch.

[0074] The process involves selecting a pixel patch in the region distribution mask image, reading all regions with a pixel value of 1 from the mask image, and extracting their corresponding patch number and pixel coordinate set. For example, patch number T001 contains 450 marked pixels. The grayscale image data of patch T001 is read from the original image. A grayscale value matrix is ​​constructed for all pixels in the patch. Each pixel is iterated through in the matrix to form grayscale pairs with its neighboring pixels. The adjacency relationship is determined using the 8-neighborhood principle, meaning each pixel has at most 8 neighboring pixels. For example, pixel (100, 150) has a grayscale value of 128, and its neighboring pixel (100, 151) has a grayscale value of 130, forming the grayscale pair (128, 130). The frequency of this grayscale pair's joint occurrence is recorded. This process continues to iterate through all pixels within the patch and counts the joint occurrence frequency of all grayscale pairs. The frequency of repeated grayscale pairs is accumulated. For example... The grayscale pair (128, 130) appeared 85 times, and the grayscale pair (130, 130) appeared 110 times. These grayscale pairs are used as row and column coordinates to construct a co-occurrence matrix. The row represents the first grayscale value, and the column represents the second grayscale value. Each element of the co-occurrence matrix records the frequency of occurrence of the corresponding grayscale pair. The co-occurrence matrix has a dimension of 256×256. If only the actual grayscale combinations are counted, a sparse matrix representation can be established. Then, the frequency values ​​of all elements in the co-occurrence matrix are traversed and sorted from high to low grayscale pair frequency. It can be observed that some grayscale pairs frequently appear in the patch, thereby identifying the strength of the grayscale value correlation in the patch. For example, if high-frequency grayscale pairs are mainly concentrated in the grayscale range of 120~135, it can be determined that there is a strong grayscale coupling relationship between the pixels inside the patch within this grayscale range. Finally, the grayscale correlation distribution characteristics of the patch are output.

[0075] S302: Based on the gray-level correlation distribution characteristics of the map patch, perform boundary closure judgment on the gray-level adjacency relationship of edge pixels within the map patch, detect whether there is a continuous gray-level connection for each edge point, filter out the boundary pixel point set with missing connection, and generate a boundary closed pixel set.

[0076] Based on the gray-level correlation distribution characteristics of the image patches, the co-occurrence matrix obtained in the previous step is read, and the gray-level connectivity of edge pixels is screened according to gray-level value combinations. First, all edge pixels in the image patch are identified. An edge pixel is a pixel that borders the mask boundary in its 8-neighborhood. For example, in image patch T001, edge pixel point B1 is located at coordinates (10, 100), and some of its neighboring pixels are located in areas with a mask value of 0. The gray-level values ​​of this point and its 8-neighborhood are extracted, and it is determined whether there is a continuous gray-level connection. That is, whether the frequency between the gray-level value of this point and the gray-level value of a certain neighboring point in the co-occurrence matrix is ​​greater than the connection threshold. The connection threshold is set to 10, that is, if the frequency of a gray-level pair in the co-occurrence matrix exceeds 10 times, it is considered that there is a connection. For example, the gray-level value of point B1 is 122, and the gray-level value of its neighboring point is 124. In the co-occurrence matrix, the gray-level pair (122, 124) has a frequency of 15. If this condition is met, the connection is considered established. If the frequency of the gray-level pair is lower than the threshold, the connection is considered missing. This judgment operation is performed on each edge pixel. If the connection frequency between the edge pixel and all pixels in its 8-neighborhood does not exceed the threshold, i.e., there is no continuous connection, the edge pixel is removed from the patch. The connection threshold used in the judgment can be set according to the overall gray-level distribution density of the patch. If the total number of pixels in the patch is 500 and the average frequency of co-occurring gray-level pairs is 12, then setting the connection threshold to 10 is reasonable. For medium-density patches, the connection threshold can be 8, and for high-density patches, the value can be increased to 15. The above operation is repeated for all edge pixels. Finally, edge pixels with at least one pair of continuous gray-level connections are retained, and the boundary closed pixel set of the patch is constructed.

[0077] S303: Call the boundary closed cell set, calculate the ratio of the total number of cells to the number of candidate cells for the remaining cells in the patch, determine whether the patch meets the density threshold condition, remove the patch in the region where the cell distribution is lower than the density ratio requirement, and obtain the structural texture screening map;

[0078] The boundary closed pixel set is invoked, and the density of the remaining unremoved pixel set in the tile is calculated. First, the total number of pixels in the tile, N_total, and the number of candidate pixels, N_candidate, within the boundary closed pixel set are counted. For example, if the original total number of pixels in tile T001 is 450, and 390 pixels are retained in the closed set, the density ratio is calculated as 390 divided by 450, yielding 0.867. This ratio is then compared with a preset density threshold of 0.85. This means that if the proportion of candidate pixels exceeds 85%, the tile is considered to meet the structural density requirements; if the ratio is lower than this threshold, it indicates that the internal structure of the tile is too sparse. The area that does not meet the requirements should be marked as a non-compliant area and removed from subsequent processing. The threshold should be set with reference to the degree of texture concentration in different structural areas. For urban areas with relatively concentrated structural textures, the density threshold can be set to 0.9, and for natural areas with slightly dispersed textures, it can be set to 0.8. The value range is generally controlled between 0.75 and 0.95. During the judgment process, the ratio result of each area should be directly compared with the corresponding threshold without further normalization or transformation. After traversing all areas, the areas that meet the density ratio requirements are recorded as retained areas, forming a structural texture screening map. Only areas with complete structure and qualified density are retained in the output layer.

[0079] Please see Figure 5 The specific steps of S4 are as follows:

[0080] S401: Extract the coordinates of the remaining pixels in the structural texture filter image, divide the image into blocks according to the image area, extract the spatial position of each pixel in each block, calculate the spatial-grayscale composite distance value, establish pairing relationships between pixels according to the distance threshold, and obtain pixel distance association pairs;

[0081] The specific formula for calculating the spatial-grayscale composite distance value is as follows:

[0082] ;

[0083] in, Indicates the first [unit] within the [block]. The pixel and the first Spatial-grayscale composite distance value between pixels, Indicates the first The row coordinates of each cell. Indicates the first The row coordinates of each cell. Indicates the first The column coordinates of each cell. Indicates the first The column coordinates of each cell. Indicates the first The grayscale value of each pixel Indicates the first The grayscale value of each pixel This represents an adjustment factor that converts grayscale differences into spatial pixel distances;

[0084] Parameter description and value acquisition process:

[0085] : These are the row and column coordinates of the pixel, read directly from the image index, in pixels. For example, pixel A is located at (42, 78), and pixel B is located at (46, 81).

[0086] : The gray value of the corresponding pixel, acquired through the image grayscale channel, with dimensionless units (8-bit image, range 0–255), such as A having a gray value of 138 and B having a gray value of 122;

[0087] : This is the conversion factor between grayscale difference and spatial distance. Assuming the patch size is 120×120 pixels, the maximum spatial distance is:

[0088] ;

[0089] The maximum difference in grayscale is 255, therefore:

[0090] ;

[0091] Substitute into the calculation:

[0092] Spatial distance item:

[0093] ;

[0094] Gray-scale difference:

[0095] ;

[0096] Total distance:

[0097] ;

[0098] Table 1 Example of pixel composite distance calculation

[0099]

[0100] As shown in Table 1, the composite distance between two pixels is 15.64 pixels.

[0101] The results show that in patch T201, although pixels A and B are only 5 pixels apart in space, the grayscale difference reaches 16, resulting in an additional distance of 10.64 after conversion, which increases the overall composite distance to 15.64. This reflects that the pixel pair lacks grayscale continuity under the condition of spatial proximity and should not be regarded as structural units in the same texture area.

[0102] The advantage of the formula lies in the introduction of a grayscale conversion factor. This enables a unified expression of grayscale and spatial attributes under different dimensions, so that the calculation of structural texture relationships no longer depends on subjective weight settings, thereby enhancing the accuracy of the subsequent adjacent boundary establishment process.

[0103] S402: Call the cell distance association pair, compare the gray value difference of the paired cells, make a judgment based on the gray value difference and the set gray value difference threshold, filter out cell pairs with gray value abrupt changes, and only retain cell combinations that meet the adjacency conditions to obtain effective gray value adjacency pairs.

[0104] After calling the pixel distance association pair, the grayscale value of each pair of pixels in the original image is read, and a pairwise comparison operation is performed to obtain the grayscale difference of each pair of pixels. This difference is then compared with a set grayscale difference threshold. If the grayscale difference between pixels is less than or equal to the threshold, the pixel pair is retained, considered to have grayscale continuity; otherwise, it is discarded. The grayscale difference threshold is set to 12. This threshold is set based on the image's grayscale dynamic range and local texture details. When the grayscale value range is 0 to 255, 12 corresponds to a grayscale proportion of approximately 4.7%, used to determine whether a local area maintains a consistent texture. This is suitable for structural areas with medium contrast features, such as buildings and ground features. For example... For example, the gray levels of the pixel pair (105, 205) and (106, 206) are 132 and 140 respectively, with a gray level difference of 8, which is less than the threshold of 12, so the pair is retained. However, the gray levels of the pixel pair (110, 210) and (111, 211) are 130 and 147 respectively, with a gray level difference of 17, which exceeds the threshold, so they are removed from the set. In the entire patch G101, there are a total of 1180 pixel pairs. After judging the gray level difference, the remaining valid pairs are 980 pairs. The remaining 200 pairs are removed due to drastic gray level changes. All the retained pixel pairs are marked as the set of valid gray level adjacency pairs, which is used to construct the graph structure.

[0105] S403: Based on the effective gray-level adjacency pairs, the cells are used as nodes in the structure graph, and the adjacency relationships between the cells are constructed as edges in the graph. All nodes and edges in the entire graph are uniformly organized to establish a connectivity network structure and obtain the spatial connection graph structure object.

[0106] Based on valid gray-level adjacency pairs, each cell is set as a node element in the structural graph, and each retained gray-level adjacency pair is used as an edge in the graph structure. Structural organization operations are performed. When constructing the graph for block G101, a node numbering table is first generated. For example, cell (105, 205) is numbered G101_001, and (106, 206) is numbered G101_002. When valid gray-level adjacency relationships exist, an edge connection is established between the two nodes. Each edge records the connection relationship without adding weight information. In block G101, 980 valid adjacency pairs will... The graph is constructed with 980 edges. Approximately 460 related pixels participate in at least one connection. The number of connections for each pixel can be obtained by counting the number of times it appears in the adjacent pairs. For example, pixel G101_012 participates in 5 edge connections, meaning it has spatial adjacency relationships with 5 pixels with the same gray level. After organizing all nodes and edges of the image patch G101 to form a local graph structure, the graph structures of multiple patches are merged according to the image sheet number to construct the spatial connectivity graph structure object corresponding to the entire image, which is used to express the connectivity state and spatial proximity features between structural pixels.

[0107] Please see Figure 6 The specific steps of S5 are as follows:

[0108] S501: Call all graph node positions and connection counts in the spatial connection graph structure object, set the starting cell point set in the vegetation mixed boundary tile, sort the nodes by connection count from smallest to largest and select the starting point set as candidate nodes, and trace adjacent nodes in the graph structure according to the connection path length to generate a candidate cluster path set.

[0109] The system retrieves the positions and connection counts of all graph nodes in the spatial connectivity graph structure object. For each graph node, it records its corresponding cell coordinates and the number of connections it has with other nodes in the graph structure. For example, graph node A has coordinates (210, 310) and a connection count of 2, while graph node B has coordinates (211, 309) and a connection count of 5. In the vegetation mixed boundary tile, a starting set of pixels is set. All nodes within the tile boundary are selected as the initial filtering range, and then these nodes are sorted from smallest to largest by connection count. The nodes with the fewest connections form a candidate starting point set. For example, the candidate starting point set can be set to the top 5% of the nodes. If there are 200 boundary nodes in the tile, then... The 10 nodes with the fewest connections are used as the starting point set, for example, the node set is {A, B, C, D, E}. Then, in the graph structure, starting from each starting node, its adjacent nodes are traced in sequence. When constructing the path, a connection number priority strategy is adopted, that is, the path is extended to the adjacent nodes with fewer connections first. The length of the tracing path is limited to a preset maximum path length, for example, 20 nodes, to prevent the path from being too long and resulting in a discrete distribution. After each tracing step, the node number and its spatial position are recorded to form a candidate path. This operation is repeated with all starting nodes as the starting point, and finally a set of candidate clustered paths is generated, for example, a total of 58 paths are generated. Each path records the node sequence and its spatial coordinate sequence.

[0110] S502: Based on the candidate clustering path set, calculate the spatial distribution variance of all nodes in the path, and calculate the average spatial distance between nodes on the path, determine whether the distribution density balance meets the set spatial distribution balance threshold condition, and obtain the spatially balanced path set.

[0111] Based on the candidate clustering path set, the spatial coordinates of all nodes in each path are extracted, and the spatial distribution variance of the path is calculated. During the statistical process, the standard deviations of the horizontal and vertical coordinates are calculated separately to assess the spatial diffusion of the path. Simultaneously, the average distance from any node in the path to all other nodes is calculated to obtain the average distance between each node and other nodes in the path. The average of all average distance values ​​is then taken as the overall average distance index of the path. For example, if a path P contains 10 nodes, its X-coordinate standard deviation is 3.2, its Y-coordinate standard deviation is 2.8, and the combined spatial distribution variance is... 3.0, with an average spacing of 4.5 pixels on the path, compares the spatial distribution variance of the path with the set spatial distribution balance threshold to determine whether the path's balance meets the conditions. The threshold is set with reference to image resolution and structural texture features. For large spatial scale tile regions, the variance threshold can be set to no more than 5.0 and the average spacing to no more than 6.0. Paths that meet both conditions are determined to be balanced paths; otherwise, they are considered discrete paths. Path numbers that meet the requirements are selected from the path set. For example, paths P01, P03, and P06 meet the conditions and form a set of spatially balanced paths for subsequent clustering and mapping operations.

[0112] S503: Call the spatial equilibrium path set, establish cluster numbers according to the pixel positions in the original image corresponding to the path nodes, divide the categories according to the path and map the pixels in each path back to the corresponding positions in the original image, output the cluster number layer, and obtain the ground-based SAR high coherence point selection scheme.

[0113] The spatially balanced path set is invoked, and clustering numbering is performed on each path in the set. Path numbers are assigned sequentially starting from 1, and each node in each path is treated as a clustering unit. The original image pixel coordinates corresponding to each node are read, and the original image is mapped in the output layer using clustering numbers. Each pixel is assigned a unique clustering ID value according to its path number. For example, if the pixel coordinates corresponding to the node in path P01 are {(210, 310), (211, 309), (212, 308)}, the corresponding values ​​for these three positions are all set to 1 in the output layer. The pixel in path P02 is set to 2, and so on until all paths are mapped. After numbering and marking on the entire image, a clustering number layer is formed. In this layer, each non-zero pixel corresponds to a clustered path region, and pixels that do not participate in path clustering are set to zero. The layer structure maintains pixel consistency with the original image, ultimately forming a ground-based SAR high coherence point selection scheme with pixel-level path aggregation identification.

[0114] 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 variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An improved method for selecting high-coherence points in ground-based SAR, characterized in that, Includes the following steps: S1: Extract the original image frames from the ground-based SAR amplitude and phase sequences. Calculate the mean and deviation of the amplitude images according to the map sheet. Perform difference expansion between phase frames and calculate the variance map. After normalization, overlay according to the regional label fusion logic to output the joint index layer map. S2: Divide the urban area and mixed vegetation area into blocks in the joint index layer map, construct a gray level histogram by statistically analyzing the gray level distribution frequency of pixels in the block, select the frequency gray level interval as the recognition range, logically judge the pixel value of each block to generate a Boolean mask map, and output the regional distribution mask map. S3: Based on the selected patch in the regional distribution mask, extract the original image to construct a gray-level co-occurrence matrix, obtain the gray-level correlation features of the patch, detect the boundary closure and density ratio, remove discontinuous and sparse regions of the boundary, and output the structural texture screening map. S4: Extract cell coordinates from the structural texture filtering map, establish the distance relationship between neighboring cells, compare the gray level difference of the path to determine the adjacency rationality, construct the relationship between nodes and edges of the structural graph, and output the spatial connection graph structural object; The specific steps of S1 are as follows: S101: Extract the original image frames from the ground-based SAR amplitude sequence and phase sequence, overlay the gray values ​​of each pixel in the amplitude image by multiple frames according to the map sheet, calculate the mean and deviation based on the total number, cumulative value and sum of squares, and generate an amplitude gray value statistics layer. S102: Call the amplitude grayscale statistics layer, calculate the inter-frame phase difference value of the phase image frame according to the time sequence, perform linear expansion processing, and calculate the variance based on the set of expanded values ​​of each pixel to generate a phase expansion variance layer. S103: Normalize the amplitude grayscale statistics layer and the phase expansion variance layer, set the superposition rules according to the region label, perform weighted superposition on the normalized value of each pixel, and obtain the joint index layer map. The specific steps of S2 are as follows: S201: Based on the urban area and vegetation mixed area block division results in the joint index layer map, the gray values ​​of all pixels in the block are statistically analyzed, the correspondence between gray values ​​and occurrence frequency is established according to the pixel gray level, the gray frequency distribution of each block is obtained, and the gray frequency statistics of the block are generated. S202: Based on the grayscale frequency statistics of the patch, extract the grayscale value range in the middle segment of the cumulative frequency distribution, determine whether the frequency span of consecutive grayscale levels meets the threshold condition according to the frequency, obtain the upper and lower limits of grayscale values ​​that meet the condition, and generate the patch grayscale recognition interval. S203: Call the grayscale recognition interval of the image block, make a logical judgment on whether the grayscale value of each pixel in the image block falls into the corresponding recognition interval, mark the judgment result as a binary Boolean value, and construct a binary logic diagram according to the image area to obtain the regional distribution mask map; The specific steps for S3 are as follows: S301: Call the selected pixel patch in the regional distribution mask image, obtain the pixel gray value combination for the corresponding patch in the original image, perform joint frequency statistics on all adjacent pixel gray value pairs in each patch, construct a co-occurrence matrix according to gray value pairs, statistically analyze the distribution relationship of gray value pairs, and obtain the gray value correlation distribution characteristics of the patch. S302: Based on the gray-level correlation distribution characteristics of the image patch, perform boundary closure judgment on the gray-level adjacency relationship of edge pixels within the image patch, detect whether there is a continuous gray-level connection for each edge point, filter out the boundary pixel point set with missing connection, and generate a boundary closed pixel set. S303: Call the boundary closed pixel set, calculate the ratio of the total number of pixels to the number of candidate pixels for the remaining pixels in the patch, determine whether the patch meets the density threshold condition, remove the patch in the region where the pixel distribution is lower than the density ratio requirement, and obtain the structural texture screening map; The specific steps of S4 are as follows: S401: Extract the coordinate positions of the remaining pixels in the structure texture screening image, divide the image into blocks according to the image area, extract the spatial position of each pixel in each block, calculate the spatial-grayscale composite distance value, establish a pairing relationship between pixels according to the distance threshold, and obtain pixel distance association pairs; S402: Call the pixel distance association pair, compare the gray value difference of the paired pixels, make a judgment based on the gray value difference and the set gray value difference threshold, filter out pixel pairs with gray value abrupt changes, and retain only pixel combinations that meet the adjacency conditions to obtain effective gray value adjacency pairs. S403: Based on the gray-level adjacency pairs, the pixels are used as nodes in the structure graph, and the adjacency relationships between pixels are constructed as edges in the graph. All nodes and edges in the entire graph are uniformly organized to establish a connectivity network structure and obtain a spatial connection graph structure object. The method further includes: S5: Based on the node positions and number of connections in the spatial connection graph structure object, set the initial pixel point set in the vegetation mixed boundary block, perform full-map aggregation path selection and distribution balance judgment, map the path results to the original image, and obtain the ground-based SAR high coherence point selection scheme.

2. The improved ground-based SAR high coherence point selection method according to claim 1, characterized in that, The joint index layer map includes an amplitude statistical feature map, a phase difference feature map, a normalization processing map, and a region fusion map. The region distribution mask map includes an urban area identification mask, a vegetation mixed area mask, a grayscale frequency filtering map, and a logical judgment result map. The structure texture filtering map includes a grayscale correlation map, a boundary integrity map, a density distribution map, and a filtering area map. The spatial connection map structure object includes a pixel node set, an adjacency relationship set, a path grayscale difference map, and a structure edge map.

3. The improved ground-based SAR high coherence point selection method according to claim 1, characterized in that, The definition of the frequency grayscale range refers to the continuous grayscale value range that appears frequently in the grayscale histogram of the image patch and represents the grayscale distribution of the pixel, and is the criterion for judging the pixel value of the identification area. The definition of the discontinuous and sparse regions of the boundary refers to the part of the verification area that has a discontinuous edge contour in the structural texture map with a pixel density lower than a set threshold, and which does not have completeness and representativeness. The definition of establishing the distance association between neighboring pixels refers to determining the rationality of the connection by calculating the spatial distance and path grayscale changes between pixels, and constructing a node-edge network that reflects the spatial topological relationship.

4. The improved ground-based SAR high coherence point selection method according to claim 3, characterized in that, The ground-based SAR high coherence point selection scheme includes a clustering path map, a node initial set, a distribution equilibrium map, and a clustering result map.

5. The improved ground-based SAR high coherence point selection method according to claim 4, characterized in that, The specific steps of S5 are as follows: S501: Call all graph node positions and connection counts in the spatial connection graph structure object, set the starting cell point set in the vegetation mixed boundary block, sort the nodes by connection count from small to large and select the starting point set as candidate nodes, and trace adjacent nodes in the graph structure according to the connection path length to generate a candidate aggregation path set. S502: Based on the candidate aggregation path set, calculate the spatial distribution variance of all nodes in the path, and calculate the average spatial distance between nodes on the path and path nodes, determine whether the distribution density balance meets the set spatial distribution balance threshold condition, and obtain the spatially balanced path set. S503: Call the spatially balanced path set, establish cluster numbers according to the pixel positions in the original image corresponding to the path nodes, classify the paths into categories, map the pixels in each path back to the corresponding positions in the original image, output the cluster number layer, and obtain the ground-based SAR high coherence point selection scheme.

Images