Polarization SAR change detection method, device, equipment and medium

Abstract

The invention discloses a polarimetric SAR change detection method, device and equipment and a medium, and relates to the technical field of radar image processing. The method comprises the following steps: acquiring a time sequence PolSAR image, and calculating similarity measurement of a multi-temporal covariance matrix by adopting JBLD divergence; calculating a time sequence edge intensity graph based on similarity measurement; initializing a clustering center by using a time sequence edge intensity graph, introducing a dynamic edge constraint mechanism in an iteration process to inhibit super-pixels from crossing image edges, and outputting a super-pixel segmentation result; constructing image topological representation fusing time sequence feature similarity, spatial adjacency and cross-time phase cross feature similarity; and constructing an energy function containing node cost and edge cost, and solving an energy minimization problem through quadratic pseudo Boolean optimization to obtain a change detection graph. According to the method, errors caused by discontinuous regions and fuzzy boundaries in super-pixel segmentation can be avoided, and high robustness is shown in natural ground features and complex urban building scenes.

Description

Technical Field

[0001] The present invention relates to the field of radar image processing technology, and in particular to a polarimetric SAR change detection method, device, equipment and medium. Background Art

[0002] Synthetic Aperture Radar (SAR) can achieve high-resolution earth observation in all weather and all day by actively emitting microwave signals and receiving backscattered waves from ground objects. It is widely used in terrain mapping, disaster monitoring and other fields. Traditional SAR obtains target scattering intensity information through a single polarization channel, but its ability to analyze complex ground objects is limited. Polarimetric Synthetic Aperture Radar (PolSAR) further expands on this basis and obtains the scattering matrix information of the target through full polarization observation. It can more comprehensively analyze the geometric structure, dielectric constant and scattering characteristics of ground objects, significantly improving the accuracy of ground object classification and dynamic monitoring. Change detection, as an important task of PolSAR image analysis, aims to identify surface changes that occur in the same area at different times, but its performance is severely restricted by factors such as high-dimensional polarization scattering characteristics of data, coherent speckle noise, and multi-phase radiation differences.

[0003] Existing PolSAR image change detection methods are mainly divided into two categories: pixel-level methods and region-level methods. Pixel-level methods (such as methods based on statistical information theory and hypothesis testing theory) directly compare the polarization covariance matrix differences of single pixels. Although they can capture subtle changes, they are highly sensitive to noise, and the detection results are prone to false alarms and missed detections. In addition, they lack spatial consistency constraints, resulting in fragmentation of the change area. The regional level method aggregates pixels into superpixel units and uses context information to suppress noise interference, but it still faces the following bottlenecks: 1. Insufficient fusion of temporal information: The existing temporal superpixel generation algorithm fails to effectively capture the joint statistical laws of multi-phase scattering characteristics, which easily leads to cross-phase boundary misalignment in gradual or weak scattering change areas, weakening segmentation consistency; 2. Lack of edge constraint mechanism: The existing superpixel method only uses edge information to initialize the cluster center, and does not dynamically suppress cross-edge misclassification during the iteration process, resulting in superpixels crossing the boundaries of real objects and destroying the integrity of the object target structure; 3. Inefficient use of topological structure information: The regional comparison-based method relies on direct feature differences and lacks modeling of image space and feature topological relationships, resulting in sensitivity to multi-phase radiation differences and insufficient detection capabilities for weak scattering change areas. In addition, the existing supervised deep learning-based methods are limited by the representativeness and coverage of training samples in complex scenes (such as urban building changes), and the generalization performance is significantly reduced. Summary of the invention

[0004] To solve the above technical problems, the present invention proposes a polarimetric SAR change detection method, device, equipment and medium. The superpixel segmentation of temporal PolSAR images based on edge constraints highlights the differences in the significant change phases by defining a new temporal polarimetric similarity metric, and suppresses the problem of inaccurate segmentation caused by superpixels crossing the image edges by introducing an edge constraint mechanism, significantly improving the fitting accuracy of superpixels to the actual ground object boundaries. The PolSAR image change detection based on graph structure energy optimization takes superpixels as the basic processing unit to construct an image topological representation that fuses temporal feature similarity and spatial adjacency. Then, the discriminant ability of weak scattering change regions is enhanced through cross-temporal cross-feature similarity measurement, and noise suppression and integrity preservation of change regions are achieved by combining with a global energy model.

[0005] To achieve the above object, the technical solution of the present invention is as follows: A polarimetric SAR change detection method, comprising the following steps: Obtain temporal PolSAR images, and calculate the similarity metric of multi-temporal covariance matrices using the JBLD divergence; calculate a temporal edge intensity map based on the similarity metric to obtain a temporal edge intensity map; initialize the clustering center using the temporal edge intensity map and introduce a dynamic edge constraint mechanism during the iteration process to suppress superpixels from crossing the image edges, and output the superpixel segmentation result; Based on the superpixel segmentation result, construct an image topological representation that fuses temporal feature similarity, spatial adjacency, and cross-temporal cross-feature similarity; based on the image topological representation, construct an energy function including node cost and edge cost, and solve the energy minimization problem through quadratic pseudo-Boolean optimization to obtain the change detection map of the dual-temporal PolSAR images.

[0006] Preferably, calculating the similarity metric of multi-temporal covariance matrices using the JBLD divergence includes the following steps: The functional expression of the JBLD divergence is: (1) Where C 1, C 2 represent two polarimetric covariance matrices, | | represents the determinant of the matrix, and log(·) represents the natural logarithm; Based on the statistical independence characteristics and the maximum value criterion between multi-temporal PolSAR images, the constructed temporal distance metric is as follows: (2) Where C i,j represents the covariance matrix of the i th phase and the j th region, Z 1 and Z 2 represent the temporal joint covariance matrices of two regions,N Indicates the number of time phases.

[0007] Preferably, the temporal edge intensity is calculated based on a similarity metric to obtain a temporal edge intensity map, including the following steps: Higher weights are assigned to pixels closer to the center position through a Gaussian-type filter, and the horizontal Gaussian window function expression is: (3) where ( x , y ) are relative coordinate values, and the weight change speeds of the window in two directions are controlled by σ x and σ y respectively; For a single-time-phase PolSAR image, the average covariance matrix of the regions on both sides of the center line is calculated as: (4) where θ f is the direction angle of the center line, and , Ω k , k = 1, 2 represent the sets of pixels in the regions on both sides of the center line; Combining Equation (2) and Equation (4), the temporal edge intensity map is obtained, and the expression is: (5) where and represent the bilateral average covariance matrix calculated using Equation (4) in the i th time phase.

[0008] Preferably, a dynamic edge constraint mechanism is introduced during the iteration process to suppress superpixels from crossing the image edge, including the following steps: Based on the temporal edge intensity map, an edge constraint term is determined, and the expression is: (6) where EDGE norm is the normalized edge intensity map, line ( i, j ) represents the line segment between any two pixels i and j , p represents the pixels on the line segment; Similarity metric for SLIC clustering of temporal PolSAR images D SLIC Generally, it is divided into three parts: feature similarity, edge constraint term, and spatial similarity, which are defined as follows: (7) where α andβ They are the coefficients of the edge constraint term and the spatial similarity weight respectively, S is the grid sampling step calculated according to the number of superpixels, D P is the maximum absolute percentage power similarity, D S is the spatial similarity.

[0009] Preferably, it further includes a post-processing optimization process, including the following steps: Merge isolated superpixel blocks with an area smaller than a preset threshold through connected region analysis to eliminate over-segmentation noise; Use morphological closing operation to smooth the superpixel edges and fill the tiny breaks caused by local similarity fluctuations.

[0010] Preferably, based on the superpixel segmentation result, construct an image topological representation that fuses temporal feature similarity, spatial adjacency, and cross-temporal cross-feature similarity, including the following steps: Use the polarimetric target decomposition method to extract the polarimetric scattering features of the PolSAR image and perform global normalization processing. Take the average polarimetric scattering feature of the superpixel block as the feature vector of the node, and the nodes of the constructed graph are: (10) Among them, the node , and respectively represent the average polarimetric scattering features of the i th superpixel in two time phases, N S represents the number of superpixels; Construct the edges of the graph according to the spatial similarity and feature similarity ε , and the expression is: (11) Among them, and are the edges constructed based on feature similarity in two time phases respectively, ε S is the edge constructed based on spatial similarity, and the definition is as follows: (12) (13) (15) Among them, represents t the K-nearest neighbor set of the feature similarity of the i th superpixel in the image at time phase 1; d s ( V i,V j ) represents the spatial distance between the geometric centers of two superpixels.

[0011] Preferably, the expression of the energy function is: (16) Wherein, θ const is a fixed offset value in the energy function, set to 0, θ i ( L i ) represents that the node V i is in the label L i at this time, the cost, θ ij ( L i ,L j ) represents that the superpixel pair with feature similarity or spatial proximity is in the label ( L i ,L j ) at this time, the cost.

[0012] Based on the above, the present invention also discloses a polarimetric SAR change detection device, including: A superpixel segmentation module, configured to obtain a time-series PolSAR image, and use the JBLD divergence to measure the similarity of multi-temporal covariance matrices; calculate a time-series edge intensity map based on the similarity measurement to obtain a time-series edge intensity map; initialize the clustering center using the time-series edge intensity map and introduce a dynamic edge constraint mechanism during the iteration process to suppress the superpixels from crossing the image edge, and output the superpixel segmentation result; A change detection module, configured to construct an image topology representation that fuses time-series feature similarity, spatial adjacency, and cross-temporal cross-feature similarity based on the superpixel segmentation result; construct an energy function including node cost and edge cost based on the image topology representation, and solve the energy minimization problem through quadratic pseudo-Boolean optimization to obtain a change detection map of the dual-temporal PolSAR image.

[0013] Based on the above, the present invention also discloses a computer device, including: a memory for storing a computer program; a processor for implementing the method as described in any one of the above when executing the computer program.

[0014] Based on the above, the present invention also discloses a readable storage medium, on which a computer program is stored, and the computer program realizes the method as described in any one of the above when executed by a processor.

[0015] Based on the above technical solution, the beneficial effects of the present invention are as follows: 1) A Jensen-Bregman Log Det (JBLD) divergence similarity measure based on the maximum value criterion in the present invention. This measure suppresses the noise interference of weakly fluctuating phases by adaptively focusing on the difference information of significantly changing phases, thereby significantly improving the accuracy of temporal segmentation. 2) A dynamic edge constraint mechanism in the present invention embeds the normalized edge intensity map into the similarity measure of Simple Linear Iterative Clustering (SLIC). By penalizing the pixel assignment across strong edges, it ensures that the superpixels strictly fit the true object boundaries, effectively maintaining the continuity and structural integrity of the object contours. 3) The present invention constructs a graph structure representation that fuses temporal feature similarity and spatial adjacency with superpixels as nodes. This structure representation innovatively introduces the cross-temporal cross-feature similarity measure. By jointly analyzing the topological correlation between cross-temporal nodes, it significantly enhances the discriminative ability for weak scattering changes and shows strong robustness in both natural object and complex urban building scenes. 4) The present invention provides a change detection framework based on graph structure energy optimization. This detection framework constructs an energy function that includes node cost (prior penalty term based on change sparsity) and edge cost (state consistency constraint based on topological correlation). It uses the Quadratic Pseudo-Boolean Optimization (QPBO) algorithm to solve the energy minimization problem and achieves efficient global convergence through the maximum flow / minimum cut theory, avoiding the problems of discontinuous regions and blurred boundaries caused by traditional threshold segmentation methods. Description of the Drawings

[0016] Figure 1 is a flowchart of a polarimetric SAR change detection method in an embodiment; Figure 2 is a schematic diagram of the feature similarity between superpixels in a polarimetric SAR change detection method in an embodiment. Detailed Embodiments

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention.

[0018] As Figure 1 shown, this embodiment provides a polarimetric SAR change detection method, which includes two parts: superpixel segmentation of temporal PolSAR images based on edge constraints and PolSAR image change detection based on graph structure energy optimization. The specific description is as follows: 1. Superpixel Segmentation of Temporal PolSAR Images Based on Edge Constraints 1.1. Calculate the similarity measure of multi-temporal covariance matrices using JBLD divergence (the similarity measure is applied in the calculation of the edge map and the SLIC clustering process). While significantly improving the computational efficiency, compared with the Symmetric Revised Wishart distance (SRW) measure, JBLD divergence maintains a comparable performance level in terms of accuracy and robustness, and it is defined as: (1) Where, C 1, C 2 represent two polarization covariance matrices, | | represents the determinant of the matrix, and log(·) represents the natural logarithm. Based on the statistical independence characteristics and the maximum value criterion between multi-temporal PolSAR images, the temporal distance measure constructed in this embodiment is as follows: (2) Where, C i,j represents the covariance matrix of the i -th time phase and the j -th region, Z 1 and Z 2 represent the temporal joint covariance matrices of two regions, N represents the number of time phases.

[0019] 1.2. Calculate the temporal edge intensity map based on a Gaussian-type filter In the edge detection framework based on a bilateral window filter, compared with a rectangular window, a Gaussian window can avoid false local maxima caused by strong noise, so it can generate more robust edges. The Gaussian-type filter assigns higher weights to pixels closer to the center position, and the horizontal Gaussian window function is defined as: (3) Where, ([[]] x , y ) are relative coordinate values, and the weight change speeds of the window in two directions are controlled by σ x and σ y respectively. For a single-temporal PolSAR image, the average covariance matrix of the regions on both sides of the center line is calculated by the following formula: (4) Where, θ f is the direction angle of the center line, and . Ω k , k= 1, 2 represents the set of pixels in the regions on both sides of the center line. Combining Equation (2) and Equation (4), the formula for calculating the edge intensity of the temporal PolSAR image is finally obtained as follows: (5) Wherein, and represent the bilateral average covariance matrix calculated using Equation (4) at the i -th time phase. By traversing the entire image using Equation (5), the temporal edge intensity map EDGE can be obtained.

[0020] 1.3. Initializing the superpixel clustering center The initial clustering centers are evenly spread across the entire image with a regular grid sampling step S , and then the clustering centers are moved to the lowest edge intensity at the 3×3 neighborhood in the temporal edge intensity map EDGE.

[0021] 1.4. K-means clustering based on the dynamic edge constraint mechanism This embodiment proposes a dynamic edge constraint mechanism, which integrates edge information into the SLIC clustering similarity metric, establishes an edge-guided similarity metric criterion, enables the generated superpixels to accurately fit the image edge features, and thus significantly improves the segmentation accuracy of potential targets. The definition of the edge constraint term is as follows: (6) Wherein, EDGE norm is the normalized edge intensity map, line ( i, j ) represents the line segment between any two pixels i and j , and p represents the pixels on this line segment. If two pixels do not cross the image edge, the edge constraint term will be lower; conversely, the edge constraint term between two pixels that cross the image edge is higher.

[0022] Using this property, the SLIC clustering similarity metric D SLIC of the temporal PolSAR image is generally divided into three parts: feature similarity, edge constraint term, and spatial similarity, and their definitions are as follows: (7) Wherein, α and β are the coefficients of the edge constraint term and the spatial similarity weight respectively, S is the grid sampling step calculated according to the number of superpixels. The feature similarity is dominated by the similarity metric defined in Equation (2), which can maximize the use of the difference information of the time phase with the largest fluctuation while improving the calculation efficiency, D P is the maximum absolute percentage power similarity, defined as: (8) Among them, P i,1 and P i,2 respectively represent the total scattering power of two points in the i th time phase. The value range of D P is [0, 1], and (1 + D P ) provides a maximum gain of twice for the feature similarity. Finally, spatial similarity is used to improve the compactness of the generated superpixels: (9) Among them, ([[]] x 1, y 1) and ([[]] x 2, y 2) represent the spatial coordinates of two points.

[0023] 1.5. Post - processing optimization After completing the edge - constraint - based SLIC clustering, post - processing optimization is performed on the initial superpixel segmentation result. First, isolated superpixel blocks with areas smaller than a preset threshold are merged through connected - component analysis to eliminate over - segmentation noise; second, morphological closing operations are used to smooth the superpixel edges and fill in the tiny breaks caused by local similarity fluctuations.

[0024] 2. PolSAR image change detection based on graph - structure energy optimization 2.1. Superpixel feature extraction To construct a graph G = { v , ε} that represents the topological structure of the PolSAR image, first, the polarization scattering features of the PolSAR image are extracted using the polarization target decomposition method F (such as obtaining F using the Yamaguchi decomposition method), and global normalization is performed. Then, the average polarization scattering feature of the superpixel block is used as the feature vector of the node. Furthermore, the nodes of the graph are constructed as: (10) Among them, , and respectively represent the average polarization scattering features of the i th superpixel in two time phases. N S represents the number of superpixels, and this feature will be used to refer to its corresponding superpixel later.

[0025] 2.2. Topological adjacency relationship construction The topological adjacency relationship of the image is also the graphG Edges ε Constructed based on spatial similarity and feature similarity: (11) Among them, and Are respectively the edges constructed based on feature similarity at two time phases ε , defined as follows: (12) (13) Among them, Represents t At time phase 1, the K-nearest neighbor set of the feature similarity of the i th superpixel in the image. Define the feature similarity between any two superpixels and as: (14) Among them, Represents the division of the corresponding positions of the vectors. For t The i th superpixel at time 1, use equation (14) to calculate the feature similarity with t Other superpixels at time 1, sort them, and take the first Minimum values to obtain . This will not be elaborated here t For the case of time 2. ε S Is the edge constructed based on spatial similarity, defined as follows: (15) Among them, d s ( V i ,V j ) represents the spatial distance between the geometric centers of two superpixels, S Is the sampling step of the superpixel grid sampling step.

[0026] 2.3. Construction of the energy function Based on the topological structure characterization of the dual-temporal PolSAR image, the PolSAR image change detection problem is transformed into an energy minimization problem, and its energy function is: (16) Among them, θ const Is the fixed offset value in the energy function, set to 0, θ i ( L i ) represents the node V i Is in the labelL i The cost at θ ij ( L i ,L j ) indicates that the superpixel pairs (i.e., the edges of the graph) with feature similarity or spatial proximity are at the label ( L i ,L j ) The cost at. Therefore, the key to the change detection method based on energy minimization is the design of the node cost and edge cost functions.

[0027] For this reason, as Figure 2 shown, in this embodiment, any two nodes are first defined as and . There are 3 categories and 6 feature similarities among the four superpixels, namely the intra-node feature similarity and , the inter-node feature similarity and , and the cross feature similarity and .

[0028] According to these 6 feature similarities, four edge cost functions are constructed as follows: (17) (18) (19) (20) Based on the fact that only a small part of the area changes and most of the area remains unchanged in the actual change detection problem, a sparse penalty term based on the change prior is designed, that is, the node cost function is: (21) In order to balance the influence of the node cost function and the edge cost function on the minimization process, this embodiment normalizes the four edge cost functions to obtain the normalized edge cost function : (22) Among them, indicates the gain coefficient when the state of the edge is ( L i ,L j ). Through the normalization operation, it can be ensured that the cost of the edge and the cost of the node are at the same order of magnitude.

[0029] 2.4. Energy optimization solution The QPBO method is adopted to solve the minimization problem of formula (16). The QPBO method is a maximum flow / minimum cut algorithm, which has excellent performance in terms of time complexity in graph optimization, can efficiently solve large-scale problems, and can ensure that the result is globally optimal. Through this optimization process, the change states (changed / unchanged) of each superpixel are finally accurately determined, and a complete change detection map of the dual-temporal PolSAR image is generated accordingly, effectively avoiding the problems of regional discontinuity and boundary blurring caused by traditional threshold segmentation.

[0030] It should be understood that although the steps in the above flowchart are shown in sequence according to the arrows, these steps do not necessarily have to be executed in the order indicated by the arrows. Unless otherwise clearly stated in this article, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the above flowchart may include multiple sub-steps or multiple stages. These sub-steps or stages do not necessarily have to be executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages does not necessarily have to be sequential, but can be executed alternately or in turn with at least a part of other steps or sub-steps or stages of other steps.

[0031] In one embodiment, a polarimetric SAR change detection device is further provided, including: A segmentation module, configured to obtain a temporal PolSAR image, calculate the similarity measure of the multi-temporal covariance matrix by using the JBLD divergence; calculate a temporal edge intensity map based on the similarity measure to obtain a temporal edge intensity map; initialize the clustering center by using the temporal edge intensity map and introduce a dynamic edge constraint mechanism during the iteration process to suppress the superpixels from crossing the image edge, and output a superpixel segmentation result; A change detection module, configured to construct an image topological representation that fuses temporal feature similarity, spatial adjacency, and cross-temporal cross-feature similarity based on the superpixel segmentation result; construct an energy function including node cost and edge cost based on the image topological representation, and solve the energy minimization problem through quadratic pseudo-Boolean optimization to obtain a change detection map of the dual-temporal PolSAR image.

[0032] The devices and modules illustrated in the above embodiments can be specifically implemented by a computer chip or an entity, or by a product with a certain function. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0033] In one embodiment, a computer device is further provided, including a memory and a processor. Computer-readable instructions are stored in the memory. When the computer-readable instructions are executed by the processor, the following steps are implemented when the processor executes a computer program: Obtain a temporal PolSAR image, and calculate the similarity measure of the multi-temporal covariance matrix using the JBLD divergence; calculate a temporal edge intensity map based on the similarity measure to obtain a temporal edge intensity map; initialize the clustering center using the temporal edge intensity map and introduce a dynamic edge constraint mechanism during the iteration process to suppress the superpixels from crossing the image edge, and output the superpixel segmentation result; Based on the superpixel segmentation result, construct an image topological representation that fuses the temporal feature similarity, spatial adjacency, and cross-temporal cross-feature similarity; based on the image topological representation, construct an energy function including node cost and edge cost, and solve the energy minimization problem through quadratic pseudo-Boolean optimization to obtain a change detection map of the dual-temporal PolSAR image.

[0034] In one embodiment, a storage medium storing computer-readable instructions is further provided. When the computer-readable instructions are executed by one or more processors, the following steps are implemented when the one or more processors execute: Obtain a temporal PolSAR image, and calculate the similarity measure of the multi-temporal covariance matrix using the JBLD divergence; calculate a temporal edge intensity map based on the similarity measure to obtain a temporal edge intensity map; initialize the clustering center using the temporal edge intensity map and introduce a dynamic edge constraint mechanism during the iteration process to suppress the superpixels from crossing the image edge, and output the superpixel segmentation result; Based on the superpixel segmentation result, construct an image topological representation that fuses the temporal feature similarity, spatial adjacency, and cross-temporal cross-feature similarity; based on the image topological representation, construct an energy function including node cost and edge cost, and solve the energy minimization problem through quadratic pseudo-Boolean optimization to obtain a change detection map of the dual-temporal PolSAR image.

[0035] A storage medium for computer-readable instructions includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transitory medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0036] The embodiments in this specification are all described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and the differences between each embodiment and other embodiments are emphasized. In particular, for the system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and reference can be made to the relevant parts of the method embodiments for the relevant parts.

Claims

1. A polarimetric SAR change detection method, characterized in that, It includes the following steps: Obtain temporal PolSAR images, and calculate the similarity measure of multi-temporal covariance matrices using JBLD divergence; calculate the temporal edge intensity map based on the similarity measure to obtain the temporal edge intensity map; initialize the clustering center using the temporal edge intensity map and introduce a dynamic edge constraint mechanism during the iteration process to suppress superpixels from crossing the image edge, and output the superpixel segmentation result; Based on the superpixel segmentation result, construct an image topological representation that fuses temporal feature similarity, spatial adjacency, and cross-temporal cross-feature similarity; Based on the image topological representation, construct an energy function that includes node cost and edge cost, and solve the energy minimization problem through quadratic pseudo-Boolean optimization to obtain the change detection map of the dual-temporal PolSAR images.

2. The polarimetric SAR change detection method according to claim 1, wherein Using JBLD divergence for the similarity measure of multi-temporal covariance matrices includes the following steps: The functional expression of the JBLD divergence is: (1) Among them, C 1, C 2 represent two polarization covariance matrices, | | represents the determinant of the matrix, and log(·) represents the natural logarithm; Based on the statistical independence characteristic and the maximum value criterion between multi-temporal PolSAR images, the constructed temporal distance measure is as follows: (2) Among them, C represents the i covariance matrix of the j th region in the Z th time phase, Z 1 and N 2 represent the temporal joint covariance matrix of two regions, represents the number of time phases. It should be noted that the specific meaning of some of these notations may need to be further understood in the context of the relevant technical content. There may be some inaccuracies in the translation due to the lack of clear context for these rather specialized notations.

3. A polarimetric SAR change detection method according to claim 2, characterized in that, Calculating the temporal edge intensity to obtain the temporal edge intensity map based on the similarity measure includes the following steps: Assign higher weights to pixels closer to the center position through a Gaussian-type filter, and the expression of the horizontal Gaussian window function is: (3) Among them, ( x , y ) are relative coordinate values, and are respectively controlled by σ x and σ y to control the weight change speed of the window in two directions; For a single-temporal PolSAR image, the calculated average covariance matrices of the regions on both sides of the center line are: (4) Wherein, θ f is the direction angle of the center line, and , Ω k , k = 1, 2 represents the set of pixels in the regions on both sides of the center line; Combining Equation (2) and Equation (4), the temporal edge intensity map is obtained, and the expression is: (5) Among them, and represent the bilateral average covariance matrix calculated by using Equation (4) at the i th time phase.

4. A polarimetric SAR change detection method according to claim 3, characterized in that Introducing a dynamic edge constraint mechanism during the iteration process to suppress superpixels from crossing the image edge includes the following steps: Based on the temporal edge intensity map, determine the edge constraint term, and the expression is: (6) Among them, EDGE norm is a normalized edge intensity map, line ( i,j ) represents a line segment between any two pixels i and j , p denotes the pixels on the line segment; Similarity Measure for SLIC Clustering of Temporal PolSAR Images D SLIC It is generally divided into three parts: feature similarity, edge constraint term, and spatial similarity, which are defined as follows: (7) Among them, α and β are the coefficients of the edge constraint term and the spatial similarity weight respectively, S is the grid sampling step calculated according to the number of superpixels, D P is the maximum absolute percentage power similarity, D S is the spatial similarity.

5. A polarimetric SAR change detection method according to claim 1, characterized in that It also includes a post-processing optimization process, which includes the following steps: Merge isolated superpixel blocks with an area smaller than a preset threshold through connected region analysis to eliminate over-segmentation noise; Use morphological closing operations to smooth the superpixel edges and fill in small breaks caused by local similarity fluctuations.

6. The polarimetric SAR change detection method according to claim 1, wherein Based on the superpixel segmentation result, constructing an image topological representation that fuses temporal feature similarity, spatial adjacency, and cross-temporal cross-feature similarity includes the following steps: Use the polarization target decomposition method to extract the polarization scattering features of the PolSAR image and perform global normalization processing. Use the average polarization scattering feature of the superpixel block as the feature vector of the node, and the constructed graph nodes are: (10) Among them, the nodes , and respectively represent the average polarization scattering characteristics of the i th superpixel in two time phases, N S represents the number of superpixels; Construct the edges of the graph based on spatial similarity and feature similarity ε , and the expression is: (11) Among them, and are respectively edges constructed based on feature similarity in two time phases, ε S is an edge constructed based on spatial similarity, and is defined as follows: (12) (13) (15) Among them, denotes t the K-nearest neighbor set of the similarity of the i th superpixel feature in the image at phase 1; d s ( V i ,V j ) represents the spatial distance between the geometric centers of two superpixels.

7. A polarimetric SAR change detection method according to claim 6, characterized in that The expression of the energy function is: (16) Among them, θ const is a fixed offset value in the energy function, set to 0. θ i ( L i ) represents the cost when the node V i is in the label L i . θ ij ( L i ,L j ) represents the cost when the superpixel pair with feature similarity or spatial proximity is in the label ( L i ,L j ).

8. A polarimetric SAR change detection device, characterized in that, It includes: A superpixel segmentation module for obtaining temporal PolSAR images and calculating the similarity measure of multi-temporal covariance matrices using JBLD divergence; Calculating the temporal edge intensity map based on the similarity measure to obtain the temporal edge intensity map; initializing the clustering center using the temporal edge intensity map and introducing a dynamic edge constraint mechanism during the iteration process to suppress superpixels from crossing the image edge, and outputting the superpixel segmentation result; A change detection module for constructing an image topological representation that fuses temporal feature similarity, spatial adjacency, and cross-temporal cross-feature similarity based on the superpixel segmentation result; Based on the image topological representation, an energy function including node cost and edge cost is constructed, and the energy minimization problem is solved through quadratic pseudo-Boolean optimization to obtain the change detection map of the dual-temporal PolSAR images.

9. A computer device, characterized in that, It includes: a memory for storing a computer program; a processor for implementing the method according to any one of claims 1 to 7 when executing the computer program.

10. A readable storage medium, characterized in that, A computer program is stored on the readable storage medium, and when the computer program is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

Images