Synthetic Aperture Radar Image Change Detection Method Based on Dual Neighborhood Heterogeneous Graphs

Abstract

This invention discloses a method for detecting changes in synthetic aperture radar (SAR) images based on a dual-neighborhood heterogeneous graph. First, spatial distance analysis is performed on pixels in two SAR images to construct a spatial neighborhood point set. Simultaneously, local constraint neighborhoods are established to construct a constraint strength neighborhood point set. Using the pixels of the SAR images as vertices of the dual-neighborhood heterogeneous graph, the vertices and their spatial neighbors are designated as spatial similarity edges, and the vertices and their constraint strength neighbors are designated as constraint strength similarity edges. Weights are assigned to the spatial and constraint strength similarity edges according to similarity rules, thus establishing a single-time-time dual-neighborhood heterogeneous graph model. Change information is measured by comparing the local structures of the dual-neighborhood heterogeneous graph, and the change measurement results are then classified to generate a change detection result image. This invention effectively suppresses speckle noise while further improving the accuracy of change detection methods.

Description

Technical Field

[0001] This invention relates to the field of synthetic aperture radar (SAR) image detection methods, specifically a SAR image change detection method based on dual-neighborhood heterogeneous graphs. Background Technology

[0002] Remote sensing image change detection is a technique that analyzes remote sensing images of the same geographical scene at different times to obtain information on changes in ground features within that scene. Essentially, it studies the spatial distribution changes of ground features, detecting their appearance, disappearance, and attribute changes. Synthetic Aperture Radar (SAR) possesses active, microwave imaging characteristics, and is not limited by atmospheric conditions or lighting conditions, allowing for all-day, all-weather imaging operations. Therefore, SAR image change detection demonstrates stable and highly timely application characteristics in both military and civilian fields.

[0003] SAR images can present extensive and rich information about ground features, but the speckle noise they contain interferes with change detection tasks. Therefore, how to achieve high-precision change detection while suppressing speckle noise has become a key issue in SAR image change detection research.

[0004] Currently, the ratio operator is the most commonly used method for SAR image change detection. It performs point division on images at two different time points to reduce the impact of speckle noise and measure the true change information of the image. However, when the noise level is high, the detection accuracy of this method is insufficient. To address this, researchers have introduced the local mean of the image into the ratio operator, proposing a SAR image change detection method based on the mean ratio. This method replaces the intensity of the original pixels with the intensity mean of the image patch and uses the ratio operator to measure the change. This algorithm has good resistance to speckle noise, but it is difficult to detect texture changes in image patches with invariant means. To address this problem, researchers have studied a SAR image change detection method based on spatial information ratios, which improves the ability to identify local texture changes by introducing higher-order statistical features. In addition, researchers have borrowed the working mode of nonlocal mean filters and combined the filter with the change measurement operator to establish a change detection method based on nonlocal mean ratios, which improves the algorithm's ability to suppress speckle noise.

[0005] By incorporating neighborhood information, the aforementioned change detection methods can effectively mitigate the negative impact of speckle noise on change detection tasks. However, these methods only employ a single neighborhood construction approach, failing to fully utilize pixel neighborhood information, resulting in slightly insufficient change measurement capabilities. Therefore, to more effectively suppress speckle noise and more comprehensively measure image changes, the construction of multiple neighborhood types and the introduction of corresponding neighborhood information are key to improving the performance of change detection methods. Summary of the Invention

[0006] This invention provides a synthetic aperture radar image change detection method based on dual-neighbor heterogeneous graphs to solve the problem of homogeneous neighborhood construction in existing change detection methods.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A synthetic aperture radar image change detection method based on dual-neighborhood heterogeneous graphs includes the following steps:

[0009] Step 1: Acquire two SAR images from synthetic aperture radar at different times;

[0010] Step 2: For each SAR image obtained in Step 1, construct the spatial neighborhood point set and constraint strength neighborhood point set for each pixel in the SAR image.

[0011] Step 3: Based on the spatial neighborhood point set and constraint strength neighborhood point set of each pixel in each SAR image obtained in Step 2, construct the dual-neighborhood heterogeneous map of each SAR image; then, by comparing the local structure of the dual-neighborhood heterogeneous maps of the two SAR images, obtain the change measurement result map.

[0012] Step 4: Classify the change measurement result map obtained in Step 3, and divide it into corresponding change classes and non-change classes to generate the final change detection result map.

[0013] In the further step 1, the two SAR images at different times are registered separately.

[0014] In the further step 2, spatial distance analysis is performed on each pixel and its neighboring pixels in each SAR image to obtain each pixel and its K neighboring pixels, thereby obtaining the spatial neighborhood point set of each pixel.

[0015] Furthermore, spatial distance analysis is performed on each pixel and its neighboring pixels based on the Euler distance function.

[0016] In the further step 2, a local constraint neighborhood is set with each pixel in each SAR image as the center, and M neighboring pixels with the closest intensity to the center pixel are selected from the local constraint neighborhood, thereby obtaining the constraint intensity neighborhood point set of each pixel.

[0017] In the further step 3, for each SAR image, each pixel of the SAR image is used as the vertex of the dual-neighborhood heterogeneous graph. Spatial similarity edges of the dual-neighborhood heterogeneous graph are constructed using each vertex and its spatial neighbors. Constraint strength similarity edges of the dual-neighborhood heterogeneous graph are established using each vertex and its constraint strength neighbors, thereby completing the connection of the two types of heterogeneous edges. Then, weights are assigned to the spatial similarity edges and constraint strength similarity edges according to the degree of correlation between vertices, thus obtaining the dual-neighborhood heterogeneous graph.

[0018] In the further step 4, the classic fuzzy C-means clustering method is used to classify the change measurement result map obtained in step 3 to generate the final change detection result map.

[0019] This invention improves the utilization of image neighborhood information by expanding the composition mode of local image neighborhoods, and enhances the noise suppression capability and change detection accuracy of the method by comparing the form of local structure of heterogeneous images to perform change measurement analysis of SAR images.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0021] 1) This invention conducts an in-depth study of the neighborhood relationships of image pixels, establishes spatial neighborhood point sets and constraint intensity neighborhood point sets, and on this basis, constructs a dual-neighborhood heterogeneous graph model to realize the comprehensive mining and expression of SAR image ground feature information.

[0022] 2) This invention uses the method of comparing the local structure of heterogeneous graphs in two neighborhoods to perform change measurement analysis. The change information of the central vertex and its adjacent vertices will be aggregated through two types of edge weights, thereby making the change measurement and detection results more robust and accurate. Attached Figure Description

[0023] Figure 1 This is a flowchart of the method according to Embodiment 1 of the present invention.

[0024] Figure 2 This is a schematic diagram of the dual-neighborhood heterogeneous graph in the method of Embodiment 1 of the present invention.

[0025] Figure 3 This is the change detection result applied to SAR images in the Huaihe River region in the experiment of Embodiment 2 of the present invention, wherein: Figure 3 (a) is a SAR image of the Huaihe River region during the dry season. Figure 3 (b) is a SAR image of the Huaihe River region during the flood season. Figure 3 (c) is a reference diagram of changes interpreted manually based on expert knowledge. Figure 3 (d) to (h) correspond to the change detection results of the ratio method, mean ratio method, spatial information ratio method, nonlocal mean ratio method, and the change detection method of Example 1, respectively.

[0026] Figure 4 This is the change detection result applied to SAR images in the Lu'an City area in the experiment of Embodiment 2 of the present invention, wherein: Figure 4 (a) is a SAR image of Lu'an City during the flood disaster. Figure 4 (b) is a SAR image of the Lu'an City area during a normal period. Figure 4 (c) is a reference diagram of changes interpreted manually. Figure 4 (d) to (h) correspond to the change detection results of the ratio method, mean ratio method, spatial information ratio method, nonlocal mean ratio method, and the change detection method of Example 1, respectively. Detailed Implementation

[0027] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0028] Example 1

[0029] like Figure 1 As shown, this embodiment discloses a synthetic aperture radar image change detection method based on dual-neighborhood heterogeneous graphs, including the following steps:

[0030] Step 1: Acquire two SAR images X1 and X2 from different times using synthetic aperture radar, and perform registration processing on the two SAR images X1 and X2.

[0031] Step 2: Construct a set of two neighboring regions, as follows:

[0032] (2.1) Construction of spatial neighborhood point set.

[0033] For each SAR image processed in step 1, define T p Let I be any pixel in the image with coordinates (x, y). Calculate the distance between pixel I and the image using the Eulerian distance function. p Spatial distance analysis is performed on pixel I and its nearby pixels to obtain the distance between pixel I and the pixel I. p The K nearest neighboring pixels are used to obtain any pixel I. p Spatial neighborhood point set N s (I p ).

[0034] (2.2) Construction of the neighborhood point set of constraint strength.

[0035] For each SAR image processed in step 1, with pixel I p Define a local constraint neighborhood of size β×β centered at the center, and select M (M≤β) from this local constraint neighborhood. 2 -1) Intensity and center pixel I p The nearest neighboring pixel is used to obtain any pixel I. pThe constraint strength neighborhood point set N c (I p ).

[0036] Step 3: Based on the spatial neighborhood point set N of each pixel in each SAR image obtained in Step 2 s (I p ) and the neighborhood point set N of the constraint strength c (I p To construct a two-neighbor heterogeneous map for each SAR image, the two-neighbor heterogeneous maps of two SAR images have the same topological structure. The process is as follows:

[0037] For each SAR image, let each pixel of the SAR image be {I1, ..., I...} n} as vertices {υ1, ..., υ1} of a two-neighbor heterogeneous graph n}, with each vertex {υ1, ..., υ n} and its spatial neighbor points construct a spatially similar edge E of a bi-neighborhood heterogeneous graph. s Let each vertex {υ1, ..., υ1} be a vertex. n} and its constraint strength neighborhood points to construct a bi-neighborhood heterogeneous graph with similar constraint strength edges E. c This completes the connection between the two types of heterogeneous edges; then, weights are assigned to spatially similar edges and constraint-similar edges according to the degree of correlation between vertices, thus obtaining the dual-neighborhood heterogeneous graph DHG, the structure of which is shown in the figure below. Figure 2 As shown. The two-neighborhood heterogeneous graph DHG = {V, E} s E c w s w c The specific components are as follows:

[0038] V = {υ1, ...,υ} n}={I1,...,I n}

[0039] E s ={(υ p υ q );υ q ∈N s (υ p )}

[0040] E c ={(υ p υ q );υ q ∈N c (υ p )}

[0041] w s (υ p υ q) = max(0.5, exp(-(υ p -υ q ) 2 )); (υ p υ q )∈E s

[0042]

[0043] In the formula: V is the vertex set of the bi-neighborhood heterogeneous graph; {υ p υ q υ l Let} represent any vertex on the bi-neighbor heterogeneous graph, and υ represent... p，i and υ l，i These originate from vertex υ p and υ l Spatial neighborhood point set; w s w represents the weight of spatially similar edges on a two-neighbor heterogeneous graph. c Then, the edge weights are similar in terms of constraint strength, and the degree of association between vertices in the heterogeneous graph is described based on the weights of the two types of edges.

[0044] Based on the above structure, two SAR images X1 and X2 are constructed using two-neighbor heterogeneous maps. Let DHG1 be the two-neighbor heterogeneous map of SAR image X1 and DHG2 be the two-neighbor heterogeneous map of SAR image X2. Both DHG1 and DHG2 have the same topological structure. After the two-neighbor heterogeneous maps DHG1 and DHG2 are constructed, information change is measured by comparing the local structures of DHG1 and DHG2, resulting in the change measurement result map CM. The calculation method is as follows:

[0045]

[0046] In the formula: Let be any vertex of the two-neighbor heterogeneous graph DHG1, and Similarly, is a vertex on the bi-neighbor heterogeneous graph DHG2 and has the same adjacency relationship; lg(·) represents the logarithmic function to the base 10, and abs(·) is the absolute value function.

[0047] Step 4: Using the classic fuzzy C-means clustering method, classify the change measurement result map CM obtained in Step 3, and divide it into corresponding change classes and non-change classes to generate the final change detection result map.

[0048] Example 2

[0049] This embodiment is an experimental example. This embodiment uses the following two sets of real SAR images for comparative experiments to further illustrate the effectiveness of the change detection method in Embodiment 1. The process is as follows:

[0050] 1. Experimental setup:

[0051] Experimental environment: MATLAB 2020, Intel(R) Core(TM) i7-10750H CPU@2.60GHz, Windows 10 Home Chinese Edition.

[0052] Experimental data: Figure 3 The images are SAR images of the Huaihe River region at two different times during the flood and drought seasons, acquired by the Gaofen-3 SAR satellite system. The resolution is 5m and the image size is 780×544 pixels. Figure 4 The image shows SAR images of Lu'an City before and after the floods, acquired by the Sentinel-1 SAR satellite system. The resolution is 10m and the image size is 429×580 pixels.

[0053] Experimental parameters: number of spatial neighborhood points K = 10, local constraint neighborhood parameter β = 7, number of strength constraint neighborhood points M = 16.

[0054] Experimental comparison methods: The ratio method, mean ratio method, spatial information ratio method, and nonlocal mean ratio method were used respectively.

[0055] Evaluation indicators for experimental results: The experimental results were objectively evaluated by selecting the number of false positives, the number of false negatives, the accuracy rate, and the Kappa coefficient.

[0056] 2. Experimental Results and Analysis:

[0057] Experiment 1 applied the ratio method, mean ratio method, spatial information ratio method, nonlocal mean ratio method, and the change detection method from Example 1 to perform change detection on SAR images of the Huaihe River region at two different time points. The experimental results are as follows: Figure 3 As shown, where Figure 3 (a) is a SAR image of the Huaihe River region during the dry season. Figure 3 (b) is a SAR image of the Huaihe River region during the flood season. Figure 3 (c) is a reference diagram of changes interpreted manually based on expert knowledge. Figure 3 (d) to (h) correspond to the change detection results of the ratio method, mean ratio method, spatial information ratio method, nonlocal mean ratio method, and the change detection method of Example 1, respectively.

[0058] Experiment 2 investigated SAR images of Lu'an City at two different time points using the ratio method, mean ratio method, spatial information ratio method, nonlocal mean ratio method, and the change detection method from Example 1. The experimental results are as follows: Figure 4 As shown, where Figure 4(a) is a SAR image of Lu'an City during the flood disaster. Figure 4 (b) is a SAR image of the Lu'an City area during a normal period. Figure 4 (c) is a reference diagram of changes interpreted manually. Figure 4 (d) to (h) correspond to the change detection results of the ratio method, mean ratio method, spatial information ratio method, nonlocal mean ratio method, and the change detection method of Example 1, respectively.

[0059] from Figure 3 , Figure 4 As can be seen from Table 1, the false detection rate and overall detection accuracy of the change detection method in Example 1 are superior to other change detection methods. This is because the change detection method in Example 1 employs a novel dual-neighborhood model, fully mining and utilizing the local neighborhood information of pixels. Based on this model, a dual-neighborhood heterogeneous map is constructed, thereby achieving a complete representation of ground feature information. Furthermore, change measurement analysis is performed by comparing the local structure of the dual-neighborhood heterogeneous map, aggregating the change information of neighborhood vertices, thus better reflecting the differences between pixels in non-changed areas and pixels in changed areas, making the change measurement results more robust and accurate. Table 1 is shown below:

[0060] Table 1. Detection results of SAR image changes in the Huaihe River and Lu'an City areas.

[0061]

[0062] The experimental results above show that the change detection method in Example 1 has better noise resistance and higher detection accuracy compared with existing change detection methods.

[0063] The embodiments described herein are merely preferred embodiments of the invention and are not intended to limit the concept and scope of the invention. Any modifications and improvements made by those skilled in the art to the technical solutions of the invention without departing from the design concept of the invention should fall within the protection scope of the invention. The technical content for which protection is sought in this invention has been fully described in the claims.

Claims

1. A method for change detection in synthetic aperture radar images based on dual-neighborhood heterogeneous graph, characterized in that, The method comprises the following steps: Step 1, obtaining two SAR images of different time points of a synthetic aperture radar; Step 2, for each SAR image obtained in step 1, constructing a spatial neighborhood point set and a constraint intensity neighborhood point set of each pixel point in the SAR image respectively; Step 3, based on the spatial neighborhood point set and the constraint intensity neighborhood point set of each pixel point in each SAR image obtained in step 2, constructing a double-neighborhood heterogeneous graph of each SAR image, including: taking each pixel point of the SAR image as a vertex; constructing a spatial similarity edge based on each vertex and its spatial neighborhood point; constructing a constraint intensity similarity edge based on each vertex and its constraint intensity neighborhood point; assigning weights to the two types of edges according to the correlation degree between the vertices; then obtaining a change measurement result map by comparing the local structures of the double-neighborhood heterogeneous graphs of the two SAR images; Step 4, classifying the change measurement result map obtained in step 3 to divide the corresponding change class and non-change class, and generating a final change detection result map.

2. The dual-neighbor heterogeneous graph-based synthetic aperture radar image change detection method of claim 1, wherein, In step 1, the two SAR images of different time points are respectively registered.

3. The dual-neighbor heterogeneous graph-based synthetic aperture radar image change detection method of claim 1, wherein, In step 2, the spatial distance of each pixel point and its nearby pixel points in each SAR image is analyzed respectively to obtain the spatial neighborhood point set of each pixel point and its K neighboring pixel points.

4. The dual-neighbor heterogeneous graph-based synthetic aperture radar image change detection method of claim 3, wherein, The spatial distance of each pixel point and its nearby pixel points is analyzed according to the Euler distance function.

5. The dual-neighbor heterogeneous graph-based synthetic aperture radar image change detection method of claim 1, wherein, In step 2, a local constraint neighborhood is set for each pixel point in each SAR image, and M neighborhood pixels with the closest intensity to the center pixel point are selected from the local constraint neighborhood, thereby obtaining the constraint intensity neighborhood point set of each pixel point.

6. The dual-neighbor heterogeneous graph-based synthetic aperture radar image change detection method of claim 1, wherein, In step 4, the change measurement result map obtained in step 3 is classified by using the classical fuzzy C-means clustering method to generate a final change detection result map.

Images