SAR image change detection method based on fusion difference map and morphological reconstruction

Abstract

The application relates to a SAR image change detection method based on fusion difference maps and morphological reconstruction, which comprises the following steps: S1, acquiring two SAR images of the same region at different times; S2, obtaining a logarithmic mean ratio difference map and a logarithmic ratio difference map according to the two SAR images; S3, performing SLIC superpixel segmentation on the logarithmic ratio difference map to obtain a superpixel segmentation difference map; S4, performing wavelet fusion processing on the logarithmic mean ratio difference map and the superpixel segmentation difference map to obtain a wavelet fusion difference map; S5, performing morphological reconstruction on the wavelet fusion difference map to obtain a final difference map; and S6, processing the final difference map by using a fuzzy C-means clustering algorithm to output a detection result map. The method can better utilize the information of neighboring pixels, has excellent detection effects on different types of SAR images, has very good robustness to noise, and can maintain the details of the image.

Description

Technical Field

[0001] This invention belongs to the field of SAR image processing, specifically relating to a SAR image change detection method based on fusion difference map and morphological reconstruction. Background Technology

[0002] Change detection in synthetic aperture radar (SAR) images aims to generate a change map that depicts the changes in two or more time phases between different calibration times. It is widely used in environmental monitoring, ecological monitoring, urban development research, agricultural and forestry monitoring, and natural disaster assessment.

[0003] Difference map generation is a crucial step in change detection, and the accuracy and effectiveness of the difference map in reflecting changes across multiple temporal phases directly impacts the quality of the final detection result. Common difference map generation methods include pixel-based operator generation, neighborhood-information-based operator generation, and methods that fuse multiple approaches. However, none of these methods can fully utilize the information in the image.

[0004] Clustering is a commonly used method for image change detection. Among them, fuzzy C-means clustering (FCM) is the most widely used. However, the FCM algorithm does not utilize neighborhood information and is very sensitive to noise, resulting in unsatisfactory segmentation results. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, this invention provides a SAR image change detection method based on the fusion of difference maps and morphological reconstruction. The technical problem to be solved by this invention is achieved through the following technical solution:

[0006] This invention provides a SAR image change detection method based on fusion of difference maps and morphological reconstruction, comprising:

[0007] S1: Acquire two SAR images of the same area at different times;

[0008] S2: Based on the two SAR images, obtain the log-mean ratio difference map and the log-ratio difference map;

[0009] S3: Perform SLIC superpixel segmentation on the logarithmic ratio difference map to obtain a superpixel segmentation difference map;

[0010] S4: Perform wavelet fusion processing on the log-mean ratio difference map and the superpixel segmentation difference map to obtain a wavelet fused difference map;

[0011] S5: Perform morphological reconstruction on the wavelet fusion difference map to obtain the final difference map;

[0012] S6: Process the final difference map using the fuzzy C-means clustering algorithm and output the detection result map.

[0013] In one embodiment of the present invention, in S2,

[0014] Generate a log-mean ratio difference plot using the log-mean ratio algorithm:

[0015]

[0016] Where Xd1 represents the log-mean-to-difference plot, (i,j) represents the coordinate point in the plot, and μ a (i,j) represents the mean gray value of the pixels within a 3*3 neighborhood window centered at coordinates (i,j) in the image with the larger sum of gray values ​​between the two SAR images, μ. b (i,j) represents the mean gray value of the pixel within a 3*3 neighborhood window centered at coordinate (i,j) in the image with the smaller sum of gray values ​​between the two SAR images. max(Xd1) represents the maximum gray value in the log-mean-to-difference image and min(Xd1) represents the minimum gray value in the log-mean-to-difference image.

[0017] The logarithmic ratio difference plot is obtained according to the following formula:

[0018]

[0019] Where Xd2 represents the logarithmic difference map, and X1(i,j) and X2(i,j) represent the pixels in the two SAR images, respectively.

[0020] In one embodiment of the present invention, S3 includes:

[0021] S31: Initialize seed points by uniformly distributing seed points within the logarithmic difference map according to a preset number of superpixels;

[0022] S32: Calculate the gradient value of all pixels in the 3*3 neighborhood of each seed point, and select the pixel with the smallest gradient value in the neighborhood as the new seed point.

[0023] S33: Assign a class label to each pixel within the neighborhood of each seed point;

[0024] S34: Within the preset search range, calculate the distance between each searched pixel and its corresponding seed point, and take the seed point with the minimum distance as the cluster center of that pixel;

[0025] S35: Repeat steps S32-S34 iteratively until the cluster center of each pixel no longer changes, and obtain the superpixel segmentation difference map.

[0026] In one embodiment of the present invention, S4 includes:

[0027] S41: Perform three-level stationary wavelet transform on the log-mean ratio difference map and the superpixel segmentation difference map respectively to obtain the low-frequency coefficients and high-frequency coefficients of the log-mean ratio difference map, and the low-frequency coefficients and high-frequency coefficients of the superpixel segmentation difference map.

[0028] S42: The low-frequency coefficients of the log-mean ratio difference map and the low-frequency coefficients of the superpixel segmentation difference map are fused to obtain the fused low-frequency coefficients;

[0029] S43: The high-frequency coefficients of the log-mean ratio difference map and the high-frequency coefficients of the superpixel segmentation difference map are fused to obtain the fused high-frequency coefficients;

[0030] S44: Perform a stationary wavelet inverse transform on the fused low-frequency coefficients and the fused high-frequency coefficients to obtain a wavelet fusion difference map.

[0031] In one embodiment of the present invention, in step S42, the fused low-frequency coefficients are obtained according to the following formula:

[0032]

[0033] Where, difL represents the low-frequency coefficients after fusion, A1 represents the low-frequency coefficients of the log-mean-ratio difference map, and A2 represents the low-frequency coefficients of the superpixel segmentation difference map.

[0034] In one embodiment of the present invention, in step S43, the fused high-frequency coefficients are obtained according to the following formula:

[0035]

[0036] Among them, difH k (i,j) represents the fused high-frequency coefficients, k∈{LH,HL,HH}, where LH,HL,HH represent the horizontal, vertical, and diagonal components of the high-frequency coefficients, respectively. This represents the local energy of the log-mean-to-difference plot. This represents the local energy of the superpixel segmentation difference map. This represents the high-frequency coefficient at coordinate (i,j) in the log-mean-to-difference plot. This represents the high-frequency coefficient at coordinate (i,j) in the superpixel segmentation difference map.

[0037] In one embodiment of the present invention, S5 includes:

[0038] The wavelet fusion difference map is reconstructed by morphological opening and morphological closing operations in sequence to obtain the final difference map.

[0039] In one embodiment of the present invention, S6 includes:

[0040] S61: Set the cluster number k, iteration error e, and fuzziness index m;

[0041] S62: Randomly initialize membership degree u kl Set all cluster centers v k =0;

[0042] S63: Set the iteration number iter = 1;

[0043] S64: Calculate cluster centers v k and membership degree u kl :

[0044]

[0045]

[0046] Where, ξ l Representing gray levels, m∈(1,∞) represents the weighting exponent, γ l Let q represent the number of pixels contained in the l-th gray level, l represent the l-th gray level, q represent the total number of gray levels in the image, j represent the j-th cluster center, and c represent the total number of cluster centers.

[0047] S65: Based on the cluster center v obtained in step S64 k and membership degree u kl Calculate the value of the objective function. If the objective function value satisfies |J iter+1 -J iter If |<e or iter>T, then return all membership degrees; otherwise, let iter = iter + 1 and return to step S64, where the objective function is:

[0048]

[0049] In the formula, ||ξ l -v k || represents the Euclidean distance between the k-th cluster center and the l-th gray level;

[0050] S66: Assign each pixel to the class with the highest membership degree and output the detection result image.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0052] The SAR image change detection method based on fusion difference map and morphological reconstruction of the present invention first constructs a novel difference map by fusing the log-mean ratio difference map and the superpixel segmentation difference map. Then, a fuzzy C-means clustering change detection algorithm based on morphological reconstruction is used to detect changes and obtain SAR image change detection results. This method makes better use of the information of neighboring pixels, has excellent detection effect on different types of SAR images, and has very good robustness to noise while preserving image details.

[0053] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of a SAR image change detection method based on fusion difference map and morphological reconstruction provided by an embodiment of the present invention;

[0055] Figure 2 This is a flowchart of a SAR image change detection method based on fusion difference map and morphological reconstruction provided by an embodiment of the present invention;

[0056] Figure 3 This is a flowchart of a morphology-based fuzzy C-means clustering algorithm provided in an embodiment of the present invention;

[0057] Figure 4 This is a SAR image of the San Francisco area provided in an embodiment of the present invention;

[0058] Figure 5 This is a SAR image of the Ottawa area provided in an embodiment of the present invention;

[0059] Figure 6 This is a map showing the differences in the San Francisco area provided in this embodiment of the invention;

[0060] Figure 7 This is a map showing the change detection results of different algorithms in the San Francisco area provided in this embodiment of the invention;

[0061] Figure 8 This is a map showing the change detection results of different algorithms in the Ottawa area provided in this embodiment of the invention; Detailed Implementation

[0062] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the following describes in detail, with reference to the accompanying drawings and specific embodiments, a SAR image change detection method based on fusion difference map and morphological reconstruction proposed according to the present invention.

[0063] The foregoing and other technical contents, features, and effects of the present invention will be clearly presented in the following detailed description of specific embodiments in conjunction with the accompanying drawings. Through the description of the specific embodiments, a more in-depth and concrete understanding can be gained of the technical means and effects adopted by the present invention to achieve its intended purpose. However, the accompanying drawings are for reference and illustration only and are not intended to limit the technical solutions of the present invention.

[0064] Example 1

[0065] Please refer to the above. Figure 1 and Figure 2 , Figure 1 This is a schematic diagram of a SAR image change detection method based on fusion difference map and morphological reconstruction provided by an embodiment of the present invention; Figure 2 This is a flowchart of a SAR image change detection method based on fused difference maps and morphological reconstruction provided by an embodiment of the present invention. As shown in the figure, the SAR image change detection method based on fused difference maps and morphological reconstruction of this embodiment includes:

[0066] S1: Acquire two SAR images of the same area at different times;

[0067] S2: Based on the two SAR images, obtain the log-mean ratio difference map and the log-ratio difference map;

[0068] Specifically, the log-mean ratio difference plot is generated using the log-mean ratio algorithm:

[0069]

[0070] Where Xd1 represents the log-mean-to-difference plot, (i,j) represents the coordinate point in the plot, and μ a (i,j) represents the mean gray value of the pixels within a 3*3 neighborhood window centered at coordinates (i,j) in the image with the larger sum of gray values ​​between the two SAR images, μ. b (i,j) represents the mean gray value of the pixel within a 3*3 neighborhood window centered at coordinate (i,j) in the image with the smaller sum of gray values ​​between the two SAR images. max(Xd1) represents the maximum gray value in the log-mean-to-difference image and min(Xd1) represents the minimum gray value in the log-mean-to-difference image.

[0071] In this embodiment, the log-mean-ratio difference map combines the advantages of the mean-ratio and log-ratio difference maps. It not only effectively utilizes the spatial neighborhood information of the image, but also non-linearly enhances the image, thereby increasing the image contrast.

[0072] Furthermore, the logarithmic ratio difference plot is obtained according to formula (2):

[0073]

[0074] Where Xd2 represents the logarithmic difference map, and X1(i,j) and X2(i,j) represent the pixels in the two SAR images, respectively.

[0075] S3: Perform SLIC superpixel segmentation on the log-ratio difference map to obtain a superpixel segmentation difference map;

[0076] Among them, the SLIC superpixel segmentation algorithm is fast in computation while effectively representing neighborhood features. Specifically, S3 includes:

[0077] S31: Initialize seed points by evenly distributing them within the log-ratio difference map according to the preset number of superpixels;

[0078] In this embodiment, assuming the image has a total of N pixels, pre-segmented into K superpixels of the same size, then the size of each superpixel is N / K, and the step size between adjacent seed points is approximately:

[0079]

[0080] S32: Calculate the gradient value of all pixels in the 3*3 neighborhood of each seed point, and select the pixel with the smallest gradient value in the neighborhood as the new seed point.

[0081] In this embodiment, this step can prevent seed points from falling on contour boundaries with large gradients, so as not to affect the subsequent clustering results.

[0082] S33: Assign a class label to each pixel within the neighborhood of each seed point;

[0083] S34: Within the preset search range, calculate the distance between each searched pixel and its corresponding seed point, and take the seed point with the minimum distance as the cluster center of that pixel;

[0084] In this embodiment, the distance includes color distance and spatial distance, and the distance calculation method is shown in formulas (4)-(6):

[0085]

[0086]

[0087] Where dc represents color distance, ds represents spatial distance, and Ns represents the maximum intra-cluster spatial distance, defined as Ns = S, applicable to each cluster. The maximum color distance Nc varies with different images and different clusters, so a fixed constant p (usually 10) is used instead.

[0088] The final distance metric is shown in formula (7):

[0089]

[0090] It should be noted that the expected superpixel size is S*S, and the search range of SLIC superpixel segmentation is limited to 2S*2S, which can accelerate the convergence of the algorithm.

[0091] S35: Repeat steps S32-S34 iteratively until the cluster center of each pixel no longer changes, and obtain the superpixel segmentation difference map.

[0092] S4: Perform wavelet fusion processing on the log-mean ratio difference map and the superpixel segmentation difference map to obtain the wavelet fused difference map;

[0093] Specifically, S4 includes:

[0094] S41: Perform three-level stationary wavelet transforms on the log-mean ratio difference map and the superpixel segmentation difference map respectively to obtain the low-frequency and high-frequency coefficients of the log-mean ratio difference map and the low-frequency and high-frequency coefficients of the superpixel segmentation difference map.

[0095] S42: The low-frequency coefficients of the log-mean ratio difference map and the low-frequency coefficients of the superpixel segmentation difference map are fused to obtain the fused low-frequency coefficients.

[0096] In this embodiment, the fused low-frequency coefficients are obtained according to formula (8):

[0097]

[0098] Where, difL represents the low-frequency coefficients after fusion, A1 represents the low-frequency coefficients of the log-mean-ratio difference map, and A2 represents the low-frequency coefficients of the superpixel segmentation difference map.

[0099] S43: The high-frequency coefficients of the log-mean ratio difference map and the high-frequency coefficients of the superpixel segmentation difference map are fused to obtain the fused high-frequency coefficients;

[0100] In this embodiment, the fused high-frequency coefficients are obtained according to formula (9):

[0101]

[0102] Among them, difH k(i,j) represents the fused high-frequency coefficients, k∈{LH,HL,HH}, where LH,HL,HH represent the horizontal, vertical, and diagonal components of the high-frequency coefficients, respectively. This represents the local energy of the log-mean-to-difference plot. This represents the local energy of the superpixel segmentation difference map. This represents the high-frequency coefficient at coordinate (i,j) in the log-mean-to-difference plot. This represents the high-frequency coefficient at coordinate (i,j) in the superpixel segmentation difference map.

[0103] Specifically, N represents local energy. i,j It is an N*N window centered at (i,j), cD k (p), p = 1, 2, ..., N×N represents the p-th coefficient within the window, and in this embodiment, N is 3.

[0104] It should be noted that in this embodiment, the low-frequency coefficients reflect the contour of the image. The weighted average fusion rule shown in formula (8) can be used for the low-frequency coefficients. The high-frequency coefficients contain rich edge and detail information, so the fusion rule of taking the local energy small shown in formula (9) is used to suppress background clutter.

[0105] S44: Perform stationary wavelet inverse transform on the fused low-frequency coefficients and the fused high-frequency coefficients to obtain the wavelet fusion difference map.

[0106] In this embodiment, a novel difference map, namely the wavelet fusion difference map, is constructed by fusing the log-mean ratio difference map and the superpixel segmentation difference map. This amplifies the characteristics of both the unchanged and changed regions, and can better utilize the information of neighboring pixels, which is beneficial for subsequent image change detection.

[0107] Furthermore, after obtaining the wavelet fusion difference map, it needs to be segmented to obtain the image change detection result. In this embodiment, morphology-based fuzzy C-means clustering is used. Fuzzy C-means clustering (FCM) is a classic change detection algorithm. It finds fuzzy clusters of given data by minimizing the objective function. However, although the FCM algorithm is fast, it does not consider neighborhood information and is very susceptible to noise. Therefore, in this embodiment, the FCM algorithm is improved. The improved FCM algorithm (i.e., morphology-based fuzzy C-means clustering algorithm) is used to segment the wavelet fusion difference map to obtain the image change detection result.

[0108] Please refer to the above. Figure 3 , Figure 3 This is a flowchart of a morphology-based fuzzy C-means clustering algorithm provided in an embodiment of the present invention, as shown in the figure, including:

[0109] S5: Perform morphological reconstruction on the wavelet fusion difference map to obtain the final difference map;

[0110] Specifically, the wavelet fusion difference map is reconstructed by performing morphological opening and morphological closing operations sequentially to obtain the final difference map.

[0111] Morphological reconstruction algorithms consist of two basic operations: dilation and erosion, where,

[0112] Image dilation: If f is the labeled image, g is the template image, B is the structuring element, and... Then the dilation operator of f with respect to g, which has a size of 1, is: This represents a first-order dilation of the original image f by the structuring element B. Therefore, the dilation operator of size n with respect to g is: in

[0113] Image erosion: If f is the labeled image, g is the template image, B is the structuring element, and... Then the erosion of f with respect to g of size 1 is: fΘB represents the first-order erosion of the original image f by the structuring element B. The erosion of f with respect to g of size is: in

[0114] By combining morphological dilation and erosion operators, some reconstruction operators with strong filtering capabilities can be obtained, such as morphological opening reconstruction operators and morphological closing reconstruction operators.

[0115] Morphological opening reconstruction first uses the original image as a template, then performs an erosion operation on the original image, then uses the eroded image as a marker image, and finally uses the marker image and the template image for dilation reconstruction. The opening reconstruction expression for an image f of size n is as follows:

[0116]

[0117] Morphological closing operation reconstruction first uses the original image as a template, then performs a dilation operation on the original image, then uses the dilated image as a marker image, and finally uses the marker image and the template image for erosion reconstruction. The closing operation reconstruction expression for an image f of size n is as follows:

[0118]

[0119] Opening operations can reduce noise in an image and remove isolated points. Closing operations can fill in small cracks in an image without changing the position and size of image patches. Therefore, alternating between opening and closing operations can remove noise from an image while keeping the image essentially unchanged.

[0120] In this embodiment, morphological filtering of the image is achieved through opening and closing reconstruction operations. The specific reconstruction operation is shown in formula (12):

[0121]

[0122] Where ξ represents the final difference plot.

[0123] S6: Use the fuzzy C-means clustering algorithm to process the final difference map and output the detection result map.

[0124] In this embodiment, S6 includes:

[0125] S61: Set the cluster number k, iteration error e, and fuzziness index m;

[0126] S62: Randomly initialize membership degree u kl Set all cluster centers v k =0;

[0127] S63: Set the iteration number iter = 1;

[0128] S64: Calculate cluster centers v k and membership degree u kl :

[0129]

[0130]

[0131] Where, ξ l Representing gray levels, m∈(1,∞) represents the weighting exponent, γ l Let represent the number of pixels contained in the l-th gray level, l represent the l-th gray level, q represent the total number of gray levels in the image, and j represent the j-th cluster center. c u represents the total number of cluster centers. kl ∈[0,1];

[0132] S65: Based on the cluster center v obtained in step S64 k and membership degree u kl Calculate the value of the objective function. If the objective function value satisfies |J iter+1 -J iter If |<e or iter>T, then return all membership degrees; otherwise, let iter = iter + 1 and return to step S64, where the objective function is:

[0133]

[0134] In the formula, ||ξ l -vk || represents the Euclidean distance between the k-th cluster center and the l-th gray level;

[0135] S66: Assign each pixel to the class with the highest membership degree and output the detection result image.

[0136] The SAR image change detection method based on fused difference map and morphological reconstruction in this embodiment first constructs a novel difference map by fusing the log-mean ratio difference map and the superpixel segmentation difference map. Then, a fuzzy C-means clustering change detection algorithm based on morphological reconstruction is used to detect changes and obtain SAR image change detection results. This method makes better use of the information of neighboring pixels, has excellent detection effect on different types of SAR images, and has very good robustness to noise while preserving image details.

[0137] Example 2

[0138] This embodiment illustrates the effectiveness of the SAR image change detection method based on fusion difference map and morphological reconstruction in Embodiment 1 through simulation experiments.

[0139] Experimental data:

[0140] In this embodiment, the simulation uses dataset 1 and dataset 2 as follows: Figure 4 and Figure 5 As shown, Figure 4 This is a SAR image of the San Francisco area provided in an embodiment of the present invention; Figure 5 This is a SAR image of the Ottawa area provided in an embodiment of the present invention. Figure 4 Figures (a) and (b) are ERS-SAR images taken in the San Francisco area of ​​the United States in August 2003 and May 2004, respectively, with a resolution of 256*256. Figure (c) is a reference image for the changes. Figure 5 Figures (a) and (b) are images obtained by Radarsat in the Ottawa region of Canada in May and August 1997, respectively. Figure (a) is an image of the Ottawa region before the rainy season, and Figure (b) is an image of the region after the rainy season and flooding. The image resolution is 290*350. Figure (c) is a reference image of the changes.

[0141] Experimental indicators:

[0142] To more objectively illustrate the effectiveness of change detection, this embodiment employs five quantitative evaluation metrics: false negatives (FN, pixels that have actually changed but are detected as unchanged), false alarms (FP, pixels that have actually remained unchanged but are detected as changed), total false detections (OE, the sum of false negatives and false alarms), accuracy (PCC, the ratio of the sum of correctly detected pixels to the total number of pixels), and the Kappa coefficient (the similarity between the detected image and the changed reference image). A good change detection method should achieve an accuracy as close to 1 as possible and a Kappa coefficient as close to 1 as possible.

[0143] OE=FP+FN (16),

[0144]

[0145]

[0146]

[0147] In the formula, TP represents the number of positively detected changes (pixels that actually changed are detected as changed pixels), TN represents the number of positively detected unchanged changes (pixels that actually did not change are detected as unchanged pixels), and N represents the total number of pixels in the image. c and N u These represent the actual number of pixels that changed and the number that did not change, respectively.

[0148] Analysis of experimental results:

[0149] All algorithms in this experiment were run on a desktop computer with an Intel(R) Core(TM) i5-10400F CPU @ 2.90GHz and 8GB of RAM, using Matlab R2018a as the simulation software. All experiments were set with a fuzzy index m = 2 and a termination threshold e = 10. -5 The structuring element used in the experiment was a rectangular structuring element.

[0150] San Francisco area Figure 4 For example, Figure 6 This is a map showing the differences in the San Francisco area provided in this embodiment of the invention. Figure 6 Figure (a) is the log-mean ratio difference map, (b) is the SLIC superpixel segmentation difference map, (c) is the wavelet fusion difference map, and (d) is the final difference map.

[0151] As can be seen from the figures, the log-mean difference map preserves image details better than the difference map, but it contains more noise. The SLIC superpixel segmentation difference map has the least noise, but loses some image details. The wavelet fusion difference map combines the advantages of both, preserving image details better with minimal noise. After morphological processing, noise is further suppressed. The gray values ​​of non-changing regions are also significantly reduced, which is more conducive to detection and analysis.

[0152] For the San Francisco area ( Figure 4 The data detection results are as follows: Figure 7 As shown in Table 1, Figure 7 This is a diagram showing the change detection results of different algorithms in the San Francisco area provided by an embodiment of the present invention. Table 1 shows the change detection results of different algorithms in the San Francisco area.

[0153] Table 1. Change detection results of different algorithms in the San Francisco area

[0154]

[0155] Analysis reveals that the FCM algorithm, which only uses distance and not neighborhood information, exhibits significant scattered noise in the image, resulting in high false alarm and false negative counts (393 and 315 pixels respectively). However, it also has the lowest total error count, PCC, and Kappa coefficient. The FLICM algorithm, utilizing neighborhood information, reduces scattered noise considerably, but still suffers from significant speckle noise. KFCM maps the original data to a high-dimensional space via kernel mapping and clusters the data in this space. Through nonlinear mapping, it effectively distinguishes, extracts, and amplifies useful features. Therefore, it performs well, significantly improving noise reduction compared to the previous two algorithms. The algorithm presented in this paper, due to morphological processing, exhibits excellent robustness to noise, resulting in a completely noise-free image. Consequently, it has the lowest false alarm count (229), far lower than the other three algorithms. Its PCC and Kappa coefficients are the highest, reaching 0.9921 and 0.9397 respectively, making this algorithm the best performing.

[0156] For the Ottawa area ( Figure 5 The data detection results are as follows: Figure 8 As shown in Table 2, Figure 8 Table 2 shows the change detection results of different algorithms in the Ottawa area provided in this embodiment of the invention.

[0157] Table 2. Change detection results of different algorithms in the Ottawa area

[0158]

[0159] It can be seen that the FCM algorithm missed too many detections in the elliptical region directly above, reaching 1548 pixels, thus exhibiting the worst performance with a Kappa coefficient of only 0.9307. The FLICM algorithm had the fewest false alarms at 74, achieving good detection results, with PCC and Kappa coefficients reaching 0.9866 and 0.9479, respectively. The KFCM algorithm performed slightly worse, with 171 false alarms and 1281 missed detections, and its PCC and Kappa coefficients were slightly lower than FLICM. The algorithm of this invention had the fewest missed detections at 760, far fewer than the other three algorithms, and its PCC and Kappa coefficients were the highest, reaching 0.9903 and 0.9630, respectively.

[0160] It should be noted that, in this document, the terms "comprising," "including," or any other variations are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0161] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for detecting changes in SAR images based on fusion of difference maps and morphological reconstruction, characterized in that, include: S1: Acquire two SAR images of the same area at different times; S2: Based on the two SAR images, obtain the log-mean ratio difference map and the log-ratio difference map; generate the log-mean ratio difference map using the log-mean ratio algorithm: ， in, This plot represents the difference between the logarithmic mean and the logarithmic mean. This represents the coordinates of the points in the diagram. This indicates that the image with the larger sum of gray values ​​among two SAR images is represented by coordinates. The average gray value of the pixels within a 3x3 neighborhood window centered on the center. This indicates that the image with the smaller total grayscale value among two SAR images is represented by coordinates. The average gray value of the pixels within a 3x3 neighborhood window centered on the center. This represents the maximum gray value in the difference graph between the logarithmic mean and the mean. This represents the minimum gray value in the logarithmic mean to difference graph. The logarithmic ratio difference plot is obtained according to the following formula: ， in, This represents a log-ratio difference plot. and These represent the pixels in the two SAR images respectively; S3: Perform SLIC superpixel segmentation on the logarithmic ratio difference map to obtain a superpixel segmentation difference map; S4: Perform wavelet fusion processing on the log-mean ratio difference map and the superpixel segmentation difference map to obtain a wavelet fused difference map; S5: Perform morphological reconstruction on the wavelet fusion difference map to obtain the final difference map; S6: Process the final difference map using the fuzzy C-means clustering algorithm and output the detection result map.

2. The SAR image change detection method based on fusion difference map and morphological reconstruction according to claim 1, characterized in that, S3 includes: S31: Initialize seed points by uniformly distributing seed points within the logarithmic difference map according to a preset number of superpixels; S32: Calculate the gradient value of all pixels in the 3*3 neighborhood of each seed point, and select the pixel with the smallest gradient value in the neighborhood as the new seed point. S33: Assign a class label to each pixel within the neighborhood of each seed point; S34: Within the preset search range, calculate the distance between each searched pixel and its corresponding seed point, and take the seed point with the minimum distance as the cluster center of that pixel; S35: Repeat steps S32-S34 iteratively until the cluster center of each pixel no longer changes, and obtain the superpixel segmentation difference map.

3. The SAR image change detection method based on fused difference map and morphological reconstruction according to claim 1, characterized in that, S4 includes: S41: Perform three-level stationary wavelet transform on the log-mean ratio difference map and the superpixel segmentation difference map respectively to obtain the low-frequency coefficients and high-frequency coefficients of the log-mean ratio difference map, and the low-frequency coefficients and high-frequency coefficients of the superpixel segmentation difference map. S42: The low-frequency coefficients of the log-mean ratio difference map and the low-frequency coefficients of the superpixel segmentation difference map are fused to obtain the fused low-frequency coefficients; S43: The high-frequency coefficients of the log-mean ratio difference map and the high-frequency coefficients of the superpixel segmentation difference map are fused to obtain the fused high-frequency coefficients; S44: Perform a stationary wavelet inverse transform on the fused low-frequency coefficients and the fused high-frequency coefficients to obtain a wavelet fusion difference map.

4. The SAR image change detection method based on fusion difference map and morphological reconstruction according to claim 3, characterized in that, In step S42, the fused low-frequency coefficients are obtained according to the following formula: ， in, A1 represents the low-frequency coefficients after fusion, and A2 represents the low-frequency coefficients of the log-mean ratio difference map.

5. The SAR image change detection method based on fusion difference map and morphological reconstruction according to claim 3, characterized in that, In step S43, the fused high-frequency coefficients are obtained according to the following formula: ， in, This represents the high-frequency coefficients after fusion. , These represent the horizontal, vertical, and diagonal components corresponding to the high-frequency coefficients, respectively. This represents the local energy of the log-mean-to-difference plot. This represents the local energy of the superpixel segmentation difference map. The coordinates in the logarithmic mean-difference graph are... High-frequency coefficients at the location, Represents the coordinates in the superpixel segmentation difference map. The high-frequency coefficient at that location.

6. The SAR image change detection method based on fusion difference map and morphological reconstruction according to claim 1, characterized in that, S5 includes: The wavelet fusion difference map is reconstructed by morphological opening and morphological closing operations in sequence to obtain the final difference map.

7. The SAR image change detection method based on fusion difference map and morphological reconstruction according to claim 1, characterized in that, S6 includes: S61: Set the cluster number k, iteration error e, and fuzziness index m; S62: Randomly initialize membership degrees Set all cluster centers =0; S63: Set the iteration count iter=1; S64: Calculate cluster centers membership degree : ， ， in, Represents grayscale level. Indicates the weighted index. Indicates the first l The number of pixels contained in each gray level l Indicates the first l One gray level, This represents the total number of gray levels in the image. Indicates the first Cluster centers, Indicates the total number of cluster centers; S65: Based on the cluster centers obtained in step S64 membership degree Calculate the value of the objective function. If the value of the objective function satisfies... or If the condition is met, then return all membership degrees; otherwise, let... Return to step S64, where the objective function is: ， In the formula, Indicates the first k The cluster centers and the first l Euclidean distance of gray levels; S66: Assign each pixel to the class with the highest membership degree and output the detection result image.

Images