Method for interpreting deformed landslide from InSAR (Interferometric Synthetic Aperture Radar) deformation graph based on morphological rule interpretation
By combining grayscale segmentation and morphological rules, the accuracy of landslide contour recognition in InSAR observation results is solved, and the accurate extraction of landslide boundaries and morphology is achieved. It is suitable for complex surface environments within a large spatial range, improving the Efficiency and accuracy of landslide identification.
Patent Information
- Application Number
- CN202411123103.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-12-06
AI Technical Summary
The prior art is difficult to quickly and accurately obtain landslide profiles from InSAR observations, resulting in challenges in disaster prevention and watershed management.
Using a method of combining grayscale segmentation and image morphology laws, the optimal screening threshold is determined through the gradient edge information of InSAR deformation value, and the image enhancement algorithm is used to optimize the segmentation results through morphological calculations to highlight the natural contour characteristics of the landslide.
It realizes the accurate extraction of the boundaries and morphology of active landslides, is suitable for complex surface environments within a large spatial range, improves the accuracy and efficiency of landslide identification, and adapts to the natural morphological characteristics of landslides.
Smart Images

Figure CN119090909A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of image processing, in particular to a method for interpreting a deformed landslide from an InSAR deformation image based on morphological rule interpretation. Background Art
[0002] InSAR technology uses the phase difference of satellite radar observation echoes from two different periods of time to calculate surface deformation. InSAR observation has the advantages of large detection area and high sensitivity, and is becoming the main technology for active landslide identification in recent years. However, how to quickly and accurately obtain landslide contours from InSAR observation results is still a key issue facing disaster prevention and watershed management. Summary of the invention
[0003] The technical problem to be solved by the present invention is to provide a method for interpreting deformed landslides from InSAR deformation images based on morphological rules. A method combining grayscale segmentation and image morphological rules is used to identify deformed bodies from InSAR observation results. First, the optimal screening threshold is determined by using the gradient edge information of the InSAR deformation value to extract the abnormal deformation area. On this basis, an image enhancement algorithm with limited image contrast is selected to improve the image contrast of the weak deformation area. Finally, the segmentation result is optimized according to the morphological rules to make it close to the contour characteristics of the natural landslide.
[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for interpreting deformation landslides from InSAR deformation images based on morphological rule interpretation, comprising the following steps: S1. Preprocessing and grayscale segmentation. To achieve the accuracy of InSAR landslide boundary recognition and solve the problem of uneven distribution of deformation pixels inside the edge of the deformed body, the gradient edge information of the phase intensity image from InSAR observation is used to determine the optimal screening threshold, and then extract the surface deformation pixels. S2, weak signal enhancement, introduces the image contrast-limited image enhancement algorithm to highlight the weak surface deformation area, and builds a morphological refinement processing flow to highlight and improve the surface deformation area; S3. Morphological calculation, by establishing surface deformation pixel screening rules and segmentation, highlighting the weak surface deformation area, and performing edge refinement processing on the original segmentation results, to further improve the accuracy of extracting the boundary and overall morphology of active landslides.
[0005] Preferably, step S1 comprises the following steps: The gradient edge information of the InSAR deformed image is used to determine the optimal screening threshold, and a gradient edge segmentation method is constructed. The first-order differential equivalent operator (Sobel) is used to obtain the gradient matrix of the input image in the horizontal and vertical directions. The input image grayscale values greater than the mean and standard deviation of the gradient matrix are then divided into potential edge points, and isolated pixels are filtered out using a Gaussian filter. Finally, the maximum mean or standard deviation of the potential edge points is taken as the optimal threshold m. After calculation by the Sobel operator, the gradients in the x and y directions of the original image are the horizontal gradients and the vertical gradient .
[0006] Preferably, step S2 comprises the following steps: The mapping function is constructed by calculating the histogram distribution in the local area, and the brightness value is reset to improve the image, and the part exceeding the threshold is "evenly" distributed on the probability density curve, thereby limiting the increase of the cumulative histogram; this processing process can stretch the dynamic range of the image, enhance the image contrast, and highlight the area with weak surface deformation; after dividing the grayscale image of the original image into several equal-sized and non-overlapping sub-blocks, first select the contrast threshold, and then evenly distribute the number of pixels in each sub-block to each gray level of the corresponding sub-block, and obtain the average number of pixels N allocated to each gray level.
[0007] Preferably, step S3 comprises the following steps: After eliminating irrelevant shapes to simplify the original extracted deformed pixels and performing image gain, the pixels that constitute the deformed body are distributed discretely, and there are still weak noise and null values in the binary image. Relying on digital morphology, the structural elements are defined and selected to compare and match with the image to be processed, so as to obtain the shape structure of the image to be processed, and then the background interference information is removed through combined calculation of morphological operators to ensure the structure in the target geometric image, thereby maintaining the image spatial structural characteristics and target geometric structure; the jagged edges and internal null value phenomena of the landslide are solved by introducing the expansion, erosion, opening and closing operations in the morphological method.
[0008] Preferably, the horizontal gradient and the vertical gradient The calculation formulas are: ; ; In the formula is the pixel in the original image, , , and They represent x-1, x+1, y-1 and y+1 respectively. The gradient amplitude can be calculated using the gradients in the x and y directions. : ; After determining the optimal screening threshold, each pixel is compared with the optimal threshold. If the pixel i in the source image is greater than m, the pixel is classified as a deformation area and its absolute value is taken and placed in Otherwise, pixel i is assigned to the background area, and the allocation calculation method is as follows: ; In the formula A deformation map without background.
[0009] In the later stage of data processing, the Min-Max method is used to map the original data to [0, 255]. This does not change the data distribution, but can easily give all data attributes equal weights, making comparison and aggregation between attributes easier and better convergence conditions for the results.
[0010] Preferably, the calculation formula is as follows: ; In the formula, is the number of pixels in the x direction of the sub-block; is the number of pixels in the y direction of the sub-block; is the number of sub-block gray levels; the threshold calculation formula is: ; In the formula, is the interception limit multiple, threshold Exceed of times; calculated using Cut the grayscale histogram of each sub-block and evenly distribute the cut redundant pixels to other gray levels; the average number of pixels distributed to each gray level is: ; is the total number of intercepted pixels; After performing histogram equalization on each sub-block after the pixels are redistributed, the pixel value of each sub-area needs to be calculated using bilinear interpolation. During the calculation process, the center point of each sub-block is used as the reference point. The pixel value is determined by its four adjacent reference points , , and To determine; during the calculation process, first perform linear interpolation in the x direction, and get and The pixel values of the points are and ,Right now: ; ; Then interpolate in the y direction and get : .
[0011] Bilinear interpolation is only for the center point with 4 reference points around it, while the pixel values at the four vertices are obtained by using mapping function transformation, and the remaining pixel calculations at the edge need to be interpolated through two adjacent reference points. The image after interpolation will eliminate the block effect of the image, and Gaussian bilateral filtering can also be used for processing to further improve the signal-to-noise ratio.
[0012] Preferably, the grayscale erosion operation will make the image darker. Defined as: ; In the formula, is a grayscale image, is a structural element (needs to define an origin); is the domain of f, is the domain of b; when exist When moving up, In The minimum value of the pixel is assigned to The origin of the binary image is used to eliminate the noise in the feature area, fill in the empty values, and smooth the lines in the feature area; After the grayscale dilation operation, the image will become brighter. Defined as: ; In the formula is the domain of f, is the domain of b; when pixel In the image When moving up, the image The maximum value on is assigned to At the origin of , remove the background information outside the feature area and only keep the valid features; Grayscale opening operation is to first perform an erosion operation on the image f(x, y), and then perform a dilation operation on the image after the erosion operation; the grayscale opening operation is defined as: ; Where ○ represents the grayscale opening operation, which can remove isolated pixels, connections between single pixels and rough edge areas in the image and smooth the image contour; Grayscale closing operation and grayscale opening operation are dual to each other, that is, Perform dilation operation, and then perform erosion operation on the image after dilation operation; the grayscale closing operation is defined as: ; Where ● represents the grayscale closing operation; this step can smooth the image contour, but unlike the opening operation, it can fill the empty values in the target and increase the spatial correlation of pixels.
[0013] The present invention provides a method for interpreting deformed landslides from InSAR deformation images based on morphological rule interpretation, and the work can be carried out only with non-contact observation results of InSAR remote sensing. Compared with automatic interpretation methods such as supervised classification, it does not require samples, does not rely on image semantic information in training data, and can make full use of abnormal deformation characteristics of all landslides; compared with unsupervised classification automatic interpretation methods such as clustering, it is adapted to the activities of landslide bodies following the special landform evolution and geological body change laws, and the interpretation results have geometric morphological diversity, and the boundaries conform to the natural morphological characteristics of the landslide. The constructed method can be used to quickly find dangerous active landslides in a large spatial range and complex surface environment, and has great room for promotion in the field of geological disaster prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 is a flow chart of the method of the present invention; Figure 2 Extracting deformation pixel effects under different threshold conditions for a single landslide in the present invention; Figure 3 This is a comparison diagram of the image effects of the present invention; Figure 4 This is the morphological processing process and effect diagram of the recognition result of the present invention. DETAILED DESCRIPTION
[0015] like Figure 1 As shown, a method for interpreting deformation landslides from InSAR deformation images based on morphological rule interpretation includes the following steps: S1. Preprocessing and grayscale segmentation. To achieve the accuracy of InSAR landslide boundary recognition and solve the problem of uneven distribution of deformation pixels inside the edge of the deformed body, the gradient edge information of the phase intensity image from InSAR observation is used to determine the optimal screening threshold, and then extract the surface deformation pixels. S2, weak signal enhancement, introduces the image contrast-limited image enhancement algorithm to highlight the weak surface deformation area, and builds a morphological refinement processing flow to highlight and improve the surface deformation area; S3. Morphological calculation, by establishing surface deformation pixel screening rules and segmentation, highlighting the weak surface deformation area, and performing edge refinement processing on the original segmentation results, to further improve the accuracy of extracting the boundary and overall morphology of active landslides.
[0016] Preferably, step S1 comprises the following steps: The gradient edge information of the InSAR deformed image is used to determine the optimal screening threshold, and a gradient edge segmentation method is constructed. The first-order differential equivalent operator (Sobel) is used to obtain the gradient matrix of the input image in the horizontal and vertical directions. The input image grayscale values greater than the mean and standard deviation of the gradient matrix are then divided into potential edge points, and isolated pixels are filtered out using a Gaussian filter. Finally, the maximum mean or standard deviation of the potential edge points is taken as the optimal threshold m. After calculation by the Sobel operator, the gradients in the x and y directions of the original image are the horizontal gradients and the vertical gradient .
[0017] Preferably, step S2 comprises the following steps: The mapping function is constructed by calculating the histogram distribution in the local area, and the brightness value is reset to improve the image, and the part exceeding the threshold is "evenly" distributed on the probability density curve, thereby limiting the increase of the cumulative histogram; this processing process can stretch the dynamic range of the image, enhance the image contrast, and highlight the area with weak surface deformation; after dividing the grayscale image of the original image into several equal-sized and non-overlapping sub-blocks, first select the contrast threshold, and then evenly distribute the number of pixels in each sub-block to each gray level of the corresponding sub-block, and obtain the average number of pixels N allocated to each gray level.
[0018] Preferably, step S3 comprises the following steps: After eliminating irrelevant shapes to simplify the original extracted deformed pixels and performing image gain, the pixels that constitute the deformed body are distributed discretely, and there are still weak noise and null values in the binary image. Relying on digital morphology, the structural elements are defined and selected to compare and match with the image to be processed, so as to obtain the shape structure of the image to be processed, and then the background interference information is removed through combined calculation of morphological operators to ensure the structure in the target geometric image, thereby maintaining the image spatial structural characteristics and target geometric structure; the jagged edges and internal null value phenomena of the landslide are solved by introducing the expansion, erosion, opening and closing operations in the morphological method.
[0019] Preferably, the horizontal gradient and the vertical gradient The calculation formulas are: ; ; In the formula is the pixel in the original image, , , and They represent x-1, x+1, y-1 and y+1 respectively. The gradient amplitude can be calculated using the gradients in the x and y directions. : ; After determining the optimal screening threshold, each pixel is compared with the optimal threshold. If the pixel i in the source image is greater than m, the pixel is classified as a deformation area and its absolute value is taken and placed in Otherwise, pixel i is assigned to the background area, and the allocation calculation method is as follows: ; In the formula A deformation map without background.
[0020] In the later stage of data processing, the Min-Max method is used to map the original data to [0, 255]. This does not change the data distribution, but can easily give all data attributes equal weights, making comparison and aggregation between attributes easier and better convergence conditions for the results.
[0021] Preferably, the calculation formula is as follows: ; In the formula, is the number of pixels in the x direction of the sub-block; is the number of pixels in the y direction of the sub-block; is the number of sub-block gray levels; the threshold calculation formula is: ; In the formula, is the interception limit multiple, threshold Exceed of times; calculated using Cut the grayscale histogram of each sub-block and evenly distribute the cut redundant pixels to other gray levels; the average number of pixels distributed to each gray level is: ; is the total number of intercepted pixels; After performing histogram equalization on each sub-block after the pixels are redistributed, the pixel value of each sub-area needs to be calculated using bilinear interpolation. During the calculation process, the center point of each sub-block is used as the reference point. The pixel value is determined by its four adjacent reference points , , and To determine; during the calculation process, first perform linear interpolation in the x direction, and get and The pixel values of the points are and ,Right now: ; ; Then interpolate in the y direction and get : .
[0022] Bilinear interpolation is only for the center point with 4 reference points around it, while the pixel values at the four vertices are obtained by using mapping function transformation, and the remaining pixel calculations at the edge need to be interpolated through two adjacent reference points. The image after interpolation will eliminate the block effect of the image, and Gaussian bilateral filtering can also be used for processing to further improve the signal-to-noise ratio.
[0023] Preferably, the grayscale erosion operation will make the image darker. Defined as: ; In the formula, is a grayscale image, is a structural element (needs to define an origin); is the domain of f, is the domain of b; when exist When moving up, In The minimum value of the pixel is assigned to The origin of the binary image is used to eliminate the noise in the feature area, fill in the empty values, and smooth the lines in the feature area; After the grayscale dilation operation, the image will become brighter. Defined as: ; In the formula is the domain of f, is the domain of b; when pixel In the image When moving up, the image The maximum value on is assigned to At the origin of , remove the background information outside the feature area and only keep the valid features; Grayscale opening operation is to first perform an erosion operation on the image f(x, y), and then perform a dilation operation on the image after the erosion operation; the grayscale opening operation is defined as: ; Where ○ represents the grayscale opening operation, which can remove isolated pixels, connections between single pixels and rough edge areas in the image and smooth the image contour; Grayscale closing operation and grayscale opening operation are dual to each other, that is, Perform dilation operation, and then perform erosion operation on the image after dilation operation; the grayscale closing operation is defined as: ; Where ● represents the grayscale closing operation; this step can smooth the image contour, but unlike the opening operation, it can fill the empty values in the target and increase the spatial correlation of pixels.
[0024] In order to verify the effectiveness of this method, the Baihetan reservoir area in the lower reaches of the Jinsha River, where landslides are frequent, was used as a test area to perform the above steps: S1: Preprocessing and grayscale segmentation, InSAR technology observation of the surface deformation range of the Baihetan reservoir area is [-0.27, 0.13] and [-0.28, 0.29]. Figure 2 The following are examples of image processing results based on five threshold segmentation methods and segmentation results for a single deformable body. It is found that the GA method can better separate the deformable body from the background area. The deformable body is a closed figure with continuous boundaries and only a small amount of isolated noise (see Appendix Figure 2 -f'). Other methods have different problems: for example, the HI method can distinguish the deformed area from the background, but there is a lot of background noise and the edges of the deformed body are rough (see Figure 2 -b'); the EN method cannot distinguish between deformable bodies and background, cannot effectively extract deformable bodies, and is disturbed by pseudo hidden dangers (rivers) (Appendix Figure 2 -c'); the background noise in the OTSU result is too much, and the deformed body cannot be completely separated from the background (Appendix Figure 2 -d'); AG cannot directly extract the deformed object from the background area, and the salt and pepper noise covers the entire deformed object (see Figure 2 -e'); S2: Weak signal enhancement, image enhancement comparison experiment results are shown in the attached Figure 3As shown in the figure, the HA algorithm can improve the fidelity of the image by supplementing the dark area information in the image, but the grayscale distribution uniformity and texture coarseness are reduced, resulting in blurring of the surrounding area of the deformed body; although the PLT and LOG algorithms improve the brightness of the image, the aggregation characteristics and distribution space characteristics of the image grayscale distribution are greatly reduced; LAP can maintain a balance between grayscale distribution uniformity and clarity, but the image contrast and the amount of information contained are greatly reduced; the CLAHE algorithm can greatly enhance the target edge information while ensuring the high-quality grayscale distribution uniformity and texture clarity in the image, thereby increasing the amount of information contained in the image. In comparison, the CLAHE algorithm has the best effect on the enhancement of deformed bodies in grayscale deformation maps.
[0025] S3: Morphological feature calculation, constructing a morphological process processing scheme to reveal the role of morphological processing in extracting surface deformation areas. The experimental objects are the enhanced intensity image of the study area, the process and result images of morphological processing of binary images (Appendix Figure 4 ), where numbers 1-4 represent the results of the whole image, single target, specific details and superimposed InSAR images, and numbers a-f represent the morphological processing process.
[0026] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.
Claims
1. A method for interpreting deformation landslides from InSAR deformation images based on morphological rules, characterized in that: The following steps are involved: S1. Preprocessing and grayscale segmentation. To achieve the accuracy of InSAR landslide boundary recognition and solve the problem of uneven distribution of deformation pixels inside the edge of the deformed body, the gradient edge information of the phase intensity image from InSAR observation is used to determine the optimal screening threshold, and then extract the surface deformation pixels. S2, weak signal enhancement, introduces the image contrast-limited image enhancement algorithm to highlight the weak surface deformation area, and builds a morphological refinement processing flow to highlight and improve the surface deformation area; S3. Morphological calculation, by establishing surface deformation pixel screening rules and segmentation, highlighting the weak surface deformation area, and performing edge refinement processing on the original segmentation results, to further improve the accuracy of extracting the boundary and overall morphology of active landslides.
2. The method for interpreting deformation landslide from InSAR deformation map based on morphological rule interpretation according to claim 1, characterized in that: The step S1 comprises the following steps: The gradient edge information of InSAR deformed images is used to determine the optimal screening threshold, and a gradient edge segmentation method is constructed. The gradient matrix of the input image in the horizontal and vertical directions is obtained using the first-order differential equivalent operator. The input image grayscale values greater than the mean and standard deviation of the gradient matrix are then divided into potential edge points, and isolated pixels are filtered out using a Gaussian filter. Finally, the maximum mean or standard deviation of the potential edge points is taken as the optimal threshold m. After calculation by the Sobel operator, the gradients in the x and y directions of the original image are the horizontal gradients and the vertical gradient .
3. The method for interpreting deformation landslide from InSAR deformation map based on morphological rule interpretation according to claim 1, characterized in that: The step S2 comprises the following steps: The mapping function is constructed by calculating the histogram distribution in the local area, and the brightness value is reset to improve the image, and the part exceeding the threshold is "evenly" distributed on the probability density curve, thereby limiting the increase of the cumulative histogram; this processing process can stretch the dynamic range of the image, enhance the image contrast, and highlight the area with weak surface deformation; after dividing the grayscale image of the original image into several equal-sized and non-overlapping sub-blocks, first select the contrast threshold, and then evenly distribute the number of pixels in each sub-block to each gray level of the corresponding sub-block, and obtain the average number of pixels N allocated to each gray level.
4. The method for interpreting deformation landslide from InSAR deformation map based on morphological rule interpretation according to claim 1, characterized in that: The step S3 comprises the following steps: After eliminating irrelevant shapes to simplify the original extracted deformed pixels and performing image gain, the pixels that constitute the deformed body are distributed discretely, and there are still weak noise and null values in the binary image. Relying on digital morphology, the structural elements are defined and selected to compare and match with the image to be processed, so as to obtain the shape structure of the image to be processed, and then the background interference information is removed through combined calculation of morphological operators to ensure the structure in the target geometric image, thereby maintaining the image spatial structural characteristics and target geometric structure; the jagged edges and internal null value phenomena of the landslide are solved by introducing the expansion, erosion, opening and closing operations in the morphological method.
5. The method for interpreting deformation landslide from InSAR deformation map based on morphological rule interpretation according to claim 2, characterized in that: The horizontal gradient and the vertical gradient The calculation formulas are: ; ; In the formula is the pixel in the original image, , , and They represent x-1, x+1, y-1 and y+1 respectively. The gradient amplitude can be calculated using the gradients in the x and y directions. : ; After determining the optimal screening threshold, each pixel is compared with the optimal threshold. If the pixel i in the source image is greater than m, the pixel is classified as a deformation area and its absolute value is taken and placed in Otherwise, pixel i is assigned to the background area, and the allocation calculation method is as follows: ; In the formula A deformation map without background.
6. The method for interpreting deformation landslide from InSAR deformation map based on morphological rule interpretation according to claim 3, characterized in that: The calculation formula is as follows: ; In the formula, is the number of pixels in the x direction of the sub-block; is the number of pixels in the y direction of the sub-block; is the number of sub-block gray levels; the threshold calculation formula is: ; In the formula, is the interception limit multiple, threshold Exceed of times; calculated using Cut the grayscale histogram of each sub-block and evenly distribute the cut redundant pixels to other gray levels; the average number of pixels distributed to each gray level is: ; is the total number of intercepted pixels; After performing histogram equalization on each sub-block after the pixels are reallocated, it is also necessary to use bilinear interpolation to calculate the pixel value of each sub-area; During the calculation process, the center point of each sub-block is used as the reference point, and the point to be calculated The pixel value is determined by its four adjacent reference points , , and To determine; during the calculation process, first perform linear interpolation in the x direction, and get and The pixel values of the points are and ,Right now: ; ; Then interpolate in the y direction and get : 。 7. The method for interpreting deformation landslide from InSAR deformation map based on morphological rule interpretation according to claim 4, characterized in that: Grayscale erosion operation will make the image darker. Grayscale image erosion operation Defined as: ; In the formula, is a grayscale image, It is a structural element; is the domain of f, is the domain of b; when exist When moving up, In The minimum value of the pixel is assigned to The origin of the binary image is used to eliminate the noise in the feature area, fill in the empty values, and smooth the lines in the feature area; After the grayscale dilation operation, the image will become brighter. Defined as: ; In the formula is the domain of f, is the domain of b; When the pixel In the image When moving up, the image The maximum value on is assigned to At the origin of , remove the background information outside the feature area and only keep the valid features; Grayscale opening operation is to first perform an erosion operation on the image f(x, y), and then perform a dilation operation on the image after the erosion operation; the grayscale opening operation is defined as: ; Where ○ represents the grayscale opening operation, which can remove isolated pixels, connections between single pixels and rough edge areas in the image and smooth the image contour; Grayscale closing operation and grayscale opening operation are dual to each other, that is, Perform dilation operation, and then perform erosion operation on the image after dilation operation; the grayscale closing operation is defined as: ; Where ● represents the grayscale closing operation; this step can smooth the image contour, but unlike the opening operation, it can fill the empty values in the target and increase the spatial correlation of pixels.
Citation Information
Patent Citations
Remote sensing image change detection method based on multi-feature evidence fusion and structural similarity
CN116912691A
Self-adaptive binarization of digital images
EP1282076A2
Automated processing and interpretation of images, supported in segmentation, and equipment for performing said process.
MX2008007657A
