Density-based parameter adaptive moving landslide space clustering identification method
By using InSAR technology and the SAO-optimized DBSCAN algorithm, parameters are adaptively adjusted, solving the problem of model parameters relying on empirical thresholds in automatic landslide identification, and improving the accuracy and performance of landslide identification.
Patent Information
- Application Number
- CN202511082721.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-04
- Publication Date
- 2025-10-17
AI Technical Summary
In existing technologies, machine learning methods for automatic landslide identification require setting model parameters through empirical thresholds, which affects clustering performance and accuracy and lacks adaptability.
InSAR technology is used to acquire deformation field data. The Sobel algorithm is used to reduce noise and extract deformed regions. MCD is used to identify moving pixels. The DBSCAN algorithm optimized by SAO is introduced to adaptively adjust the core parameters MinPts and Radius to achieve adaptive clustering.
It improves the performance and accuracy of automatic landslide identification, reduces reliance on empirical thresholds, and enhances the adaptability and accuracy of clustering.
Smart Images
Figure CN120804746A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of disaster identification, and particularly relates to a parameter self-adaptive active landslide spatial clustering identification method based on density. BACKGROUND
[0002] As a new emerging earth observation technology, InSAR technology has the ability to capture small surface deformation, and its precision can reach millimeter to meter level, and is widely used in the field of geological disaster identification. Landslide disasters are often sudden and have great destructive power, which seriously threatens the safety of people's lives and property, so it is crucial to identify landslide hazards in advance. Existing research uses machine learning method (DBSCAN) to carry out automatic identification of active landslides, which has achieved good results. However, the core parameters of the model need to be set by empirical threshold, which significantly affects the clustering performance and precision of DBSCAN. Therefore, there is an urgent need for a landslide spatial clustering identification method with adaptive parameters to improve the performance of automatic landslide hazard identification. SUMMARY
[0003] Based on the above purpose, the application provides a parameter self-adaptive active landslide spatial clustering identification method based on density, which solves the parameter self-adaptive problem of the existing DBSCAN clustering model through the steps of deformation field data acquisition, background noise removal and deformation area extraction, motion pixel positioning, automatic clustering of landslide motion pixels and active landslide identification.
[0004] In order to achieve the above effect, the technical scheme of the application discloses a parameter self-adaptive active landslide spatial clustering identification method based on density, comprising the following steps:
[0005] S1. Obtain an average deformation rate dataset image of the to-be-identified region by InSAR technology;
[0006] S2. Perform noise reduction and extraction processing on the average deformation rate dataset image by using the Sobel algorithm to obtain a fine deformation rate field;
[0007] S3. Locate the motion pixels of the target region by using the fine deformation rate field, and identify the motion pixels by using MCD;
[0008] S4. Perform automatic clustering on the identified motion pixels by using the SOA-optimized DBSCAN algorithm to obtain the final clustering result.
[0009] Further, the S2 is the step of denoising and extracting the average deformation rate dataset by the Sobel algorithm, specifically: first, the Sobel algorithm is used to obtain the gradient matrix of the average deformation rate dataset image in 8 directions, then the Gaussian filter is used to remove the isolated pixels in the gradient matrix, the mean and standard deviation of the gradient matrix are calculated, and the maximum value of the two is taken as the mask threshold of the average deformation rate dataset image, and a fine deformation rate field is obtained.
[0010] Further, the S4 is the step of automatically clustering the identified moving pixels by the DBSCAN algorithm optimized by the SAO, specifically:
[0011] ① defining P as a core pixel in the moving pixels, assigning all the moving pixels MinPts of the radius neighborhood Radius of P to the same cluster as P, and calculating the boundary pixels and noise pixels thereof;
[0012] ② defining the maximum contour coefficient as the SAO optimization function of the optimization target;
[0013] In the initialization stage, the iteration process starts from a randomly generated population, and the upper and lower bounds of the two optimization parameters MinPts and Radius of DBSCAN are predetermined;
[0014] In the exploration and development stage, the probability density function of normal distribution is used to simulate the generation of irregular Brownian motion phenomenon when the snowmelt water is converted into steam, and the snowmelt model is used to simulate the focusing of the optimal solution when the snowmelt is converted into liquid water; at the same time, a double population mechanism is designed to balance the exploration and development stages;
[0015] ③ calculating the fitness value of the optimization target function of the current iteration of the snowmelt optimization algorithm, and retaining the optimal fitness value of the position of the snow particle in the current iteration; simulating the sublimation and melting behavior of snow, and updating the MinPts and Radius parameter solution;
[0016] ④ judging whether the current target function is the optimal solution according to the contour coefficient of the iteration, if yes, inputting the current MinPts and Radius optimal parameters to identify all the moving pixels that should be clustered with P, if not, executing ② to continue iteration until the optimal solution of the MinPts and Radius parameters is obtained, and all the moving pixels are clustered.
[0017] Further, the SAO optimization function is:
[0018] fitness=1-C;
[0019] Wherein C is the contour coefficient, and the value range of C is [-1, 1], the greater the value of C, the better the clustering effect.
[0020] Further, the formula for determining the upper and lower bounds of the two optimization parameters MinPts and Radius of DBSCAN is:
[0021]
[0022] Wherein, L and U represent the lower and upper limits of the optimization parameters respectively; μ is a random number in the range of [0, 1]; N represents the population size; S is a snow cluster.
[0023] Further, after obtaining the final clustering result in S4, the method further comprises filtering the clustering result according to the self-defined slope of the geographical condition of the to-be-recognized region.
[0024] Beneficial effects: the application provides an active landslide spatial clustering identification method, which first obtains the deformation field data of the to-be-measured region, i.e., the annual average deformation rate, through the InSAR technology; then based on the deformation rate result of InSAR, an edge detection algorithm (Sobel operator) is used to obtain the gradient matrix of the input image in 8 directions, a Gaussian filter is used to remove isolated pixels in the gradient matrix, the mean value and standard deviation of the gradient matrix are calculated, and the maximum value of the two is taken as the mask threshold of the input image to obtain a refined deformation rate field; secondly, the refined deformation rate field is used to locate the motion pixels that may be active landslides, and the minimum covariance action (MCD) method in the LIBRA software is used to statistically identify the motion pixels; after locating all the motion pixels that may be active landslides, a snow ablation optimizer (SAO) is introduced to optimize the core parameters of DBSCAN, and a parameter-adaptive SAO-DBSCAN spatial clustering method is formed to automatically cluster the motion pixels into landslide bodies, which greatly improves the performance and accuracy of landslide automatic clustering; finally, after automatically clustering all the pixels that may be landslides, 5° is selected as the screening condition of the landslide according to the geographical condition, and only the clustering results with an average slope greater than 5° are retained, which further improves the accuracy of active landslide spatial clustering identification. BRIEF DESCRIPTION OF DRAWINGS
[0025] Figure 1 A flowchart of an active landslide spatial clustering identification method provided for an embodiment of the application.
[0026] Figure 2 A deformation field schematic diagram of an active landslide spatial clustering identification method provided for an embodiment of the application.
[0027] Figure 3 An adaptive parameter SAO-DBSCAN spatial clustering schematic diagram of an active landslide spatial clustering identification method provided for an embodiment of the application.
[0028] Figure 4A comparison diagram of active landslide identification results of an active landslide spatial clustering identification method provided by an embodiment of the present invention;
[0029] Figure 5 This is the recognition result of the SAO-DBSCAN spatial clustering method with adaptive parameters according to an embodiment of the present invention. DETAILED DESCRIPTION
[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.
[0031] Unless otherwise specifically stated, the relative arrangement of the parts and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present application. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the drawings are not drawn according to actual proportional relationships. The techniques, methods and equipment known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the techniques, methods and equipment should be considered as part of the authorization specification. In all examples shown and discussed here, any specific values should be interpreted as being merely exemplary and not as limitations. Therefore, other examples of the exemplary embodiments may have different values. It should be noted that similar numbers and letters represent similar items in the following figures, and therefore, once an item is defined in one figure, it does not need to be further discussed in subsequent figures.
[0032] Unless otherwise specified, the meanings of technical terms in this specification are the same as those generally understood by those skilled in the art. However, if there is any conflict, the definitions in this specification shall prevail.
[0033] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.
[0034] The following describes in detail the technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems using specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The following embodiments of the present invention are described in conjunction with the accompanying drawings.
[0035] The relevant technical terms in this application are as follows:
[0036] InSAR: Interferometric Synthetic Aperture Radar (Synthetic Aperture Radar Interferometry)
[0037] HYF: Haiyuan Fault
[0038] LPSF: Liupanshan Fault
[0039] DBSCAN: Density-Based Spatial Clustering of Applications with Noise
[0040] SAO: SnowAblation Optimizer (Snow Ablation Optimizer)
[0041] MCD: Minimum Covariance Determinant
[0042] Example 1
[0043] like Figure 1 As shown, this embodiment shows a flowchart of a method for spatial clustering identification of active landslides, including:
[0044] (1) Collect SAR data and InSAR data, process them and obtain deformation field datasets.
[0045] like Figure 2 As shown in the figure, firstly, the Sentinel-1A radar data from the ascending and descending orbit, the digital elevation model of the area to be measured and the atmospheric external correction data (GACOS) are obtained, and the average deformation rate dataset of the area to be identified is obtained by using the synthetic aperture radar interferometry InSAR technology.
[0046] (2) Reduce the impact of background noise and quickly extract the deformed area
[0047] like Figure 3 As shown in the figure, based on the deformation rate results of InSAR, the gradient matrix of the input image in 8 directions is obtained through the edge detection algorithm (Sobel operator). The isolated pixels in the gradient matrix are removed using Gaussian filtering. The mean and standard deviation of the gradient matrix are calculated, and the maximum value of the two is taken as the mask threshold of the input image to obtain the refined deformation rate field.
[0048] (3) Locating the motion pixels of active landslides
[0049] As Figure 3 shown, the use of fine deformation rate field to locate the moving pixels of the active landslide, and use the minimum covariance determinant (MCD) method in LIBRA software to statistically identify the moving pixels, which does not require empirical threshold and adaptive automatic detection of moving pixels.
[0050] (4) Automatic clustering of landslide moving pixels
[0051] As Figure 3 shown, based on the identified moving pixels, the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm is used to automatically cluster into landslide mass. DBSCAN is a powerful clustering algorithm in statistics, but its application in the field of landslide is less. Unlike the commonly used K-means division algorithm, this algorithm is based on the spatial density of pixels, and does not need to define the number of clusters in advance, which improves the adaptability of clustering. In addition, the K-means algorithm forces clustering of all contained pixels, and is easily affected by noise, while the DBSCAN algorithm can exclude noise pixels that lack enough connected neighbors. DBSCAN defines three types of pixels: core pixels, boundary pixels and noise pixels. Core pixels are located within a cluster, and at least a minimum number of moving pixels (MinPts) within a fixed radius (Radius) around the cluster. These moving pixels are called core pixel neighborhood moving pixels. Noise pixels are not adjacent to any core pixels.
[0052] The performance of DBSCAN is highly dependent on two key parameters: core point number (MinPts) and radius (Radius), and manual setting can significantly affect the clustering effect. In order to adaptively find the optimal parameters according to the pixel density of the landslide moving points and improve the clustering accuracy, in the present invention, the snow melting optimization algorithm (SAO) is used to adaptively improve the core parameters MinPts and Radius of the DBSCAN algorithm, and the contour coefficient clustering evaluation index is used as the fitness function to obtain the optimal parameters, and then the high-precision landslide clustering result is obtained. The SAO algorithm is a novel meta-heuristic algorithm that simulates the sublimation and melting behavior of snow to find the optimal solution of multi-objective optimization problem. The specific steps are as follows:
[0053] ① Define P as a core pixel in the moving pixels, assign all radius (Radius) neighborhood moving pixels (MinPts) of P to the same cluster as P, and calculate its boundary pixels and noise pixels.
[0054] ② Define the maximum contour coefficient as the SAO optimization function of the optimization target, the formula is as follows:
[0055] fitness = 1 - C (1)
[0056] Where C is the silhouette coefficient, which takes a value in the range [-1, 1], the larger the value of C, the better the clustering effect.
[0057] In the initialization phase, the iteration process starts from a randomly generated population, and the upper and lower bounds of the two optimization parameters MinPts and Radius of DBSCAN are predetermined. The formula is as follows:
[0058]
[0059] Where L and U represent the lower and upper limits of the optimization parameters, respectively. μ is a random number in the range [0, 1]. N represents the population size. S is a snow cluster.
[0060] In the exploration and development phase, the probability density function of normal distribution is used to simulate the generation of irregular Brownian motion phenomenon when snowmelt water is converted into steam, and the snowmelt model is used to simulate the focusing of the optimal solution when snowmelt is converted into liquid water. At the same time, a double population mechanism is designed to balance the exploration and development phases.
[0061] ③Calculate the fitness value of the optimization objective function of the current iteration of the snowmelt optimization algorithm, and retain the optimal fitness value of the snow particle position in the current iteration, while simulating the sublimation and melting behavior of snow, and updating the MinPts and Radius parameter solution.
[0062] ④According to the silhouette coefficient of the iteration, determine whether the current objective function is the optimal solution, if so, input the current MinPts and Radius optimal parameters to identify all motion pixels that should be clustered with P, if not, execute ② to continue iteration until the optimal solution of MinPts and Radius parameters is obtained, and all moving pixels are clustered.
[0063] (5) Screening active landslide spatial clustering recognition results
[0064] As shown in Figure 3 , after automatic clustering of all pixels that may be landslides, 5° slope is selected as the screening condition according to the geographical conditions, and only the clustering results with an average slope greater than 5° are retained.
[0065] Example 2
[0066] Taking the Haiyuan region of China as an example, as shown in Figure 4 , the InSAR deformation rate spatial distribution map of three tracks in the Haiyuan region is obtained. (a-c) are the annual average deformation rate results of ascending track (T157), descending track (T62), and descending track (T164), respectively.
[0067] Figure 5The SAO-DBSCAN spatial clustering method is used to identify the results of the adaptive parameters. 203 active landslides are automatically identified by using the SAO-DBSCAN method. The active landslides identified based on the sentinel-1A results have obvious spatial heterogeneity, and most of the active landslides are distributed near the faults or along the fault strike.
[0068] The above is an active landslide spatial clustering identification method provided by one or more embodiments of the present specification.
Claims
1. A density-based parameter adaptive active landslide spatial clustering identification method, characterized by: The steps include: S1. Obtain an image of the average deformation rate dataset of the area to be identified using InSAR technology; S2. Use the Sobel algorithm to perform noise reduction and extraction on the average deformation rate dataset image to obtain a refined deformation rate field; S3. Use the refined deformation rate field to locate the moving pixels in the target area and use MCD to identify the moving pixels; S4. Use the SAO-optimized DBSCAN algorithm to automatically cluster the identified motion pixels and obtain the final clustering result.
2. The identification method according to claim 1, characterized in that: S2 describes the denoising and extraction processing of the average deformation rate dataset using the Sobel algorithm as follows: first, the Sobel algorithm is used to obtain the gradient matrix of the average deformation rate dataset image in 8 directions, and then the Gaussian filter is used to remove isolated pixels in the gradient matrix, and the mean and standard deviation of the gradient matrix are calculated. The maximum value of the two is taken as the mask threshold of the average deformation rate dataset image to obtain a fine deformation rate field.
3. The identification method according to claim 1, characterized in that: S4 uses the SAO-optimized DBSCAN algorithm to automatically cluster the identified motion pixels, specifically: ① Define P as the core pixel in the moving pixel, assign all moving pixels MinPts of the radius neighborhood Radius of P to the same cluster as P, and calculate its boundary pixels and noise pixels; ② Define the SAO optimization function with the maximum silhouette coefficient as the optimization target; In the initialization phase, the iterative process starts with a randomly generated population and predetermines the upper and lower bounds of the two optimization parameters of DBSCAN, MinPts and Radius; During the exploration and development phases, a normally distributed probability density function was used to simulate the irregular Brownian motion phenomenon when snowmelt water was converted into steam, and a snowmelt model was used to simulate the focus of the optimal solution when snowmelt was converted into liquid water. At the same time, a dual-population mechanism was designed to balance the exploration and development phases. ③ Calculate the fitness value of the optimization objective function of the current iteration of the snowmelt optimization algorithm, retain the optimal fitness value of the snow particle position in the current iteration; simulate the sublimation and melting behavior of snow, and update the MinPts and Radius parameter solutions; ④ Determine whether the current objective function is the optimal solution based on the iterative silhouette coefficient. If so, input the current optimal MinPts and Radius parameters to identify all moving pixels that should be clustered with P. If not, execute ② and continue iterating until the optimal solution of MinPts and Radius parameters is obtained and all moving pixels are clustered.
4. The identification method according to claim 3, characterized in that: The SAO optimization function is: fitness=1-C; Where C is the silhouette coefficient, and its value range is between [-1, 1]. The larger the value of C, the better the clustering effect.
5. The identification method according to claim 3, characterized in that: The formula for predetermining the upper and lower bounds of the two optimization parameters MinPts and Radius of DBSCAN is: Where L and U represent the lower and upper bounds of the optimization parameters, respectively; μ is a random number in the range [0,1]; N represents the group size; and S is a snow cluster.
6. The identification method according to claim 1, characterized in that: After the final clustering result is obtained as described in S4, the clustering result is further screened by customizing the slope according to the geographical conditions of the area to be identified.