A high and steep slope deformation monitoring system based on slope radar monitoring

Abstract

The application belongs to the technical field of intelligent slope monitoring, and provides a high and steep slope deformation monitoring system. In order to solve the problem of low accuracy of high and steep slope deformation monitoring in the traditional technology, the system determines a plurality of monitoring units corresponding to the target high and steep slope, determines the current radar image and the reference radar image corresponding thereto, and then judges whether the current radar image and the reference radar image are similar based on a pre-trained slope deformation visual model of difference perception, to judge whether the monitoring unit has "suffered from slope deformation", so as to judge whether the target high and steep slope has suffered from slope deformation. Thus, data focusing and model focusing based on the monitoring unit are realized. Since the data focused on the monitoring unit has more obvious data features, the monitoring model focused on the monitoring unit can improve the accuracy of visual recognition of the model, and the fine mapping between the monitoring unit and the monitoring model can improve the accuracy of high and steep slope monitoring.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and slope monitoring technology, and in particular to a high and steep slope deformation monitoring system based on slope radar monitoring. Background Technology

[0002] High and steep slope deformation monitoring refers to the systematic and periodic measurement and data analysis of surface and internal displacement, settlement, tilt, and other shape changes of artificially excavated or naturally formed high and steep slopes (such as mountain slopes, road cut slopes, open-pit mine slopes, reservoir banks, etc.) to monitor slopes, predict slope hazards, and take corresponding preventive measures.

[0003] Traditionally, slope monitoring is typically conducted using close-up photography, GPS / GNSS monitoring, total stations, and radar. With the development of automation, from the perspective of monitoring efficiency and convenience, especially for steep slopes, there is a growing trend to use GPS / GNSS monitoring and synthetic aperture radar (SAR) for remote, non-contact, area-based slope monitoring. This mainly involves data acquisition and transmission through appropriate information acquisition equipment, followed by data processing and analysis, and then slope monitoring based on the analysis results.

[0004] However, the inventors realized that in traditional technologies, when monitoring slopes remotely, non-contactly, and in a surface manner, the overall data of the slope is usually collected and analyzed in a comprehensive manner. However, for steep slopes with large heights and complex and diverse terrains, this direct and holistic slope monitoring method reduces the accuracy of monitoring the deformation of steep slopes.

[0005] Therefore, improving the accuracy of deformation monitoring of steep slopes has become an urgent problem to be solved in the field of slope monitoring. Summary of the Invention

[0006] The technical problem solved by this invention is to address the low accuracy of deformation monitoring of steep slopes in traditional technologies.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a high-steep slope deformation monitoring system based on slope radar monitoring, comprising: a first determining module, used to determine a plurality of monitoring units corresponding to a target high-steep slope; a first acquiring module, used to acquire the current radar signal corresponding to the monitoring unit based on a preset slope radar; a first conversion module, used to convert the current radar signal into a corresponding current radar image according to a preset data image conversion method; a second determining module, used to determine a reference radar image corresponding to the monitoring unit; a third determining module, used to determine a corresponding pre-trained slope deformation visual model based on difference perception based on the monitoring unit; a first judging module, used to judge whether the current radar image and the reference radar image are similar based on the current radar image and the pre-trained slope deformation visual model based on difference perception; and a first determination module, used to determine that the monitoring unit "has undergone slope deformation" if the above judgment is negative.

[0008] As a preferred embodiment of the high and steep slope deformation monitoring system based on slope radar monitoring according to the present invention, the first determining module includes: a preprocessing submodule, used to acquire the original three-dimensional point cloud data of the target area and preprocess the original three-dimensional point cloud data to obtain ground point cloud data; a first generating submodule, used to generate a target digital elevation model based on the ground point cloud data and a raster interpolation algorithm; a first calculation submodule, used to calculate a slope map based on the target digital elevation model and a GIS spatial analysis algorithm; a first determining submodule, used to determine the area range of the target high and steep slope from the slope map according to a preset slope threshold; a second calculation submodule, used to calculate several topographic factor maps within the area range of the target high and steep slope based on the target digital elevation model; and a first generating submodule, used to combine the several topographic factor maps into a multi-band image and use an unsupervised machine learning algorithm to segment and cluster the multi-band image to generate monitoring units corresponding to dividing the target high and steep slope into several partitions.

[0009] The beneficial effects of this invention are as follows: By identifying several monitoring units corresponding to a target steep slope and determining the corresponding current radar image and reference radar image, and then, based on a pre-trained slope deformation visual model with difference perception, judging whether the current radar image and the reference radar image are similar, it is possible to determine whether the monitoring unit has "experienced slope deformation," thereby determining whether the target steep slope has undergone slope deformation. Thus, by focusing on the monitoring unit and determining its corresponding current radar image and reference radar image, data focusing and image focusing based on the monitoring unit are achieved. Because the data and images focused on the monitoring unit possess more obvious data features and image characteristics... Based on this, the accuracy of slope deformation identification in monitoring units can be improved. Furthermore, by performing a corresponding and targeted fine mapping between the monitoring units and the pre-trained slope deformation visual model based on difference perception, the focus of the corresponding model can be achieved. This not only simplifies the model structure, but also improves the accuracy of model visual recognition because the model only needs to identify the difference perception of a single monitoring area. Thus, the geological features such as topography and geomorphology of high and steep slopes are combined with unit monitoring and its corresponding monitoring model. From the perspective of combining the focus of monitoring units and the focus of monitoring models and achieving a fine mapping between the two, the accuracy and precision of high and steep slope monitoring can be improved. Attached Figure Description

[0010] Figure 1 A schematic block diagram of a high and steep slope deformation monitoring system based on slope radar monitoring provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the overall concept of a high and steep slope deformation monitoring system based on slope radar monitoring provided in an embodiment of the present invention; Figure 3 This is the first sub-schematic block diagram of a high and steep slope deformation monitoring system based on slope radar monitoring provided in an embodiment of the present invention; Figure 4 This is a second schematic block diagram of a high and steep slope deformation monitoring system based on slope radar monitoring provided in an embodiment of the present invention. Detailed Implementation

[0011] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0012] This invention provides a high-steep slope deformation monitoring system based on slope radar monitoring. The system can be applied to devices including but not limited to laptops, desktop computers, servers, cloud platforms, and edge computing devices, and is used in slope monitoring applications, including but not limited to slope monitoring. The invention will be described in detail below through specific embodiments.

[0013] Example 1, please refer to Figure 1 and Figure 2 , Figure 1 This is a schematic block diagram of a high and steep slope deformation monitoring system based on slope radar monitoring provided in an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the overall concept of a high-steep slope deformation monitoring system based on slope radar monitoring, provided in an embodiment of the present invention. Figure 1 As shown, in this embodiment, the high and steep slope deformation monitoring system 100 includes a first determination module 101, a first acquisition module 102, a first conversion module 103, a second determination module 104, a third determination module 105, a first judgment module 106, a first determination module 107, and a second determination module 108; the above functional modules are described in detail below.

[0014] The first determining module 101 is used to determine several monitoring units corresponding to the target steep slope.

[0015] Explained, for target steep slopes such as open-pit mine slopes, mountain slopes, dams, reservoir banks, and high slopes along railways or highways, based on the current topography, slope deformation, and other geological characteristics of the target steep slope, the slopes with similar properties and spatial continuity are divided into several different slope regions. Each slope region is then used as an independent monitoring unit. The monitoring unit is the smallest monitoring unit with similar properties and spatial continuity. This results in several monitoring units corresponding to the target steep slope. It should be noted that the monitoring units are not fixed but dynamically change according to the changes in the topography, slope deformation, and other geological characteristics of the target steep slope. This ensures the integrity and independence of the monitoring units in terms of similar properties and spatial continuity. Therefore, dividing steep slopes into different monitoring units is a way to implement refined, efficient, and accurate monitoring. The key first step in slope deformation monitoring embodies the core principles of the technical solution: "zonal differentiated monitoring, minimization of monitoring units, standardization of geological features, reduction of interference, focus and precision, and improvement of monitoring accuracy and precision." Among these, "zonal differentiated monitoring" is based on the different geological features of steep slopes, such as topography, terrain undulation, and slope deformation. Instead of treating the entire steep slope "equally and holistically," it divides the steep slope into different, independent, and easily uniformly minimized atomic monitoring areas (i.e., monitoring units) according to its topography, terrain undulation, and slope deformation. Each monitoring unit is then monitored for slope deformation based on a pre-trained slope deformation visual model that uses differential perception, achieving targeted and precise monitoring and improving the adaptability, accuracy, and precision of the monitoring.

[0016] Based on the above concept and description, the target steep slope is first divided into several monitoring units. These units can be determined by relevant personnel based on the geographical location of the target steep slope, or they can be automatically divided into monitoring units by a corresponding algorithm. These monitoring units may include, but are not limited to, crack monitoring units, landslide backwall monitoring units, bulging monitoring units, wetland (seepage zone) monitoring units, gully monitoring units, and ridge monitoring units. This further enhances the automation and intelligence of steep slope deformation monitoring. Based on this, in this embodiment of the invention, the first determining module 101 first determines several monitoring units corresponding to the target steep slope. For example, as... Figure 2 As shown, in Figure 2 The target steep slope includes monitoring area 1, monitoring area 2, monitoring area 3 and monitoring area 4.

[0017] The first acquisition module 102 is used to acquire the current radar signal corresponding to the monitoring unit based on a preset slope radar.

[0018] Explained, slope radar is an advanced device that uses synthetic aperture radar (SAR) technology and interferometric measurement principles to monitor the surface deformation of slopes (such as mine slopes, mountains, dams, road cuts, etc.) in an all-weather, long-distance, and high-precision manner. In the fields of mining, civil engineering, and geological disaster prevention, slope instability is a major safety risk. Traditional monitoring methods (such as total stations, GPS, and manual inspections) have certain limitations. Therefore, for the high and steep slope deformation monitoring involved in the embodiments of this invention, slope radar can achieve long-distance and high-precision monitoring of surface deformation of high and steep slopes.

[0019] Based on the above concept and description, in this embodiment of the invention, the first acquisition module 102 acquires radar signals of a target steep slope based on a preset slope radar, and acquires the current radar signal corresponding to the monitoring unit. The original radar signal data is a complex number, containing amplitude and phase information. However, the original radar signal data can also be converted into amplitude diagrams, coherence diagrams, deformation diagrams / interferograms, and then processed accordingly. Furthermore, it should be noted that the term "current radar signal" is only used to distinguish and identify different radar signals and is not used to limit the term "radar signal." Similar terms in this embodiment of the invention are used similarly and will not be repeated here.

[0020] The first conversion module 103 is used to convert the current radar signal into a corresponding current radar image according to a preset data image conversion method.

[0021] Explained, the data image conversion method is preset, that is, the preset data image conversion method. The preset data image conversion method refers to the way to convert the original "complex" data of the radar signal into the corresponding "image" data such as amplitude diagram, coherence diagram, deformation diagram / interferogram, etc. The relevant conversion methods can be referred to the relevant existing technical means, which will not be elaborated here.

[0022] Based on the above concept and setup, in this embodiment of the invention, the first conversion module 103 converts the current radar signal into a corresponding current radar image according to a preset data image conversion method. The present invention does not limit which type of "image" the current radar image is, such as an amplitude diagram, coherence diagram, deformation diagram, or interferogram. Those skilled in the art, with the necessary technical knowledge and R&D capabilities, can determine which type of "image" to convert the current radar signal into, such as an amplitude diagram, coherence diagram, deformation diagram, or interferogram, based on their needs.

[0023] The second determining module 104 is used to determine the reference radar image corresponding to the monitoring unit.

[0024] Explanatoryly, a reference radar image corresponding to the monitoring unit is determined. The reference radar image represents a standard image used as a comparison reference to determine whether the monitoring unit represented by the current radar image has undergone slope deformation. The reference radar image can be a historical normal state radar image corresponding to the monitoring unit, or a historical radar image (i.e., an adjacent historical radar image) corresponding to the monitoring unit and adjacent to the current radar image. The historical normal state radar image represents the radar image corresponding to the monitoring unit in a normal state without deformation.

[0025] The third determining module 105 is used to determine the corresponding pre-trained slope deformation visual model based on difference perception according to the monitoring unit.

[0026] Explained as described above, based on the technical concept of dividing monitoring units according to different geological features such as topography and slope deformation of steep slopes, a pre-trained slope deformation visual model based on difference perception is pre-set for each possible scenario of the monitoring unit. This pre-trained slope deformation visual model based on difference perception is only used for slope deformation monitoring under the geological feature scenario corresponding to the corresponding monitoring unit. Thus, for each monitoring unit, a corresponding pre-trained slope deformation visual model based on difference perception is set up to achieve a refined mapping between the monitoring unit and the pre-trained slope deformation visual model based on difference perception. Since the pre-trained slope deformation visual model based on difference perception is only used for slope deformation monitoring under the geological feature scenario corresponding to the corresponding monitoring unit, the focus of each pre-trained slope deformation visual model based on difference perception is achieved on the basis of the technical concept of "zonal differentiated monitoring," thereby improving the accuracy and monitoring precision of each pre-trained slope deformation visual model based on difference perception.

[0027] The pre-trained slope deformation visual model based on difference perception needs to possess two core capabilities: 1) powerful feature extraction capability (provided by the pre-trained model), which can be identified as a "feature extraction module"; 2) efficient difference comparison capability (provided by a specific network structure design), which can be identified as a "difference comparison module". The "feature extraction module" can be constructed using models including, but not limited to, CNN-based models (convolutional neural networks), VisionTransformers (ViT), and their variants. The "difference comparison module" can be constructed using, but not limited to, models based on Siamese networks, encoder-decoder structures (based on feature fusion), temporal fusion models based on recurrent neural networks or 3D CNNs, and generative adversarial networks (GAN) paradigms, indicating a focus on understanding and quantifying changes between two or more images.

[0028] Based on the above concept and setup, in this embodiment of the invention, the third determining module 105 determines the corresponding pre-trained slope deformation visual model based on difference perception according to the slope area conditions corresponding to different geological features such as topography and slope deformation of the monitoring unit. This enables adaptive and refined mapping between the monitoring unit and the pre-trained slope deformation visual model based on difference perception, thereby improving the accuracy of slope deformation monitoring of the monitoring unit.

[0029] The first judgment module 106 is used to determine whether the current radar image and the reference radar image are similar based on the current radar image and the reference radar image, and based on the pre-trained slope deformation visual model of difference perception; the first judgment module 107 is used to determine that the monitoring unit "has undergone slope deformation" if the above judgment is negative; the second judgment module 108 is used to determine that the monitoring unit "has not undergone significant slope deformation" if the above judgment is positive.

[0030] Explained as described above, based on the adaptive correspondence mapping between the monitoring unit and the pre-trained slope deformation visual model based on difference perception, in this embodiment of the invention, the first judgment module 106 determines whether the current radar image and the reference radar image are similar, i.e., whether the feature difference between the current radar image and the reference radar image is large enough, to determine whether the monitoring unit corresponding to the current radar image has undergone significant slope deformation. The first judgment module 107 determines whether the monitoring unit "has undergone slope deformation" if the above judgment is negative, i.e., the current radar image and the reference radar image are not similar, or the feature difference between the current radar image and the reference radar image is large enough. Otherwise, the second judgment module 108 determines whether the monitoring unit "has undergone significant slope deformation" if the above judgment is positive, i.e., the current radar image and the reference radar image are similar, or the feature difference between the current radar image and the reference radar image is not large or obvious enough. The second judgment module 108 then determines that the monitoring unit "has not undergone significant slope deformation," and by default, a portion of the steep slope corresponding to the monitoring unit is in a stable state without slope deformation.

[0031] In this embodiment of the invention, several monitoring units corresponding to a target steep slope are identified, along with their corresponding current radar images and reference radar images. Then, based on a pre-trained slope deformation visual model using difference perception, the similarity between the current and reference radar images is determined to ascertain whether the monitoring unit has undergone slope deformation. This process determines whether the target steep slope has experienced slope deformation. Thus, by focusing on monitoring units and determining their corresponding current and reference radar images, data and image focusing based on monitoring units is achieved. Because the data and images focused on the monitoring units possess more obvious data and image characteristics... Based on this, the accuracy of slope deformation identification in monitoring units can be improved. Furthermore, by performing a corresponding and targeted fine mapping between the monitoring units and the pre-trained slope deformation visual model based on difference perception, the focus of the corresponding model can be achieved. This not only simplifies the model structure, but also improves the accuracy of model visual recognition because the model only needs to identify a single type of difference perception. Thus, the geological features such as topography and geomorphology of high and steep slopes are combined with unit monitoring and its corresponding monitoring model. From the perspective of combining the focus of the monitoring units and the focus of the monitoring models and achieving a fine mapping between the two, the accuracy and precision of high and steep slope monitoring can be improved.

[0032] In one embodiment, please refer to Figure 3 , Figure 3 This is the first sub-schematic block diagram of a high-steep slope deformation monitoring system based on slope radar monitoring provided in an embodiment of the present invention. Figure 3 As shown, in this embodiment, the first determining module 101 includes: a preprocessing submodule 301, used to acquire the original three-dimensional point cloud data of the target area and preprocess the original three-dimensional point cloud data to obtain ground point cloud data; a first generating submodule 302, used to generate a target digital elevation model based on the ground point cloud data and a raster interpolation algorithm; a first calculation submodule 303, used to calculate a slope map based on the target digital elevation model and a GIS spatial analysis algorithm; a first determining submodule 304, used to determine the area range of the target steep slope from the slope map according to a preset slope threshold; a second calculation submodule 305, used to calculate several terrain factor maps within the area range of the target steep slope based on the target digital elevation model; and a second generating submodule 306, used to combine the several terrain factor maps into a multi-band image and use an unsupervised machine learning algorithm to segment and cluster the multi-band image to generate monitoring units corresponding to dividing the target steep slope into several partitions.

[0033] Explained, the first determining module 101 includes a preprocessing submodule 301, used to acquire raw 3D point cloud data of the target area. This raw 3D point cloud data can be acquired through various sensing technologies (such as slope radar, lidar, photogrammetry, etc.) and consists of a large number of dense and disordered 3D coordinate points (X, Y, Z). These points collectively and directly depict the surface morphology of the scene corresponding to the target area. The raw 3D point cloud data undergoes preprocessing such as denoising and filtering (separating ground points from non-ground points) to obtain ground point cloud data. This ground point cloud data represents point cloud data based on ground points in the target area. Specifically, acquiring ground point cloud data based on slope radar typically involves "slope radar + external digital elevation model (DEM)" or "interferometric radar + photogrammetry."

[0034] The first generation submodule 302 is used to generate a high-resolution target digital elevation model, i.e. a grayscale image containing elevation information, based on ground point cloud data and raster interpolation algorithms (such as inverse distance weighting, Kriging interpolation, etc.). The digital elevation model, abbreviated as DEM, is a dataset that uses digital form to express and simulate the undulation of ground elevation.

[0035] The first calculation submodule 303 is used to calculate a series of terrain factors for each raster cell based on the target digital elevation model and through GIS spatial analysis algorithms (such as neighborhood analysis, convolution kernel, etc.). These factors are core features describing the terrain and landforms (such as slope, aspect, plane curvature, profile curvature, topographic relief, surface roughness, etc.). Generally, in GIS software (such as ArcGIS, QGIS) or professional remote sensing processing software, slope calculation algorithms are used to analyze each raster cell in the DEM. The algorithm calculates the rate of change of elevation between the cell and its eight neighboring cells, and finally obtains the slope value of each point (the unit can be degrees ° or percentage %). The calculated slope values ​​are then rendered using color gradients to obtain a slope map. The slope map is a thematic map that uses color, contour lines, or numerical values ​​to intuitively express the spatial distribution of the slope at each location on the land surface (specifically the slope). GIS spatial analysis algorithms refer to a series of mathematical and logical calculation steps and rules used in Geographic Information Systems (GIS) to process, analyze, model, and interpret geospatial data (i.e., data with location information).

[0036] The first determining submodule 304 is used to determine the area range of the target high and steep slope from the slope map according to a preset slope threshold. For example, the area range of the target high and steep slope is extracted from the slope map according to the preset slope threshold and continuous area threshold.

[0037] The second calculation submodule 305 is used to calculate several topographic factor maps based on the target digital elevation model within the area of ​​the target steep slope. The topographic factor maps include at least one of the following: slope map, aspect map, plane curvature map, and profile curvature map. The calculation of topographic factors can refer to relevant existing technical means, which will not be elaborated here.

[0038] The second generation submodule 306 is used to combine several topographic factor maps into a multi-band image, and use unsupervised machine learning algorithms (such as K-Means, DBSCAN, mean shift algorithm) to segment and cluster the multi-band image. Based on the spatial variability of topographic features (slope, aspect, curvature, etc.), the continuous slope surface is divided into multiple internally homogeneous and clearly defined "geomorphic units". These units are the preliminary monitoring zones (i.e. monitoring units). Furthermore, unsupervised clustering algorithms incorporate spatial proximity constraints during the clustering process. For example, this can be achieved by using pixel spatial coordinates as additional feature input, or by using a custom distance metric function. Alternatively, spatial proximity constraints can be implemented using density-based spatial clustering (DBSCAN). Since DBSCAN can handle clusters of arbitrary shapes well and conforms to the distribution characteristics of geological units, it groups pixels with similar features and spatial proximity into the same category, thus simultaneously satisfying the feature similarity and spatial continuity of the slope area. Based on this, introducing spatial constraints into the clustering process using unsupervised machine learning algorithms ensures that the generated monitoring units are spatially continuous, thereby generating monitoring units corresponding to dividing the target steep slope into several partitions. Among these, K-Means / ISODATA clustering groups multiple terrain features of each pixel into a feature vector for clustering. In the clustering results, continuous areas of the same category constitute a monitoring unit. Mean-Shift clustering, on the other hand, does not require a preset number of categories and can automatically determine the number of clusters, and is not sensitive to noise.

[0039] Furthermore, the preprocessing submodule 301 includes: a first acquisition submodule, used to acquire the original three-dimensional point cloud data of the target area through airborne lidar or UAV photogrammetry technology; and a first preprocessing submodule, used to perform noise reduction, filtering and classification preprocessing on the original three-dimensional point cloud data to separate the ground point cloud data representing the ground surface.

[0040] Specifically, the first acquisition submodule is used to acquire the original 3D point cloud data of the target area through airborne LiDAR or UAV photogrammetry technology. LiDAR (Light Detection and Ranging) emits a laser beam towards the target, measures the time it takes for the laser to reflect back, calculates the precise distance based on the speed of light, and calculates the 3D coordinates of each point by combining the scanner's own attitude and position (usually through GNSS / IMU positioning). UAV photogrammetry technology is based on the UAV taking a large number of overlapping photos of the same target area from different angles, finding the same feature points through computer vision algorithms, and calculating the 3D coordinates of these feature points based on the principle of parallax.

[0041] The first preprocessing submodule is used to perform noise reduction, filtering and classification preprocessing on the original 3D point cloud data to separate the ground point cloud data that represents the ground surface.

[0042] The denoising process can employ methods including, but not limited to, statistical outlier removal and radius outlier removal to remove obvious and irregular error points (outliers). Statistical outlier removal is based on the statistical characteristics of the point cloud. For each point, the average distance to its K nearest neighbors is calculated. Assuming that these distances follow a Gaussian distribution throughout the point cloud, a standard deviation multiple is set as a threshold. Points with an average distance exceeding the threshold are considered outliers and are removed. Radius outlier removal counts the number of neighboring points within a specified radius of a given point. If the number is lower than a set threshold, the point is considered an isolated noise point.

[0043] Filtering can employ methods including, but not limited to, progressive triangulation encryption algorithms, cloth simulation filtering, and slope filtering to separate “ground points” from “non-ground points” (mainly low vegetation and small obstacles).

[0044] Classification is typically not a standalone step, but rather an extension and refinement of the filtering step. It involves semantically classifying the point cloud and assigning a category label to each point (e.g., ground, vegetation, building, low point / noise). In the slope monitoring of this embodiment, the primary focus is on ground points.

[0045] This invention, through a combination of digital elevation model (DEM) and GIS spatial analysis algorithms, automatically identifies target steep slopes within a target area. Furthermore, by combining DEM and segmentation clustering, the target steep slopes are automatically divided into different monitoring units. This allows for the fully automated, objective, and efficient generation of several monitoring units corresponding to the target steep slope, while maintaining similarity in terrain attributes and spatial continuity, even in the face of complex steep slope topography. This addresses the unique challenges posed by the complexity of steep slope terrain (such as large elevation differences, numerous shadows, and high data noise) to automated processing. Based on these monitoring units, deformation monitoring of steep slopes is achieved, further improving the automation and efficiency of steep slope deformation monitoring while enhancing its accuracy and precision.

[0046] In one embodiment, the second generation submodule 306 includes: a segmentation and clustering submodule, used to combine several topographic factor maps into a multi-band image, and use an unsupervised machine learning algorithm to segment and cluster the multi-band image to generate several initial monitoring units corresponding to the target steep slope; a despotting submodule, used to use morphological opening and closing operations to remove noisy fine patches from the initial monitoring units to obtain despotted monitoring units; and a conversion submodule, used to convert the despotted monitoring units into vector polygons to obtain several monitoring units corresponding to the target steep slope.

[0047] Explained, the segmentation and clustering submodule is used to combine several topographic factor maps into a multi-band image, and to segment and cluster the multi-band image using an unsupervised machine learning algorithm to generate several initial monitoring units corresponding to the target steep slope, as described above, and will not be repeated here.

[0048] The speckle removal submodule uses morphological opening and closing operations to remove noisy fine speckles from the initial monitoring units, resulting in speckled monitoring units. In the morphological opening and closing operations, the opening operation (erosion followed by dilation) effectively removes scattered small speckles (noise), such as isolated, irrelevant pixels within a unit. The closing operation (dilation followed by erosion) fills small holes within the unit and smooths the unit boundaries, making them more natural and consistent with the actual terrain's continuous morphology. Because unsupervised clustering algorithms (such as K-Means) directly generate raster partitioning results, their boundaries... Often jagged and containing numerous isolated misclassified points of only one or a few pixels (i.e., "salt-and-pepper noise"), this is completely unreasonable for geological engineering applications. A monitoring unit should be a continuous and complete area. However, after morphological opening and closing operations, each unit of the raster partition map is a smooth-bounded and internally continuous "block," which is visually and logically closer to an area divided by a human expert. This can eliminate "salt-and-pepper noise," optimize partition boundaries, and further improve the accuracy and precision of high and steep slope monitoring.

[0049] The conversion submodule is used to convert the spot monitoring units into vector polygons. Generally, the grid units are converted into vector polygons through vectorization. Each polygon represents a monitoring unit and has its own clear boundary coordinates, thus obtaining several monitoring units corresponding to the target steep slope.

[0050] Furthermore, the second generation submodule 306 also includes a terrain feature attribute calculation submodule, used to calculate the average terrain feature attribute of each monitoring unit based on the target digital elevation model, wherein the average terrain feature attribute includes at least one of the following: average slope, dominant aspect, maximum elevation, minimum elevation, elevation difference, and surface roughness.

[0051] Specifically, the terrain feature attribute calculation submodule is used to calculate the average terrain feature attribute of each monitoring unit based on the target digital elevation model. The average terrain feature attribute includes at least one of the following: average slope, dominant aspect, maximum elevation, minimum elevation, elevation difference, and surface roughness. The calculation of the average terrain feature attribute can transform the output result from "algorithm result" into "usable engineering product", which greatly improves the practicality and completeness of the technical solution of the present invention embodiment.

[0052] It should be noted that the calculation of the above average terrain feature attributes can refer to relevant methods in the existing technology, and will not be repeated here. Furthermore, although each of the above calculations is existing, combining them for the specific purpose of "quantitatively assigning attributes to automatically divided slope monitoring units" gives these calculations a new and valuable application scenario and technical effect. These calculated attribute values ​​are intended to automatically and quantitatively guide subsequent monitoring behaviors (such as risk classification, alarm threshold setting, etc.).

[0053] In this embodiment of the invention, by performing morphological post-processing and vectorization on the partitions of the target steep slope, not only can the partition boundaries be optimized, but each monitoring unit also has its own clear boundary coordinates, which can improve the output quality and availability of the monitoring units, thereby further improving the accuracy and precision of steep slope monitoring.

[0054] In one embodiment, the pre-trained slope deformation visual model based on difference perception includes a pre-trained feature extraction module and a pre-trained difference fusion module; the first judgment module 106 includes: a feature extraction submodule, used to extract features from the current radar image based on the pre-trained feature extraction module to obtain a first feature map, and to extract features from the reference radar image to obtain a second feature map; a difference feature calculation submodule, used to calculate the difference features between the first feature map and the second feature map based on the pre-trained difference fusion module to obtain a difference feature map; a decoding submodule, used to decode the difference feature map to obtain a corresponding deformation heatmap; and a first judgment submodule, used to determine whether the current radar image and the reference radar image are similar based on the deformation heatmap.

[0055] Explained, the feature extraction submodule is used to extract features from the current radar image based on the pre-trained feature extraction module to obtain a first feature map, and to extract features from the reference radar image to obtain a second feature map. The pre-trained feature extraction module is described in the "Feature Extraction Module" above and will not be repeated here.

[0056] The differential feature calculation submodule is used to calculate the differential features between the first feature map and the second feature map based on the pre-trained differential fusion module, and obtain the differential feature map. The pre-trained differential fusion module is described in the "Differential Comparison Module" above, and will not be repeated here.

[0057] The decoding submodule is used to input the difference feature map into a lightweight decoder (consisting of a series of upsampling and convolutional layers), and finally output a deformation heatmap of the same size as the input image. The value of each pixel in the deformation heatmap represents the probability or degree of deformation at that location.

[0058] The first judgment submodule is used to determine whether the current radar image is similar to the reference radar image based on the deformation heat map, so as to determine whether the monitoring unit has undergone deformation. It can take measures including but not limited to pixel-level threshold segmentation, regional morphological filtering, and time-series trend analysis on the deformation heat map generated by decoding, and finally comprehensively determine whether the local high and steep slope deformation corresponding to the monitoring area has occurred, its location, range and severity level.

[0059] Further, please refer to Figure 4 , Figure 4 This is the second sub-schematic block diagram of a high-steep slope deformation monitoring system based on slope radar monitoring provided in an embodiment of the present invention. (See diagram below.) Figure 4 As shown, in this embodiment, the first judgment submodule 400 includes: a binarization processing submodule 401, used to binarize the deformation heatmap using a preset probability threshold to obtain a binarized mask, wherein a first value in the binarized mask identifies potential deformation regions and a second value identifies undeformed regions; a removal submodule 402, used to perform morphological processing and connected component analysis on the binarized mask to remove noise regions with an area smaller than a preset area threshold, obtaining several candidate deformation regions corresponding to the current radar image; and a continuous detection submodule 403, used to continuously detect the current radar image. The preceding radar image is the initial image. Based on the deformation detection results of multiple consecutive images, a temporal deformation curve of the candidate deformation region is generated. A first determination submodule 404 is used to determine that the current radar image is similar to the reference radar image based on the changing trend of the temporal deformation curve, where the changing trend indicates that the temporal deformation curve fluctuates randomly near the zero value. A second determination submodule 405 is used to determine that the current radar image is not similar to the reference radar image when the changing trend indicates that the temporal deformation curve does not fluctuate randomly near the zero value.

[0060] Specifically, the binarization processing submodule 401 is used to binarize the deformation heatmap using a preset probability threshold to obtain a binarized mask. The first value in the binarized mask identifies potential deformation areas, and the second value identifies undeformed areas. Since the decoder ultimately outputs a deformation heatmap corresponding to the spatial size of the input image, the value of each pixel on the deformation heatmap (usually a floating-point number between 0 and 1) represents the probability or relative degree of deformation at that location. Therefore, the following judgment method can be used to determine whether deformation has occurred at a single pixel, achieving pixel-level judgment of "where might have changed?". The judgment method is as follows: {Set a probability threshold (Threshold, α); if pixel value > α: then the pixel is determined as a "suspected deformation point"; else: the pixel is determined as an "undeformed point"}.

[0061] The removal submodule 402 is used to perform morphological processing and connected component analysis on the binarized mask, removing noise regions with an area smaller than a preset area threshold to obtain several candidate deformation regions corresponding to the current radar image. Since the activation of a single pixel may be noise (such as cloud shadows, changes in illumination, etc.), and the true slope deformation is usually continuous and patchy, based on the above single-pixel level judgment, the following judgment method can be used to achieve the region-level judgment "Is this a meaningful deformation region?", the judgment method is as follows: {Judgment method: morphological processing and connected component analysis; Denoising: Perform morphological opening operation (erosion followed by dilation) on the above binarized mask to remove isolated, small-area noise points; Clustering: Find all connected components (Blob), that is, regions composed of interconnected "suspected deformation points"; Filtering: Set an area threshold (β) for these connected components; only regions with an area greater than β will be retained as "candidate deformation regions"}.

[0062] The continuous detection submodule 403 is used to generate a temporal deformation curve of candidate deformation regions based on the deformation detection results of multiple consecutive images, using the current radar image as the first image. The first determination submodule 404 is used to determine that the current radar image is similar to the reference radar image based on the changing trend of the temporal deformation curve, where the changing trend indicates that the temporal deformation curve fluctuates randomly near the zero value. The second determination submodule 405 is used to determine that the current radar image is not similar to the reference radar image, where the changing trend indicates that the temporal deformation curve does not fluctuate randomly near the zero value.

[0063] Since true deformation typically exhibits temporal continuity, directionality, and trend, based on the aforementioned regional-level judgment, the following judgment method can be used to achieve a time-series-level judgment: "Is this a meaningful deformation region?" The judgment method is as follows: {Judgment method: multi-period data analysis and trend fitting; it is not sufficient to rely solely on two periods of imagery, but requires the introduction of third, fourth, or even longer-term sequence data; plot the curve of the average pixel value (or deformation amount) within the candidate deformation region over time; analyze the trend: stable / no trend: the curve fluctuates randomly around zero → no significant deformation has occurred; linear trend: the curve rises slowly and continuously → uniform deformation has occurred; accelerating trend: the slope of the curve is increasing → deformation is accelerating, and the warning level is high!}

[0064] Furthermore, the difference feature calculation submodule includes: a subtraction submodule, used to subtract the first feature map from the second feature map element by element based on the pre-trained difference fusion module, and calculate the absolute value of the subtraction to obtain an initial difference feature map; and a stitching submodule, used to stitch the initial difference feature map and the second feature map together channel by channel to fuse the original background feature information of the deformation starting point corresponding to the reference radar image, and to fuse the stitched features through a convolutional layer to obtain a difference feature map.

[0065] Specifically, the subtraction submodule is used to subtract the first feature map from the second feature map element by element based on the pre-trained difference fusion module, and calculate the absolute value of the subtraction to obtain the initial difference feature map. The pre-trained difference fusion module is as described above and will not be repeated here.

[0066] The stitching submodule performs channel-wise stitching between the initial difference feature map and the second feature map to fuse the original background feature information of the deformation starting point corresponding to the reference radar image. The stitched features are then fused through a convolutional layer to obtain the difference feature map, which contains the difference information between the images. Specifically, the channel-wise stitching with the second feature map provides the difference feature map with contextual information corresponding to the "starting point of change," i.e., "what this location looked like before the change," which helps the network distinguish between true and false changes.

[0067] Since slope deformation difference detection is not a typical change detection (such as building construction or vegetation changes), it requires extremely high precision to perceive deformations at the millimeter or centimeter level. Simultaneously, it must severely suppress spurious changes caused by factors such as lighting, shadows, seasonal vegetation, and sensor noise. This is a very challenging and complex engineering problem. Furthermore, simple differential operations are very fragile, extremely sensitive to noise and registration errors, and generate a large number of invalid signals in complex natural environments. Directly using them for high-precision slope monitoring yields poor results. The above embodiment, based on the initial differential, stitches back the second feature map corresponding to the reference radar image for convolutional fusion. Thus, "stitching" provides contextual information about the "starting point of change," that is, it provides the semantic background of "what is changing," essentially telling the convolutional... The layered structure "not only represents the region of difference, but also the original background from which deformation begins before the difference occurs." Based on this, it provides crucial information for the network to distinguish between real deformation and spurious changes, and provides the necessary contextual information for the network to determine the true and false changes. Thus, through learning, it can weigh the difference information and contextual information from the spliced ​​features, and autonomously learn how to suppress noise and enhance the real deformation signal. This is a more advanced and intelligent splicing and fusion achieved in response to the special challenges of slope monitoring scenarios (small deformation, high noise, and high reliability requirements). It can greatly improve the model's ability to suppress spurious changes (noise resistance). Noise such as changes in illumination, shadow movement, vegetation growth, and water fluctuations can be effectively filtered out because the semantics of the ground background they are attached to are clear.

[0068] In this embodiment of the invention, a pre-trained slope deformation visual model based on difference perception is realized by combining feature extraction and difference comparison to determine whether the current radar image has undergone deformation. By integrating powerful feature extraction capabilities with efficient difference comparison capabilities, the invention focuses on understanding and quantifying whether the current radar image has changed.

[0069] It should be noted that the high and steep slope deformation monitoring system based on slope radar monitoring described in the above embodiments can be recombined with the technical features included in different embodiments as needed to obtain a combined implementation scheme, but all of them are within the protection scope claimed by this invention.

[0070] The modules in the aforementioned high and steep slope deformation monitoring system based on slope radar can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0071] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0072] The software tools, components, or models not belonging to our company that appear in the embodiments of this invention are merely illustrative examples and do not represent actual use.

[0073] The data collection in this embodiment of the invention complies with the requirements of relevant laws and regulations, such as China's Personal Information Protection Law, GDPR (General Data Protection Regulation of the European Union), or information security standards of other countries and regions.

[0074] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A high-steep slope deformation monitoring system based on slope radar monitoring, characterized in that, include: The first determination module is used to determine several monitoring units corresponding to the target steep slope; The first acquisition module is used to acquire the current radar signal corresponding to the monitoring unit based on a preset slope radar. The first conversion module is used to convert the current radar signal into a corresponding current radar image according to a preset data image conversion method. The second determining module is used to determine the reference radar image corresponding to the monitoring unit; The third determining module is used to determine the corresponding pre-trained slope deformation visual model based on difference perception according to the monitoring unit. The first judgment module is used to determine whether the current radar image and the reference radar image are similar based on the current radar image and the reference radar image, and based on the pre-trained slope deformation visual model with difference perception. The first determination module is used to determine that the monitoring unit "has undergone slope deformation" if the above determination is negative.

2. The high-steep slope deformation monitoring system based on slope radar monitoring as described in claim 1, characterized in that, The first determining module includes: a preprocessing submodule, used to acquire the original three-dimensional point cloud data of the target area and preprocess the original three-dimensional point cloud data to obtain ground point cloud data; a first generating submodule, used to generate a target digital elevation model based on the ground point cloud data and a raster interpolation algorithm; a first calculation submodule, used to calculate a slope map based on the target digital elevation model and a GIS spatial analysis algorithm; a first determining submodule, used to determine the area range of the target steep slope from the slope map according to a preset slope threshold; a second calculation submodule, used to calculate several topographic factor maps within the area range of the target steep slope based on the target digital elevation model; and a second generating submodule, used to combine the several topographic factor maps into a multi-band image and use an unsupervised machine learning algorithm to segment and cluster the multi-band image to generate monitoring units corresponding to dividing the target steep slope into several partitions.

3. The high-steep slope deformation monitoring system based on slope radar monitoring as described in claim 2, characterized in that, The preprocessing submodule includes: a first acquisition submodule, used to acquire the original three-dimensional point cloud data of the target area through airborne lidar or UAV photogrammetry technology; and a first preprocessing submodule, used to perform noise reduction, filtering and classification preprocessing on the original three-dimensional point cloud data to separate the ground point cloud data representing the ground surface.

4. The high and steep slope deformation monitoring system based on slope radar monitoring as described in claim 2, characterized in that, The topographic factor map includes at least one of the following: slope map, aspect map, plan curvature map, and profile curvature map.

5. The high-steep slope deformation monitoring system based on slope radar monitoring as described in claim 2, characterized in that, The second generation submodule includes: a segmentation and clustering submodule, used to combine several topographic factor maps into a multi-band image, and use an unsupervised machine learning algorithm to segment and cluster the multi-band image to generate several initial monitoring units corresponding to the target steep slope; a despotting submodule, used to use morphological opening and closing operations to remove noisy small patches from the initial monitoring units to obtain despotted monitoring units; and a conversion submodule, used to convert the despotted monitoring units into vector polygons to obtain several monitoring units corresponding to the target steep slope.

6. The high-steep slope deformation monitoring system based on slope radar monitoring as described in claim 5, characterized in that, The second generation submodule further includes: a terrain feature attribute calculation submodule, used to calculate the average terrain feature attribute of each monitoring unit based on the target digital elevation model, wherein the average terrain feature attribute includes at least one of the following: average slope, dominant aspect, maximum elevation, minimum elevation, elevation difference, and surface roughness.

7. The high and steep slope deformation monitoring system based on slope radar monitoring as described in claim 1, characterized in that, The pre-trained slope deformation visual model based on difference perception includes a pre-trained feature extraction module and a pre-trained difference fusion module. The first judgment module includes: a feature extraction submodule, used to extract features from the current radar image based on the pre-trained feature extraction module to obtain a first feature map, and to extract features from the reference radar image to obtain a second feature map; a difference feature calculation submodule, used to calculate the difference features between the first feature map and the second feature map based on the pre-trained difference fusion module to obtain a difference feature map; a decoding submodule, used to decode the difference feature map to obtain a corresponding deformation heatmap; and a first judgment submodule, used to determine whether the current radar image and the reference radar image are similar based on the deformation heatmap.

8. The high-steep slope deformation monitoring system based on slope radar monitoring as described in claim 7, characterized in that, The first judgment submodule includes: a binarization processing submodule, used to binarize the deformation heatmap using a preset probability threshold to obtain a binarized mask, wherein a first value in the binarized mask identifies potential deformation regions and a second value identifies undeformed regions; a removal submodule, used to perform morphological processing and connected component analysis on the binarized mask to remove noise regions with an area smaller than a preset area threshold, obtaining several candidate deformation regions corresponding to the current radar image; a continuous detection submodule, used to generate a temporal deformation curve of the candidate deformation regions based on the deformation detection results of multiple consecutive images, using the current radar image as the first image; a first judgment submodule, used to determine that the current radar image is similar to the reference radar image based on the changing trend of the temporal deformation curve, where the changing trend indicates that the temporal deformation curve fluctuates randomly near zero; and a second judgment submodule, used to determine that the current radar image is not similar to the reference radar image, where the changing trend indicates that the temporal deformation curve does not fluctuate randomly near zero.

9. The high-steep slope deformation monitoring system based on slope radar monitoring as described in claim 7 or 8, characterized in that, The differential feature calculation submodule includes: a subtraction submodule, used to subtract the first feature map from the second feature map element by element based on the pre-trained differential fusion module, and calculate the absolute value of the subtraction to obtain an initial differential feature map; and a stitching submodule, used to stitch the initial differential feature map and the second feature map together by channel to fuse the original background feature information of the deformation starting point corresponding to the reference radar image, and to fuse the stitched features through a convolutional layer to obtain a differential feature map.

10. The high-steep slope deformation monitoring system based on slope radar monitoring as described in claim 1, characterized in that, The system further includes a second determination module, used to determine that the monitoring unit "has not experienced significant slope deformation" when the current radar image is similar to the reference radar image.

Images