Slope monitoring and early warning method based on oblique photography and radar technology

Abstract

The invention discloses a slope monitoring and early warning method based on oblique photography and radar technologies. The method comprises the following steps: acquiring three-dimensional point cloud data and surface texture data of a slope; constructing a risk identification model; inputting the three-dimensional point cloud data and the surface texture data into a registration fusion module for registration fusion to obtain three-dimensional image data of the slope; the three-dimensional image data of the side slope are input into an image recognition module for crack recognition, and a crack image is obtained and comprises crack positions, sizes and the number; the crack image is input into a comparison module, and whether risk warning is carried out or not is judged based on a preset threshold value; and further dynamically adjusting the scanning frequency of the synthetic aperture radar through the crack image. According to the method, the three-dimensional point cloud data acquired by the multi-line laser radar and the surface texture data acquired by the oblique photography unmanned aerial vehicle are fused, so that high-precision and comprehensive identification of the slope crack can be realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of slope monitoring and early warning, and specifically relates to a slope monitoring and early warning method based on oblique photography and radar technology. Background Art

[0002] Slope monitoring is an important means to prevent geological disasters and ensure the safety of infrastructure. Traditional slope monitoring technologies mainly rely on single sensors or monitoring methods, such as surface displacement sensors, rain gauges, or total stations. Although these methods can provide certain monitoring data, they have many limitations in complex environments. For example, surface monitoring equipment is difficult to penetrate vegetation cover and cannot obtain information on the interior or hidden areas of slopes; while satellite remote sensing technology, although having a wide coverage, its resolution and data accuracy will be greatly affected in areas with dense vegetation or complex terrain.

[0003] In recent years, with the development of unmanned aerial vehicle (UAV) technology and radar technology, slope monitoring means have been significantly enriched. UAV oblique photography technology can quickly obtain high-resolution surface texture data of slopes, and through three-dimensional modeling, the topography and crack distribution of slopes can be intuitively displayed. However, when facing vegetation cover such as trees, this technology may lead to data loss in some areas. Radar technology, especially synthetic aperture radar (SAR) and ground penetrating radar (GPR), can penetrate vegetation and soil to obtain the internal structure information of slopes, but the resolution of radar images is relatively low, and it is difficult to directly reflect the details of the slope surface.

[0004] How to fuse multi-source data obtained by oblique photography technology and radar technology is a key technical problem in the current field of slope monitoring. On the one hand, the high-resolution surface texture data provided by oblique photography technology can clearly show the location, size, and shape of cracks, but it cannot penetrate vegetation cover; on the other hand, although radar technology can break through vegetation cover to obtain the internal structure information of slopes, its data resolution is low and it is difficult to directly use it for precise crack identification. In addition, the heterogeneity of multi-source data (such as differences in data format, dimension, time resolution, etc.) also poses challenges to data fusion.

[0005] Furthermore, in traditional slope monitoring, the synthetic aperture radar (SAR) scan and crack image data have been in a fragmented state for a long time: the synthetic aperture radar mechanically scans the slope in zones according to a fixed grid, unable to associate with crack risk characteristics, resulting in insufficient scanning of high-risk areas and wasted resources in low-risk areas. Summary of the Invention

[0006] In view of this, the present invention provides a slope monitoring and early warning method based on oblique photography and radar technology, which realizes the all-element real-time monitoring of high-steep slopes and the accurate early warning of disaster risks through multi-source data fusion, dynamic threshold classification, and intelligent prediction algorithms.

[0007] To achieve the above object, the technical solution of the present invention is realized as follows: A slope monitoring and early warning method based on oblique photography and radar technology, comprising: S1: Obtain the three-dimensional point cloud data of the slope through a multi-line lidar; obtain the surface texture data of the slope through an oblique photography unmanned aerial vehicle; the surface texture data includes the position, size and quantity of cracks on the slope surface; S2: Construct a risk identification model, the model includes: a registration and fusion module, an image recognition module and a comparison module; Input the three-dimensional point cloud data and the surface texture data into the registration and fusion module for registration and fusion to obtain the three-dimensional image data of the slope; Input the three-dimensional image data of the slope into the image recognition module for crack recognition to obtain a crack image, the crack image includes the crack position, size and quantity; Input the crack image into the comparison module, and judge the risk type and perform hierarchical warning based on a preset threshold; S3: Perform zonal scanning detection on the slope through a synthetic aperture radar, and optimize the zones based on the crack image in S2 to obtain an optimized detection area; dynamically adjust the scanning frequency of the synthetic aperture radar based on the risk type of the optimized detection area; Among them, obtaining the optimized detection area is expressed as: ; Among them, represents the optimized detection area; represents the position coordinates of the pixel points in the optimized detection area; represents the crack in the crack image, the central position coordinates, represents any crack in the crack image, the central position coordinates; represents the Euclidean distance; represents the crack in the corresponding crack image, the risk type of the risk scaling factor; represents any crack in the corresponding crack image, the risk type of the risk scaling factor, represents the risk scaling factor.

[0008] Further, specifically, obtaining the three-dimensional point cloud data of the slope through a multi-line lidar in step S1 is as follows: Adopt a multi-line lidar, set the horizontal angular resolution, vertical angular resolution and point cloud density; obtain the three-dimensional point cloud data of the slope through multi-station scanning and stitching , expressed as: ; Among them, represents the -th point's axis coordinate in the three-dimensional point cloud data, represents the -th point's axis coordinate in the three-dimensional point cloud data, represents the -th point's axis coordinate in the three-dimensional point cloud data; The surface texture data of the slope obtained by the oblique photography UAV in step S1 is specifically: The oblique photography UAV flies crosswise at a preset inclination angle to generate an orthophoto image, enhances the crack texture of the orthophoto image through the HSV algorithm, and obtains the surface texture data of the slope.

[0009] Furthermore, the three-dimensional image data of the slope obtained in step S2 is specifically: Spatial registration, expressed as: ; Among them, represents the rotation matrix, represents the translation vector, represents the surface texture data; By minimizing the corresponding point distance error , solve the rotation matrix and translation vector, so as to geometrically align the surface texture data and the three-dimensional point cloud data; Fuse the surface texture data and the three-dimensional point cloud data after registration to obtain the three-dimensional image data of the slope.

[0010] Furthermore, in step S2, input the three-dimensional image data of the slope into the image recognition module for crack recognition to obtain the crack image, specifically: Step S21: Preprocess and enhance the three-dimensional image data of the slope to obtain an enhanced two-dimensional grayscale image; Step S22: Perform multi-scale feature extraction, attention enhancement and decoding on the enhanced two-dimensional grayscale image to obtain a crack mask image, that is, a crack image; during this process, obtain the count label of the crack mask image; Step S23: Perform morphological closing operation on the crack mask image to optimize the crack boundary, and filter noise based on the threshold of the connected domain area and the image size, and finally obtain the noise-filtered crack mask image; Step S24: Obtain the position, size and quantity of the cracks based on the noise-filtered crack mask image.

[0011] Further, step S21 is specifically as follows: Preprocess the three-dimensional image data of the input slope, and expand the three-dimensional image data into a two-dimensional grayscale image to retain the brightness difference between the crack and the image background, denoted as; ; wherein, are respectively the red channel value, green channel value and blue channel value of the three-dimensional image data, are all preset weight parameters; Through algorithm to enhance the local contrast of the two-dimensional grayscale image to obtain the enhanced two-dimensional grayscale image , denoted as: ; wherein, represents the contrast-limited adaptive histogram equalization function.

[0012] Further, step S22 is specifically as follows: Perform multi-scale feature extraction on the enhanced two-dimensional grayscale image to obtain multi-scale features , denoted as: ; wherein, represents a lightweight convolutional layer for extracting the local texture of the crack; represents the atrous spatial pyramid pooling operation for simultaneously capturing the features of thin cracks and wide cracks; Perform attention enhancement on the multi-scale features to obtain enhanced multi-scale features , denoted as: ; wherein, represents channel attention for suppressing image interference; Generate a crack mask image through the decoder , denoted as: ; wherein, represents the decoder, represents the Sigmoid activation function, Separate the crack mask image to obtain individual crack images and perform annotation, denoted as: ; wherein, represents the connected component labeling function for distinguishing sticky or crossing cracks, Counting label indicating a single crack image Indicates a preset threshold value

[0013] Furthermore, step S23 is specifically as follows: Perform morphological filtering on the crack mask image to obtain an optimized crack mask image, denoted as: ; Among them, denotes a closing operation; denotes a dilation function for expanding the boundary of the crack image; denotes an erosion function for shrinking the boundary of the crack image; denotes the optimized crack mask image; Perform noise filtering on the optimized crack mask image to obtain a noise-filtered crack mask image , denoted as: ; Among them, denotes an indicator function, denotes the connected domain area of the input optimized crack mask image, denotes the height of the input optimized crack mask image, denotes the width of the input optimized crack mask image, is a preset noise coefficient.

[0014] Furthermore, step S24 is specifically as follows: Obtain the position coordinates of the noise-filtered crack mask image , denoted as: ; Among them, denotes the crack coordinates of the th pixel point in, denotes the noise-filtered crack mask image total number of pixels; denotes the connected domain area of the input optimized crack mask image; Perform size calculation on the noise-filtered crack mask image to obtain the length of the crack, the average width and the maximum width , denoted as: ; ; ; Among them, It represents the th sequence of path points generated by skeletonizing a single crack; It represents the th sequence of path points generated by skeletonizing a single crack; It represents the Euclidean distance; It represents the ranging function; For the denoised crack mask image perform quantity statistics to obtain the quantity of cracks , which is expressed as: ; Among them, represents the maximum value function.

[0015] Furthermore, the process of judging whether to perform risk warning based on a preset threshold is specifically as follows: Obtain the risk coefficient , which is expressed as: ; Among them, , and are all preset weight parameters, represents the preset length of the crack, represents the preset maximum width of the crack, represents the preset maximum quantity of cracks; Based on the risk coefficient obtain the risk type to perform hierarchical alarm, specifically as follows: ; Among them, 0, 1, and 2 are the labels corresponding to each risk type.

[0016] Furthermore, the risk scaling factor in S3 corresponds to the risk type of the area, specifically as follows: When the risk type is the high-risk area, the risk scaling factor > 1; When the risk type is the medium-risk area, the risk scaling factor = 1; When the risk type is the low-risk area, the risk scaling factor < 1.

[0017] The beneficial effects of the present invention are as follows:

[0018] 1. By fusing the three-dimensional point cloud data obtained by the multi-line lidar and the surface texture data obtained by the oblique photography unmanned aerial vehicle, it is possible to achieve high-precision and comprehensive identification of slope cracks.

[0019] The specific manifestations are as follows: Precise extraction of crack information: The three-dimensional point cloud data provides accurate spatial geometric information of the slope, enabling the accurate positioning of the location and size of cracks; the surface texture data clearly shows the shape and quantity of cracks through high-resolution images. The fusion of the two makes crack identification no longer limited to a single dimension, but can accurately locate and quantify crack features in three-dimensional space; Breaking through occlusion limitations: Although oblique photography technology can obtain high-resolution surface textures, it has limitations when facing occlusions such as trees and other vegetation. The three-dimensional point cloud data of lidar can penetrate partial vegetation occlusions and obtain topographic information of the occluded area. Through data fusion, the missing crack information in the occluded area can be effectively filled, realizing comprehensive monitoring of slope cracks. Enhancing the quality of model input: The three-dimensional image data after multi-source data fusion is used as the input of the risk identification model, providing the model with richer and more accurate crack feature information. This helps improve the model's ability to identify crack risks and prediction accuracy, thereby more reliably evaluating the stability of the slope.

[0020] 2. The present invention dynamically scales the partition range based on the crack risk type, improving the monitoring coverage of the high-risk area of the slope, compressing the coverage of the low-risk area, and completely solving the problem of data fragmentation; and dynamically adjusts the scanning frequency of the synthetic aperture radar based on this, improving the monitoring efficiency while reducing costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The present invention provides a slope monitoring and early warning method based on oblique photography and radar technology. Referring to Figure 1 , a slope monitoring and early warning method based on oblique photography and radar technology includes: S1: Obtain the three-dimensional point cloud data of the slope through a multi-line lidar; obtain the surface texture data of the slope through an oblique photography unmanned aerial vehicle; the surface texture data includes the location, size, and quantity of cracks on the slope surface. S2: Construct a risk identification model, which includes: a registration and fusion module, an image recognition module, and a comparison module; Input the three-dimensional point cloud data and the surface texture data into the registration and fusion module for registration and fusion to obtain the three-dimensional image data of the slope; Input the three-dimensional image data of the slope into the image recognition module for crack identification to obtain a crack image, which includes the crack location, size, and quantity; Input the crack image into the comparison module, and judge the risk type and perform hierarchical warning based on a preset threshold. S3: Use synthetic aperture radar to perform zonal scanning detection on the slope, and optimize the zones based on the crack images in S2 to obtain the optimized detection areas. Dynamically adjust the scanning frequency of the synthetic aperture radar based on the risk types of the optimized detection areas. Among them, obtaining the optimized detection areas is expressed as: ; Among them, represents the optimized detection areas; represents the position coordinates of the pixel points in the optimized detection areas; represents the cracks in the crack images of the central position coordinates, represents any one crack in the crack images of the central position coordinates; represents the Euclidean distance; represents the corresponding cracks in the crack images of the risk types of the risk scaling factors; represents any one crack in the corresponding crack images of the risk types of the risk scaling factors, represents the risk scaling factor.

[0023] Furthermore, specifically, the three-dimensional point cloud data of the slope obtained by the multi-line lidar in step S1 is as follows: Adopt a multi-line lidar, set the horizontal angular resolution, vertical angular resolution and point cloud density; obtain the three-dimensional point cloud data of the slope through multi-station scanning and stitching , which is expressed as: ; Among them, represents the th point in the three-dimensional point cloud data of the axis coordinate, represents the th point in the three-dimensional point cloud data of the axis coordinate, represents the th point in the three-dimensional point cloud data of the axis coordinate; Specifically, the surface texture data of the slope obtained by the oblique photography UAV in step S1 is as follows: The oblique photography UAV flies crosswise at a preset inclination angle to generate an orthophoto image, and enhances the crack texture of the orthophoto image through the HSV algorithm to obtain the surface texture data of the slope.

[0024] Furthermore, specifically, the three-dimensional image data of the slope obtained in step S2 is as follows: Spatial registration, denoted as: ; wherein, represents the rotation matrix, represents the translation vector, represents the surface texture data; By minimizing the distance error of corresponding points , the rotation matrix and the translation vector are solved, so as to geometrically align the surface texture data and the three-dimensional point cloud data; Fuse the registered surface texture data and the three-dimensional point cloud data to obtain the three-dimensional image data of the slope.

[0025] Furthermore, in step S2, the three-dimensional image data of the slope is input into the image recognition module for crack recognition to obtain a crack image, specifically: Step S21: Preprocess and enhance the data of the three-dimensional image data of the slope to obtain an enhanced two-dimensional grayscale image; Step S22: Perform multi-scale feature extraction, attention enhancement and decoding on the enhanced two-dimensional grayscale image to obtain a crack mask image, that is, a crack image; during this process, the count label of the crack mask image is obtained; Step S23: Perform morphological closing operation on the crack mask image to optimize the crack boundary, and perform noise filtering based on the threshold of the connected domain area and the image size, and finally obtain the noise-filtered crack mask image; Step S24: Obtain the position, size and quantity of the cracks based on the noise-filtered crack mask image.

[0026] Furthermore, step S21 is specifically: Preprocess the three-dimensional image data input into the slope, and expand the three-dimensional image data into a two-dimensional grayscale image , so as to retain the brightness difference between the crack and the image background, denoted as; ; wherein, are respectively the red channel value, the green channel value and the blue channel value of the three-dimensional image data, are all preset weight parameters; By algorithm to enhance the local contrast of the two-dimensional grayscale image to obtain an enhanced two-dimensional grayscale image , denoted as: ; wherein, represents the limited contrast adaptive histogram equalization function.

[0027] Furthermore, step S22 is specifically: Perform multi-scale feature extraction on the enhanced two-dimensional grayscale image to obtain multi-scale features , which is expressed as: ; Among them, represents a lightweight convolutional layer for extracting local textures of cracks; represents an atrous spatial pyramid pooling operation for simultaneously capturing features of thin and wide cracks; Perform attention enhancement on the multi-scale features to obtain enhanced multi-scale features , which is expressed as: ; Among them, represents channel attention for suppressing image interference; Generate a crack mask image through a decoder , which is expressed as: ; Among them, represents the decoder, represents the Sigmoid activation function, Separate the crack mask image to obtain individual crack images and perform annotation, which is expressed as: ; Among them, represents the connected component labeling function for distinguishing adhered or crossed cracks, represents the counting label of individual crack images, represents a preset threshold.

[0028] Furthermore, step S23 is specifically: Perform morphological filtering on the crack mask image to obtain an optimized crack mask image, which is expressed as: ; Among them, represents closing operation; represents the dilation function for expanding the boundary of the crack image; represents the erosion function for shrinking the boundary of the crack image; represents the optimized crack mask image; Perform noise filtering on the optimized crack mask image to obtain a noise-filtered crack mask image , which is expressed as: ; Among them, represents the indicator function, Represents the connected component area of the optimized crack mask image of the input. Represents the height of the optimized crack mask image of the input. Represents the width of the optimized crack mask image of the input. Is a preset noise coefficient.

[0029] Further, step S24 is specifically as follows: Obtain the crack mask image after noise filtering The position coordinates of , which is expressed as: ; Among them, Represents The th crack coordinate of the pixel point in Represents the crack mask image after noise filtering Total number of pixels; Represents the connected component area of the optimized crack mask image of the input; Perform size calculation on the crack mask image after noise filtering To obtain the length , average width And maximum width , which is expressed as: ; ; ; Among them, Represents the th path point sequence generated by skeletonizing a single crack; Represents the th path point sequence generated by skeletonizing a single crack; Represents the Euclidean distance; Represents the ranging function; Perform quantity statistics on the crack mask image after noise filtering To obtain the number of cracks, which is expressed as: ; Among them, Represents the maximum value function.

[0030] Further, the process of determining whether to issue a risk warning based on a preset threshold is specifically as follows: Obtain the risk coefficient , which is expressed as: ; Among them, , and are all preset weight parameters, represents the preset length of the crack, represents the preset maximum width of the crack, represents the preset maximum number of cracks; Based on the risk coefficient obtain the risk type to perform hierarchical alarm, specifically: ; where 0, 1, and 2 are the labels corresponding to each risk type.

[0031] Furthermore, the risk scaling factor in S3 corresponds to the risk type of the region, specifically: When the risk type is a high-risk area, the risk scaling factor > 1; When the risk type is a medium-risk area, the risk scaling factor = 1; When the risk type is a low-risk area, the risk scaling factor < 1.

[0032] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A slope monitoring and early warning method based on oblique photography and radar technology, characterized in that Including: S1: Obtain the three-dimensional point cloud data of the slope through a multi-line lidar; obtain the surface texture data of the slope through an oblique photography UAV; The surface texture data includes the position, size, and quantity of cracks on the slope surface; S2: Construct a risk identification model, the model includes: a registration and fusion module, an image recognition module, and a comparison module; Input the three-dimensional point cloud data and the surface texture data into the registration and fusion module for registration and fusion to obtain the three-dimensional image data of the slope; Input the three-dimensional image data of the slope into the image recognition module for crack recognition to obtain a crack image, and the crack image includes the crack position, size, and quantity; Input the crack image into the comparison module, and judge the risk type and perform hierarchical warning based on a preset threshold; S3: Perform zonal scanning detection on the slope through a synthetic aperture radar, and optimize the zones based on the crack images in S2 to obtain an optimized detection area; dynamically adjust the scanning frequency of the synthetic aperture radar based on the risk type of the optimized detection area; Among them, obtaining the optimized detection area is expressed as: ； Among them, represents the optimized detection area; represents the position coordinates of the pixel points in the optimized detection area; represents the crack in the crack image of the central position coordinates, represents any crack in the crack image of the central position coordinates; represents the Euclidean distance; represents the crack in the corresponding crack image of the risk type of the risk scaling factor; represents any crack in the corresponding crack image of the risk type of the risk scaling factor, represents the risk scaling factor.

2. The slope monitoring and early warning method based on oblique photography and radar technology according to claim 1, characterized in that Specifically, obtaining the three-dimensional point cloud data of the slope through the multi-line lidar in step S1 is: Adopt a multi-line lidar, set the horizontal angular resolution, vertical angular resolution and point cloud density; obtain the three-dimensional point cloud data of the slope through multi-station scanning and stitching , expressed as: ； Among them, represents the -axis coordinate of the th point in the three-dimensional point cloud data, represents the -axis coordinate of the th point in the three-dimensional point cloud data, represents the -axis coordinate of the th point in the three-dimensional point cloud data; Specifically, obtaining the surface texture data of the slope through the oblique photography UAV in step S1 is: The oblique photography UAV flies crosswise at a preset inclination angle to generate an orthophoto image, enhances the crack texture of the orthophoto image through the HSV algorithm, and obtains the surface texture data of the slope.

3. The slope monitoring and early warning method based on oblique photography and radar technology according to claim 2, characterized in that, Specifically, obtaining the three-dimensional image data of the slope in step S2 is: Spatial registration, expressed as: ； Among them, represents the rotation matrix, represents the translation vector, represents the surface texture data; By minimizing the corresponding point distance error , solve the rotation matrix and translation vector, thereby geometrically aligning the surface texture data and the 3D point cloud data; Fuse the registered surface texture data and the three-dimensional point cloud data to obtain the three-dimensional image data of the slope.

4. A slope monitoring and early warning method based on oblique photography and radar technology according to claim 1, characterized in that In step S2, input the three-dimensional image data of the slope into the image recognition module for crack recognition to obtain a crack image, specifically: Step S21: Preprocess and enhance the data of the three-dimensional image data of the slope to obtain an enhanced two-dimensional grayscale image; Step S22: Perform multi-scale feature extraction, attention enhancement, and decoding on the enhanced two-dimensional grayscale image to obtain a crack mask image, that is, a crack image; in this process, obtain the counting label of the crack mask image; Step S23: Perform morphological closing operation on the crack mask image to optimize the crack boundary, and filter out noise based on the threshold of the connected domain area and the image size to finally obtain the noise-filtered crack mask image; Step S24: Obtain the position, size, and quantity of cracks based on the noise-filtered crack mask image.

5. The slope monitoring and early warning method based on oblique photography and radar technology according to claim 4, characterized in that Specifically, step S21 is: Preprocess the three-dimensional image data of the input slope, and unfold the three-dimensional image data into a two-dimensional grayscale image , to retain the brightness difference between the cracks and the image background, expressed as; ； wherein, are respectively the red channel value, the green channel value, and the blue channel value of the three-dimensional image data, are all preset weight parameters; By Algorithm to enhance the local contrast of a two-dimensional grayscale image to obtain an enhanced two-dimensional grayscale image , denoted as: ； Among them, represents the limited contrast adaptive histogram equalization function.

6. The slope monitoring and early warning method based on oblique photography and radar technology according to claim 5, wherein, Specifically, step S22 is: Enhance a two-dimensional grayscale image Perform multi-scale feature extraction to obtain multi-scale features , which is expressed as: ; Among them, represents a lightweight convolutional layer for extracting the local texture of cracks; represents an atrous spatial pyramid pooling operation for simultaneously capturing the features of thin and wide cracks; Perform attention enhancement on multi-scale features to obtain enhanced multi-scale features , which is expressed as: ； Among them, represents channel attention and is used to suppress image interference; Generate a crack mask image through the decoder , expressed as: ； Among them, represents a decoder, represents a Sigmoid activation function, Separate the crack mask image to obtain individual crack images and perform annotation, expressed as: ； Among them, represents a connected component labeling function, which is used to distinguish between adhered or crossed cracks. represents a counting label for a single crack image. represents a preset threshold.

7. A slope monitoring and early warning method based on oblique photography and radar technology according to claim 6, characterized in that, Specifically, step S23 is: For the crack mask image Perform morphological filtering to obtain the optimized crack mask image, denoted as: ; Among them, represents a closing operation; represents a dilation function for expanding the boundary of the crack image; represents an erosion function for shrinking the boundary of the crack image; represents the optimized crack mask image; Denoise the optimized crack mask image to obtain the denoised crack mask image , expressed as: ； Among them, represents the indicator function, represents the connected domain area of the optimized crack mask image of the input, represents the height of the optimized crack mask image of the input, represents the width of the optimized crack mask image of the input, is a preset noise factor.

8. A slope monitoring and early warning method based on oblique photography and radar technology according to claim 7, characterized in that, Specifically, step S24 is: Obtain the crack mask image after noise filtering of the position coordinates , expressed as: ； Among them, represents the th crack coordinate of the pixel point in represents the crack mask image after noise filtering total number of pixels; represents the connected component area of the input optimized crack mask image; For the filtered crack mask image perform size calculation to obtain the length of the crack , average width and maximum width , expressed as: ； ； ； Among them, represents the th sequence of path points generated by skeletonizing a single crack; represents the th sequence of path points generated by skeletonizing a single crack; represents the Euclidean distance; represents the ranging function; For the filtered crack mask image perform a quantity count to obtain the number of cracks , expressed as: ； Among them, represents the maximum value function.

9. The slope monitoring and early warning method based on oblique photography and radar technology according to claim 8, wherein, Specifically, the process of judging whether to perform risk warning based on a preset threshold is: Obtain the risk coefficient , expressed as: ； Among them, , and are all preset weight parameters, represents the preset length of the crack, represents the maximum width preset for the crack, represents the maximum number of preset cracks; Based on the risk coefficient Obtain the risk type To perform hierarchical alarm, specifically as follows: ； Among them, 0, 1, and 2 are the labels corresponding to each risk type.

10. A slope monitoring and warning method based on oblique photography and radar technology according to claim 9, characterized in that Specifically, the risk scaling factor in S3 corresponds to the risk type of the area, specifically: When the risk type is a high-risk area, the risk scaling factor > 1; When the risk type is a medium-risk area, the risk scaling factor = 1; When the risk type is a low-risk area, the risk scaling factor < 1.

Images