Multi-physics coupling slope monitoring method and system based on unmanned aerial vehicle technology

Abstract

The invention discloses a multi-physics field coupling slope monitoring method and system based on an unmanned aerial vehicle technology, and belongs to the technical field of slope monitoring. According to the method, contour recognition is carried out on the satellite remote sensing image, the accurate slope effective area can be obtained in the satellite remote sensing image in a large range, then feature recognition is carried out on the slope effective area, and therefore surface feature data of the slope are recognized; by judging the correlation between the physical fields and the mutability of the physical fields, simple physical field feature data which is more closely related to slope monitoring can be screened out, and slope monitoring misjudgment caused by massive physical field feature data is avoided; physical field feature data and surface feature data are subjected to data coupling, so that comprehensive coverage of slope monitoring feature data from macroscopic contour recognition to microcosmic physical field analysis is achieved, and it is ensured that a slope monitoring neural network model can output a more accurate slope monitoring result; and the accuracy of slope monitoring is improved.

Description

Technical Field

[0001] This invention belongs to the field of slope monitoring technology, and particularly relates to a slope monitoring method and system based on multi-physics coupling of unmanned aerial vehicle (UAV) technology. Background Technology

[0002] As a critical structure widely present in civil engineering, mining, water conservancy and hydropower, and transportation construction, slope stability plays a vital role in the safety of surrounding areas and facilities. Currently, protective works such as vegetation, retaining walls, and concrete facing are often added to the surface of slopes to further improve their stability. By monitoring slopes, the risk of slope instability can be identified and early warnings can be issued, thereby ensuring the safe operation of relevant projects and disaster prevention and mitigation.

[0003] Currently, slope monitoring commonly employs manual inspections and drone monitoring. While manual inspections of slope protection works by staff provide a direct view of the slope, the complex environment of slopes means that the risk of landslides or collapses is not always directly correlated with the protection works, leading to inaccuracies in monitoring. With the development of drone technology and neural network model prediction, more and more slope monitoring solutions utilize drones equipped with sensors to collect massive amounts of physical field data about the slope. This data is then used as the basis for monitoring via neural network models. However, because these massive physical field data often contain a large amount of redundant or irrelevant information, directly inputting this data into the neural network model can lead to misjudgments and inaccurate slope monitoring. Therefore, there is an urgent need for a multi-physics coupled slope monitoring method and system based on drone technology to overcome the shortcomings of existing technologies. Summary of the Invention

[0004] This invention aims to provide a slope monitoring method and system based on multi-physics coupling using UAV technology to solve the aforementioned technical problems. It obtains surface feature data through satellite remote sensing images, performs correlation and abrupt change analysis on the physical field monitoring data to obtain physical field feature data, and improves the accuracy of slope monitoring by coupling the physical field feature data and surface feature data.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide a slope monitoring method based on multi-physics coupling using unmanned aerial vehicle (UAV) technology, comprising: Based on the pre-set UAV technology, monitoring data of several physical fields of the slope are collected, and several satellite remote sensing images of the slope are obtained. The satellite remote sensing images are contoured according to a preset feature recognition algorithm to obtain the effective slope area of ​​each satellite remote sensing image, and the effective slope area is feature-recognized to obtain the surface feature data of the slope. Based on the preset physical field feature analysis method and the monitoring data, the correlation score between each pair of physical fields and the abrupt change score of each physical field are obtained. Based on the correlation score and the abrupt change score, the monitoring data of the physical fields are filtered to obtain the physical field feature data of the slope. Based on the preset multi-source heterogeneous data coupling method, and combined with the mutation score, the physical field feature data and surface feature data are coupled to obtain slope monitoring feature data. A slope monitoring neural network model is constructed based on the slope monitoring feature data, and the slope monitoring feature data is input into the slope monitoring neural network model to obtain the slope monitoring results.

[0006] Understandably, this invention, by performing contour recognition on satellite remote sensing images, can obtain precise effective slope areas within a large range of satellite remote sensing images. Then, feature recognition is performed on the effective slope areas to identify the surface feature data of the slope. Through physical field feature analysis methods and physical field monitoring data, the correlation between physical fields and the abrupt changes within the physical fields themselves are determined, thereby enabling the selection of more closely related and concise physical field feature data for slope monitoring, avoiding misjudgments caused by massive amounts of physical field feature data. By coupling physical field feature data and surface feature data, the slope monitoring feature data achieves comprehensive coverage from macroscopic contour recognition to microscopic physical field analysis, ensuring that the slope monitoring neural network model can output more accurate slope monitoring results; thus improving the accuracy of slope monitoring.

[0007] As a preferred embodiment, the step of performing contour recognition on the satellite remote sensing images according to a preset feature recognition algorithm to obtain the effective slope area of ​​each satellite remote sensing image, and performing feature recognition on the effective slope area to obtain the surface feature data of the slope, includes: acquiring the color channel value of each pixel in each satellite remote sensing image, and converting each satellite remote sensing image into a satellite remote sensing grayscale image based on the color channel value; identifying the first slope contour and the second slope contour of each satellite remote sensing grayscale image based on a preset sliding window and a preset convolution kernel, and determining the effective slope area of ​​each satellite remote sensing grayscale image based on the first slope contour and the second slope contour; and performing feature recognition on the effective slope area of ​​each satellite remote sensing grayscale image to obtain the surface feature data of the slope.

[0008] This preferred solution reduces noise interference and image complexity by converting satellite remote sensing images into satellite remote sensing grayscale images, while preserving key image features, thus improving the efficiency and reliability of subsequent image processing. Through the coordinated operation of sliding windows and convolution kernels, the accuracy of effective slope region extraction is significantly improved. Using the effective slope region avoids interference from background areas in the entire satellite remote sensing grayscale image for slope surface feature identification. Furthermore, locating the effective slope region reduces data processing volume, ensuring more accurate and efficient feature identification based on the effective slope region. This improves the accuracy of slope surface feature data, thereby enhancing the accuracy of subsequent multi-source heterogeneous data coupling and slope monitoring.

[0009] As a preferred embodiment, the step of identifying the first slope contour and the second slope contour of each satellite remote sensing grayscale image based on a preset sliding window and a preset convolution kernel, and determining the effective slope region of each satellite remote sensing grayscale image based on the first slope contour and the second slope contour, includes: sliding the preset sliding window sequentially across each satellite remote sensing grayscale image based on a preset sliding step frequency; after each slide, calculating the average grayscale value of all pixels within the sliding window and the grayscale value of the pixel corresponding to the center of the sliding window; if the grayscale value of the pixel corresponding to the center of the sliding window is greater than the average grayscale value of all pixels within the sliding window, then marking the pixel corresponding to the center of the sliding window as the first pixel, until the sliding window has completed sliding across each satellite remote sensing grayscale image, obtaining each The first set of pixels in each of the satellite remote sensing grayscale images is obtained; based on the first set of pixels in each of the satellite remote sensing grayscale images, a first slope contour of each of the satellite remote sensing grayscale images is obtained; a preset convolution kernel is convolved with each pixel in each of the satellite remote sensing grayscale images to obtain the response value of each pixel in each of the satellite remote sensing grayscale images; if the response value of the pixel is greater than a preset response threshold, the pixel is marked as a second pixel to obtain a second set of pixels in each of the satellite remote sensing grayscale images; based on the second set of pixels in each of the satellite remote sensing grayscale images, a second slope contour of each of the satellite remote sensing grayscale images is obtained; based on the first slope contour and the second slope contour of each of the satellite remote sensing grayscale images, the effective slope area of ​​each of the satellite remote sensing grayscale images is determined.

[0010] This preferred scheme determines the first slope contour using a sliding window and the second slope contour using a convolution kernel. The first and second slope contours are then combined to obtain the effective slope region, significantly improving the accuracy of the effective slope region. The linkage between the sliding window and the convolution kernel not only preserves the boundaries of abrupt grayscale changes but also captures high-frequency details of the convolution response, avoiding edge blurring and over-segmentation problems in traditional image segmentation algorithms. This results in a more complete, coherent, and accurate delineation of the effective slope region. Consequently, the accuracy of the slope's surface feature data is improved, which in turn enhances the accuracy of subsequent multi-source heterogeneous data coupling and slope monitoring.

[0011] As a preferred embodiment, the step of performing feature recognition on the effective slope area of ​​each satellite remote sensing grayscale image to obtain the surface feature data of the slope includes: constructing a grayscale histogram of the effective slope area of ​​each satellite remote sensing grayscale image based on the grayscale values ​​of all pixels in the effective slope area of ​​each satellite remote sensing grayscale image, and calculating the total number of pixels and the sum of grayscale values ​​of the effective slope area of ​​each satellite remote sensing grayscale image; determining several grayscale levels and several feature separation candidate thresholds for the effective slope area of ​​each satellite remote sensing grayscale image based on the grayscale histogram, wherein one grayscale level is associated with one feature separation candidate threshold. The following steps should be taken: Initialize the feature separation threshold, feature classification variance, cumulative pixel count, and cumulative grayscale value sum for the effective slope area of ​​each satellite remote sensing grayscale image; sequentially traverse the feature separation candidate thresholds for the effective slope area of ​​each satellite remote sensing grayscale image; in each traversal, based on the grayscale level corresponding to the currently traversed feature separation candidate threshold, update the cumulative pixel count and the cumulative grayscale value sum, and based on the updated cumulative grayscale value sum, updated cumulative pixel count, total number of pixels, and the grayscale value sum, calculate the target proportion, target average grayscale value, and non-target proportion of the currently traversed feature separation candidate threshold. The average grayscale value of the target and non-target components; based on the target ratio, target average grayscale value, non-target ratio, and non-target average grayscale value, calculate the separation variance of the currently traversed feature separation candidate thresholds; compare the separation variance of the currently traversed feature separation candidate thresholds with the current feature classification variance; if the separation variance of the currently traversed feature separation candidate thresholds is greater than the current feature classification variance, then update the feature separation threshold based on the currently traversed feature separation candidate thresholds, and update the feature classification variance based on the separation variance of the currently traversed feature separation candidate thresholds, until all feature separation candidate thresholds have been traversed; based on the slope of each of the satellite remote sensing grayscale images... The feature separation threshold obtained from the last update of the effective area is used as the optimal feature separation threshold for the effective area of ​​the slope in each satellite remote sensing grayscale image. Based on the optimal feature separation threshold for the effective area of ​​the slope in each satellite remote sensing grayscale image, the pixels of the effective area of ​​the slope are classified to obtain the protective engineering category pixels of the effective area of ​​the slope. Based on the number of protective engineering category pixels, the protective engineering integrity rate of the effective area of ​​the slope in each satellite remote sensing grayscale image is calculated. Based on the data sampling time of each satellite remote sensing grayscale image, the protective engineering integrity rate is sorted to generate the surface feature data of the slope.

[0012] This preferred scheme uses adaptive feature separation based on grayscale histograms to dynamically calculate the optimal feature separation threshold, enabling accurate identification of protective works within the effective area of ​​the slope, thus obtaining a more accurate protection work integrity rate (surface feature data). By iterating the feature classification variance and feature separation threshold using grayscale histograms and separation variance, and using variance to measure the feature separation threshold, the optimal feature separation threshold can be dynamically generated based on the actual situation of the effective area of ​​the slope in each satellite remote sensing grayscale image. This improves the classification accuracy of pixels in the protective works category, thereby improving the accuracy of surface feature data and subsequent slope monitoring.

[0013] As a preferred embodiment, the step of obtaining the correlation score between each pair of physical fields and the abrupt change score of each physical field based on the preset physical field feature analysis method and the monitoring data includes: calculating the average monitoring data of each physical field based on the monitoring data of each physical field; calculating the average difference of the monitoring data of each physical field based on the average monitoring data and the monitoring data; calculating the covariance of the monitoring data between each pair of physical fields based on the average difference of the monitoring data of each physical field, and calculating the standard deviation of the monitoring data of each physical field; determining the correlation score between each pair of physical fields based on the covariance and standard deviation of the monitoring data of each physical field; constructing an absolute difference sequence of the monitoring data of each physical field based on the monitoring data of each physical field, and calculating the variance of the monitoring data of each physical field based on the absolute difference sequence of the monitoring data; and determining the abrupt change score of each physical field based on the variance of the monitoring data of each physical field.

[0014] This preferred scheme achieves the screening of monitoring data from multiple different physical fields through joint analysis of correlation scores and abrupt change scores, avoiding slope monitoring errors caused by the use of massive physical field data in traditional methods. Correlation scores are calculated based on covariance and standard deviation to measure the strength of the association between two physical fields and identify the coupling relationship between different physical fields. Abrupt change scores are obtained based on absolute difference sequences and variance differences to measure the degree of abnormal fluctuation in the monitoring data of each physical field, focusing on the degree of change in the monitoring data of the physical field itself. The combination of correlation scores and abrupt change scores forms the basis for subsequent data screening, thereby improving the accuracy of physical field characteristic data and thus improving the accuracy of subsequent data coupling and slope monitoring.

[0015] As a preferred embodiment, the step of filtering the monitoring data of the physical field based on the correlation score and the abrupt change score to obtain the physical field characteristic data of the slope includes: calculating the mean abrupt change score based on the abrupt change score of each physical field; filtering the physical fields based on the mean abrupt change score and the abrupt change score of each physical field to obtain several abrupt change physical fields; constructing an undirected weighted complete graph with each abrupt change physical field as a node and the correlation score between every two abrupt change physical fields as edge weights; determining the optimal path length based on the correlation score between every two abrupt change physical fields; solving the undirected weighted complete graph based on a preset state compression dynamic programming algorithm and the optimal path length, with the goal of maximizing the sum of correlation scores, to obtain the optimal path; marking the abrupt change physical field corresponding to each node in the optimal path as the slope monitoring physical field, and obtaining the physical field characteristic data of the slope based on the monitoring data corresponding to the slope monitoring physical field.

[0016] This preferred scheme dynamically filters physical fields by using the mean of mutation scores to obtain abrupt physical fields. The degree of mutation in the monitoring data of a physical field can characterize the degree of change of the slope in the corresponding physical field, and the degree of change can affect the monitoring results of the slope. By using the mean of mutation scores to filter out physical fields with a higher degree of mutation, physical fields with high correlation to slope monitoring can be retained. Then, an undirected weighted complete graph is constructed based on the correlation scores. Through the state compression dynamic programming algorithm, highly correlated slope monitoring physical fields can be filtered out from the highly correlated physical fields (i.e., abrupt physical fields) of slope monitoring. This balances the abnormal fluctuations of a single physical field with the synergistic effect between multiple physical fields, ensuring that the slope monitoring physical fields have both high individual mutation and high overall correlation. This not only enables the final physical field characteristic data to comprehensively and accurately reflect the changes in the slope state, but also further simplifies the data, avoids redundant data, and significantly improves the accuracy of subsequent data coupling and slope monitoring.

[0017] As a preferred embodiment, the step of coupling the physical field feature data and surface feature data according to a preset multi-source heterogeneous data coupling method and in conjunction with the abrupt change score to obtain slope monitoring feature data includes: determining the abrupt change weight of each slope monitoring physical field based on the abrupt change score of each slope monitoring physical field; normalizing the monitoring data of each slope monitoring physical field to obtain normalized monitoring data of each slope monitoring physical field; and adjusting the normalization of the normalized monitoring data of each slope monitoring physical field based on the data sampling time of the normalized monitoring data of each slope monitoring physical field. Data is extracted from the monitoring data to generate an initial physical field monitoring vector for the slope at each data sampling time. Based on the abrupt change weight of each physical field of the slope monitoring, the initial physical field monitoring vector is weighted to obtain a first physical field monitoring vector for the slope at each data sampling time. Based on the data sampling time of the surface feature data, the surface feature data is concatenated with the first physical field monitoring vector to obtain a monitoring feature vector for the slope at each data sampling time. Based on the monitoring feature vector of the slope at each data sampling time, slope monitoring feature data is obtained.

[0018] In the data coupling process, this preferred scheme eliminates the dimensional differences in monitoring data from different physical fields through normalization, improving the overall stability and consistency of the data. By converting abrupt change scores into abrupt change weights, physical fields with significant abrupt changes can occupy a higher proportion in data coupling, thereby highlighting the potential hazards of the slope. By splicing surface feature data with the first physical field monitoring vector, a unified-dimensional monitoring feature vector can be formed, realizing spatiotemporal correlation. This effectively integrates the dynamic relationship between the slope surface protection engineering situation and the internal physical field monitoring data in the time dimension, avoiding information loss and improving the comprehensiveness and accuracy of slope monitoring feature data, thereby improving the accuracy of subsequent slope monitoring.

[0019] As a preferred embodiment, the step of constructing a slope monitoring neural network model based on the slope monitoring feature data and inputting the slope monitoring feature data into the slope monitoring neural network model to obtain slope monitoring results includes: filtering a preset training dataset based on the slope monitoring feature data to generate a slope monitoring training set; training a preset neural network model based on the slope monitoring training set and a preset slope label set to obtain a slope monitoring neural network model; and inputting the slope monitoring feature data into the slope monitoring neural network model to obtain slope monitoring results.

[0020] This preferred solution uses slope monitoring feature data to select a slope monitoring training set that better matches the actual situation of the slope. This reduces the interference of irrelevant training data on model training, improves the accuracy of the slope monitoring neural network model, and thus improves the accuracy of the slope monitoring results.

[0021] Accordingly, this invention provides a slope monitoring system based on UAV technology with multi-physics coupling, including: a multi-source heterogeneous data acquisition module, an image feature recognition module, a physical field feature recognition module, a multi-source heterogeneous data coupling module, and a slope monitoring module; The multi-source heterogeneous data acquisition module is used to collect monitoring data of several physical fields of the slope based on preset UAV technology, and to acquire several satellite remote sensing images of the slope. The image feature recognition module is used to perform contour recognition on the satellite remote sensing image according to a preset feature recognition algorithm to obtain the effective area of ​​the slope in each satellite remote sensing image, and to perform feature recognition on the effective area of ​​the slope to obtain the surface feature data of the slope. The physical field feature identification module is used to obtain the correlation score between each pair of physical fields and the abrupt change score of each physical field according to the preset physical field feature analysis method and the monitoring data, and to filter the monitoring data of the physical fields based on the correlation score and the abrupt change score to obtain the physical field feature data of the slope. The multi-source heterogeneous data coupling module is used to perform data coupling on the physical field feature data and surface feature data according to the preset multi-source heterogeneous data coupling method and in combination with the mutation score, so as to obtain slope monitoring feature data. The slope monitoring module is used to construct a slope monitoring neural network model based on the slope monitoring feature data, and input the slope monitoring feature data into the slope monitoring neural network model to obtain the slope monitoring results.

[0022] As a preferred embodiment, the image feature recognition module includes: an image feature recognition unit; the image feature recognition unit is used to acquire the color channel value of each pixel in each satellite remote sensing image, and convert each satellite remote sensing image into a satellite remote sensing grayscale image based on the color channel value; based on a preset sliding window and a preset convolution kernel, identify the first slope contour and the second slope contour of each satellite remote sensing grayscale image, and determine the effective slope area of ​​each satellite remote sensing grayscale image based on the first slope contour and the second slope contour; perform feature recognition on the effective slope area of ​​each satellite remote sensing grayscale image to obtain the surface feature data of the slope.

[0023] Understandably, this system, by performing contour recognition on satellite remote sensing images, can acquire precise effective slope areas within a large range of satellite remote sensing images. Then, it performs feature recognition on these effective slope areas to identify the surface feature data of the slope. Through physical field feature analysis methods and physical field monitoring data, it determines the correlation between physical fields and the abrupt changes within the physical fields themselves. This allows it to filter out more relevant and concise physical field feature data for slope monitoring, avoiding misjudgments caused by massive amounts of physical field feature data. By coupling physical field feature data and surface feature data, the system achieves comprehensive coverage of slope monitoring feature data, from macroscopic contour recognition to microscopic physical field analysis, ensuring that the slope monitoring neural network model can output more accurate slope monitoring results, thus improving the accuracy of slope monitoring. Attached Figure Description

[0024] Figure 1 A flowchart illustrating the steps of a slope monitoring method based on multi-physics coupling using unmanned aerial vehicle (UAV) technology, provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a slope monitoring system based on multi-physics coupling using unmanned aerial vehicle (UAV) technology, provided as an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example 1 Please refer to Figure 1 , Figure 1 The flowchart of a slope monitoring method based on multi-physics coupling using UAV technology provided in this embodiment of the invention includes steps S101 to S105.

[0027] Step S101: Based on the preset UAV technology, collect monitoring data of several physical fields of the slope and acquire several satellite remote sensing images of the slope.

[0028] It should be noted that Unmanned Aerial Vehicle Technology (Drone Technology) refers to the technology that utilizes unmanned aerial vehicles and related systems. It typically includes the drone platform itself, flight control system, navigation system, payload (such as cameras and sensors), data link communication system, and ground control station. In slope monitoring, drones primarily serve as flexible and efficient aerial platforms, carrying various sensors to collect information at close range with high precision. A physical field refers to a continuously distributed physical quantity in space. These physical quantities can be data such as temperature, pressure, density, displacement, velocity, stress, and seepage rate. Monitoring changes in the physical field is central to slope monitoring. Satellite remote sensing imagery refers to the use of devices such as optical cameras, radar, and infrared sensors to detect and record the electromagnetic wave energy reflected or emitted by the Earth's surface and process it into visualized image data.

[0029] In an optional embodiment, the data sampling interval is set to 10 minutes, that is, the interval between every two data sampling times is 10 minutes; a sensor is mounted on the drone (the type of sensor can be determined based on the data of the physical field to be collected, which is not limited in this embodiment), and data is collected every 10 minutes based on drone technology, thereby obtaining monitoring data of several physical fields; similarly, satellite remote sensing images are collected every 10 minutes. In this embodiment, the resolution of the satellite remote sensing images is set to 1024×960, that is, there are 1024 pixels in each row and 960 pixels in each column. The rows and columns of each collected satellite remote sensing image are aligned, and the position of each pixel in the satellite remote sensing image can be determined, that is, the pixel in which row and column.

[0030] Step S102: Perform contour recognition on the satellite remote sensing images according to the preset feature recognition algorithm to obtain the effective slope area of ​​each satellite remote sensing image, and perform feature recognition on the effective slope area to obtain the surface feature data of the slope.

[0031] In this embodiment, the step of performing contour recognition on the satellite remote sensing images according to a preset feature recognition algorithm to obtain the effective slope area of ​​each satellite remote sensing image, and performing feature recognition on the effective slope area to obtain the surface feature data of the slope, includes: Obtain the color channel value of each pixel in each of the satellite remote sensing images, and convert each of the satellite remote sensing images into a satellite remote sensing grayscale image based on the color channel value; Based on a preset sliding window and a preset convolution kernel, the first slope contour and the second slope contour of each satellite remote sensing grayscale image are identified, and the effective slope area of ​​each satellite remote sensing grayscale image is determined based on the first slope contour and the second slope contour. Feature identification is performed on the effective area of ​​the slope in each of the aforementioned satellite remote sensing grayscale images to obtain the surface feature data of the slope.

[0032] This embodiment reduces noise interference and image complexity by converting satellite remote sensing images into satellite remote sensing grayscale images, while preserving key image features, thus improving the efficiency and reliability of subsequent image processing. Through the coordinated operation of sliding windows and convolution kernels, the accuracy of effective slope region extraction is significantly improved. Using the effective slope region avoids interference from background areas in the entire satellite remote sensing grayscale image for slope surface feature identification. Furthermore, locating the effective slope region reduces data processing volume, ensuring more accurate and efficient feature identification based on the effective slope region. This improves the accuracy of slope surface feature data, thereby enhancing the accuracy of subsequent multi-source heterogeneous data coupling and slope monitoring.

[0033] In one optional embodiment, the color channel value of each pixel in the satellite remote sensing image refers to the RGB color space, that is, each pixel has three color channels, namely red (R), green (G) and blue (B), and the color channel value ranges from 0 to 255; at this time, the color channel value with the largest value is selected as the grayscale value of the pixel; thereby converting the satellite remote sensing image into a satellite remote sensing grayscale image.

[0034] In this embodiment, the step of identifying the first slope contour and the second slope contour of each satellite remote sensing grayscale image based on a preset sliding window and a preset convolution kernel, and determining the effective slope region of each satellite remote sensing grayscale image based on the first slope contour and the second slope contour, includes: Based on a preset sliding step frequency, the preset sliding window is sequentially slid across each of the aforementioned satellite remote sensing grayscale images; After each slide, the average gray value of all pixels in the sliding window and the gray value of the pixel corresponding to the center of the sliding window are calculated. If the gray value of the pixel corresponding to the center of the sliding window is greater than the average gray value of all pixels in the sliding window, the pixel corresponding to the center of the sliding window is marked as the first pixel. This process continues until the sliding window is completed on each satellite remote sensing grayscale image, resulting in the first set of pixels for each satellite remote sensing grayscale image. Based on the first set of pixels in each of the satellite remote sensing grayscale images, the first slope profile of each of the satellite remote sensing grayscale images is obtained; A preset convolution kernel is convolved with each pixel of each of the satellite remote sensing grayscale images to obtain the response value of each pixel in each of the satellite remote sensing grayscale images; If the response value of the pixel is greater than the preset response threshold, the pixel is marked as the second pixel, thus obtaining the second pixel set for each of the satellite remote sensing grayscale images; Based on the second pixel set of each of the satellite remote sensing grayscale images, the second slope profile of each of the satellite remote sensing grayscale images is obtained; Based on the first and second slope contours of each of the satellite remote sensing grayscale images, the effective slope area of ​​each of the satellite remote sensing grayscale images is determined.

[0035] This embodiment determines the first slope contour using a sliding window and the second slope contour using a convolution kernel. The first and second slope contours are then combined to obtain the effective slope region, significantly improving the accuracy of the effective slope region. The linkage between the sliding window and the convolution kernel not only preserves the boundaries of abrupt grayscale changes but also captures high-frequency details of the convolution response, avoiding edge blurring and over-segmentation problems in traditional image segmentation algorithms. This results in a more complete, coherent, and accurate delineation of the effective slope region, thereby improving the accuracy of the slope's surface feature data and consequently enhancing the accuracy of subsequent multi-source heterogeneous data coupling and slope monitoring.

[0036] In one optional embodiment, a preset sliding window is obtained. The size of the sliding window is 3×3, indicating that there are theoretically 9 pixels in the window. The sliding step frequency is 1, that is, one pixel is slid at a time. The center of the sliding window is aligned with the pixels in the first row and first column of the satellite remote sensing grayscale image, and the sliding is performed in order from left to right and from top to bottom. After each slide, the grayscale values ​​of all pixels in the sliding window are summed and averaged. In particular, there are cases where there are no pixels in the sliding window. For example, when the center of the sliding window is sliding in the first row, the grayscale value of the position with no pixels is set to 0, so as to ensure that the grayscale values ​​of 9 grids are summed. If the grayscale value of the pixel corresponding to the center of the sliding window is greater than the average grayscale value of all pixels in the sliding window, the pixel corresponding to the center of the sliding window is marked as the first pixel. The sliding window is continued to slide on each of the satellite remote sensing grayscale images until the first set of pixels in each satellite remote sensing grayscale image is obtained. Then, the contour can be drawn on the satellite remote sensing grayscale image using all the first pixels in the first set of pixels, so as to obtain the first slope contour.

[0037] In one optional embodiment, a convolution operation is a mathematical operation that uses a sliding window mechanism to perform element-wise multiplication and summation of the input data and the convolution kernel. The default convolution kernel is also set to 3×3, specifically: Similar to the sliding window process, if a situation with no pixels is encountered, such as in the first row, the grayscale value of the position with no pixels is set to 0. The convolution kernel is used to traverse each pixel to obtain the response value of each pixel. In this embodiment, the response threshold is set to 4. If the response value of a pixel is greater than 4, the pixel is marked as the second pixel, and the second pixel set of each satellite remote sensing grayscale image is obtained. Then, the contour can be drawn on the satellite remote sensing grayscale image through all the second pixels in the second pixel set, thereby obtaining the second slope contour.

[0038] In an optional embodiment, based on the first slope profile and the second slope profile of each of the satellite remote sensing grayscale images, the effective slope area of ​​each satellite remote sensing grayscale image is determined, including: comparing the area corresponding to the first slope profile and the area corresponding to the second slope profile to obtain the overlapping area between the first slope profile and the second slope profile, and taking the overlapping area as the effective slope area of ​​the satellite remote sensing grayscale image. Furthermore, since the first and second slope contours are drawn on the same satellite remote sensing grayscale image, it is only necessary to find the overlapping areas by identifying the repeated pixels within the regions enclosed by the first and second slope contours.

[0039] In this embodiment, the step of performing feature recognition on the effective area of ​​the slope in each of the satellite remote sensing grayscale images to obtain the surface feature data of the slope includes: Based on the gray values ​​of all pixels in the effective area of ​​the slope in each of the satellite remote sensing grayscale images, a grayscale histogram of the effective area of ​​the slope in each of the satellite remote sensing grayscale images is constructed, and the total number of pixels and the sum of gray values ​​in the effective area of ​​the slope in each of the satellite remote sensing grayscale images are calculated. Based on the grayscale histogram, several grayscale levels and several feature separation candidate thresholds are determined for the effective area of ​​the slope in each satellite remote sensing grayscale image, wherein one grayscale level corresponds to one feature separation candidate threshold. Initialize the feature separation threshold, feature classification variance, cumulative pixel count, and cumulative grayscale value of the effective area of ​​the slope in each of the aforementioned satellite remote sensing grayscale images; The feature separation candidate thresholds for the effective area of ​​the slope in each of the aforementioned satellite remote sensing grayscale images are sequentially traversed; In each traversal, based on the gray level corresponding to the feature separation candidate threshold of the current traversal, the cumulative number of pixels and the total cumulative gray value are updated, and based on the updated total cumulative gray value, the updated cumulative number of pixels, the total number of pixels and the total gray value, the target proportion, target average gray value, non-target proportion and non-target average gray value of the feature separation candidate threshold of the current traversal are calculated. The separation variance of the current feature separation candidate threshold is calculated based on the target ratio, target average gray value, non-target ratio, and non-target average gray value. Compare the separation variance of the currently traversed feature separation candidate thresholds with the current feature classification variance. If the separation variance of the currently traversed feature separation candidate thresholds is greater than the current feature classification variance, then update the feature separation threshold based on the currently traversed feature separation candidate thresholds, and update the feature classification variance based on the separation variance of the currently traversed feature separation candidate thresholds, until all feature separation candidate thresholds have been traversed. The feature separation threshold obtained from the last update of the effective slope area in each of the satellite remote sensing grayscale images is used as the optimal feature separation threshold for the effective slope area in each of the satellite remote sensing grayscale images. Based on the optimal feature separation threshold of the effective slope area in each satellite remote sensing grayscale image, the pixels of the effective slope area are classified to obtain the protective engineering type pixels of the effective slope area, and the protective engineering integrity rate of the effective slope area in each satellite remote sensing grayscale image is calculated based on the number of the protective engineering type pixels. Based on the data sampling time of each satellite remote sensing grayscale image, the integrity rate of the protection project is sorted to generate the surface feature data of the slope.

[0040] It should be noted that a histogram is a statistical chart used to display the distribution of continuous data. It uses a series of bars of varying heights to represent the frequency or relative frequency of data within different intervals, visually reflecting the central tendency, dispersion, and distribution shape of the data.

[0041] In one optional embodiment, a grayscale histogram is constructed based on the grayscale values ​​of all pixels within the effective area of ​​the slope, with the horizontal axis representing grayscale values ​​and the vertical axis representing the number of pixels. After constructing the grayscale histogram, the total number of pixels within the effective area of ​​the slope is counted. Then, the gray values ​​of each pixel are summed to obtain the total gray values ​​of the effective area of ​​the slope. Based on the horizontal axis of the grayscale histogram, several grayscale levels (0 to 255) and several feature separation candidate thresholds (0 to 255) are determined, with one grayscale level corresponding to one feature separation candidate threshold; then, the feature separation thresholds for the effective slope area of ​​each of the aforementioned satellite remote sensing grayscale images are initialized. =0, feature classification variance =0, cumulative pixel count =0 and the sum of accumulated gray values =0; Then, starting from 0 to 255, the candidate thresholds for feature separation are traversed. In each traversal, the number of pixels at the gray level corresponding to the current candidate threshold for feature separation is compared with the cumulative number of pixels. Add them together to update the cumulative pixel count. The sum of the gray values ​​of all pixels corresponding to the gray level of the current feature separation candidate threshold is compared with the cumulative sum of gray values. Add them together to update the cumulative grayscale value sum. ; The updated cumulative pixel count will then be used. Divide by the total number of pixels The proportion of non-target thresholds for feature separation candidates in the current iteration is obtained. non-target ratio Non-target average gray value This is the sum of the updated cumulative grayscale values. Divide by cumulative pixel count ,Right now Target ratio The non-target ratio is 1 minus the target ratio. Similarly, the target average gray value The calculation process is as follows: ; Then, based on the target proportion, target average gray value, non-target proportion, and non-target average gray value, the separation variance of the current feature separation candidate threshold is calculated. , If the separation variance of the current feature separation candidate threshold is... Greater than the feature classification variance Then the separation variance of the current feature separation candidate threshold will be used. As a feature classification variance The new value updates the feature classification variance. Furthermore, the current feature separation candidate threshold is used as the feature separation threshold. This updates the feature separation threshold. Repeat the above operation until all candidate thresholds for feature separation have been traversed; then update the feature separation threshold obtained from the last update. , as the optimal feature separation threshold; Then, based on the optimal feature separation threshold, if the gray value of a pixel in the effective area of ​​the slope is greater than the optimal feature separation threshold, then the pixel is a protective engineering pixel, thus obtaining the number of protective engineering pixels in the effective area of ​​the slope; the number of protective engineering pixels is then divided by the total number of pixels. The integrity rate of the protective project is obtained; then, based on the data sampling time of each satellite remote sensing grayscale image, the integrity rates of the protective project are sorted in order from first to last to generate the surface feature data of the slope.

[0042] It should be noted that the process of solving the optimal feature separation threshold described in this embodiment will be further explained here. Within the effective area of ​​the slope, the purpose of this embodiment is to solve the coverage rate of the protective engineering. Therefore, it is necessary to identify the pixels corresponding to the protective engineering (i.e., protective engineering class pixels). Thus, the protective engineering class pixels are the target (target proportion, target average gray value) to be solved, while any non-protective engineering class pixels are all classified as non-target (non-target proportion, non-target average gray value). Based on this, variance is used to characterize the dispersion of data distribution. This embodiment introduces the idea of ​​variance. When a certain gray value is used as a threshold, the larger the variance between the target and non-target, the greater the separation degree between the protective engineering class pixels and the non-protective engineering class pixels, that is, the gray value can maximize the separation degree. Therefore, the process of solving the optimal feature separation threshold is to solve for a gray value as the boundary between the non-protective engineering class pixels and the protective engineering class pixels, thereby maximizing the separation degree between the non-protective engineering class pixels and the protective engineering class pixels. Furthermore, we introduce the cumulative pixel count and the cumulative sum of grayscale values, initialized to 0. Therefore, we iterate from 0 to 255. Taking a grayscale value of 0 as an example, we assume the current feature separation candidate threshold is the optimal feature separation threshold, i.e., the optimal feature separation threshold is 0. Then, pixels with grayscale values ​​greater than 0 within the effective area of ​​the slope are considered protective engineering pixels (i.e., targets), while pixels with grayscale values ​​less than or equal to 0 are considered non-protective engineering pixels. We can then further calculate the cumulative pixel count. and the sum of cumulative gray values Cumulative number of pixels This refers to the number of non-protected engineering pixels (i.e., the total grayscale values ​​of non-target pixels), and the cumulative total grayscale values. This refers to the sum of grayscale values ​​of non-protected engineering pixels (i.e., the sum of grayscale values ​​of non-target pixels), and therefore can be calculated based on the formula. To calculate the proportion of non-protection engineering pixels to all pixels in the effective area of ​​the slope, based on the formula... To calculate the average grayscale value of pixels in the non-protective engineering area, we can use the following method: Similarly, knowing the proportion of non-protective engineering pixels to all pixels in the effective area of ​​the slope, we can calculate the proportion of protective engineering pixels to all pixels in the effective area of ​​the slope (i.e., the target proportion). It can also calculate the average gray value of pixels in the protection project (i.e., the target average gray value). Therefore, the separation variance can be solved at this point. Using separated variance The magnitude represents the degree of separation between pixels belonging to the protective engineering category and those belonging to non-protective engineering categories, assuming a grayscale value of 0 is used as the optimal feature separation threshold; the separation variance is also considered. A larger value indicates a higher degree of separation; in this case, the separation variance is compared. and feature classification variance If the variance is separated Greater than the feature classification variance This indicates a higher degree of separation, requiring an update to the feature classification variance. In particular, due to the variance of feature classification Since it is initialized to 0, the feature classification variance will definitely be updated when iterating through grayscale values ​​of 0. and feature separation threshold ; Next, assuming the optimal feature separation threshold is a grayscale value of 1, pixels with a grayscale value greater than 1 within the effective area of ​​the slope are considered protective engineering pixels (i.e., targets), while pixels with a grayscale value less than or equal to 1 are considered non-protective engineering pixels. Since we have already traversed the case where the optimal feature separation threshold is a grayscale value of 0, we only need to compare the number of pixels with a grayscale value of 1 with the cumulative number of pixels. By adding them together, we can directly obtain the number of pixels with a grayscale value less than or equal to 1; the cumulative number of pixels. Similarly, by analogy, we can obtain the separation variance when the optimal feature separation threshold is a grayscale value of 1. If the separation variance at this time Greater than the feature classification variance This indicates that when the grayscale value is 1, the separation degree between protective engineering pixels and non-protective engineering pixels is higher than when the grayscale value is 0. Therefore, a grayscale value of 1 is more suitable as the optimal feature separation threshold. In this case, the feature separation threshold should be... Updated to 1, and the feature classification variance is also updated. Updated to the optimal feature separation threshold is the separation variance when the gray value is 1. This process continues until all grayscale values ​​are 255, thus allowing us to determine the largest feature classification variance. This leads to the corresponding optimal feature separation threshold. Thus, the optimal feature separation threshold was obtained.

[0043] This embodiment utilizes adaptive feature separation based on grayscale histograms to dynamically calculate the optimal feature separation threshold, enabling accurate identification of protective works within the effective area of ​​a slope. This results in a more accurate protection work integrity rate (surface feature data). By iterating over the feature classification variance and feature separation threshold using grayscale histograms and separation variance, and measuring the feature separation threshold with variance, the optimal feature separation threshold can be dynamically generated based on the actual situation of the effective slope area in each satellite remote sensing grayscale image. This improves the classification accuracy of protective works pixels, thereby enhancing the accuracy of surface feature data and subsequent slope monitoring.

[0044] Step S103: Based on the preset physical field feature analysis method and the monitoring data, obtain the correlation score between each pair of physical fields and the abrupt change score of each physical field. Based on the correlation score and the abrupt change score, filter the monitoring data of the physical fields to obtain the physical field feature data of the slope.

[0045] In this embodiment, obtaining the correlation score between each pair of physical fields and the abrupt change score of each physical field based on the preset physical field feature analysis method and the monitoring data includes: Based on the monitoring data for each of the physical fields, calculate the average value of the monitoring data for each of the physical fields; Based on the average value of the monitoring data and the monitoring data, calculate the average difference of the monitoring data for each physical field; Based on the average difference of the monitoring data for each of the physical fields, the covariance of the monitoring data between each pair of physical fields is calculated, and the standard deviation of the monitoring data for each of the physical fields is calculated. Based on the covariance of the monitoring data between each pair of physical fields and the standard deviation of the monitoring data for each physical field, a correlation score is determined between each pair of physical fields. Based on the monitoring data of each physical field, an absolute difference sequence of monitoring data for each physical field is constructed, and the variance of the monitoring data for each physical field is calculated based on the absolute difference sequence of monitoring data. Based on the variance of the monitoring data for each physical field, a mutation score for each physical field is determined.

[0046] In one optional embodiment, the monitoring data for each physical field are summed and then averaged to obtain the average monitoring data for each physical field; then, based on the average monitoring data and the monitoring data, the average difference of the monitoring data for each physical field is calculated, followed by the calculation of the covariance of the monitoring data between each pair of physical fields, and the calculation of the standard deviation of the monitoring data for each physical field, thereby obtaining the correlation score between each pair of physical fields; for example, assuming Indicates the first Such physical fields Indicates the first Such physical fields, therefore use Indicates the first The physical field in the first Monitoring data from each data sampling time, using Indicates the first The physical field in the first Monitoring data for each data sampling time; Indicates the first Monitoring data for each data sampling time; The number of data sampling times represents the average values ​​of the monitoring data for the two physical fields, respectively. and Therefore, the first Type of physical field and the first The average difference in monitoring data of various physical fields can be expressed as and ;No. Type of physical field and the first The covariance of monitoring data between different physical fields can be calculated using the following formula: ;No. The standard deviation of monitoring data for a certain physical field can be calculated using the following formula: , No. The standard deviation of monitoring data for a certain physical field can be calculated using the following formula: ; Then the numerator of the covariance of the monitoring data (i.e. The correlation score between the two physical fields is calculated using the standard deviation of the monitoring data. For example, the first... Type of physical field and the first Correlation score between physical fields The calculation formula is: ; Among them, the correlation score The range is The closer the value is to 1, the higher the correlation between the two physical fields. Then, based on the monitoring data of each physical field, an absolute difference sequence of monitoring data for each physical field is constructed, and the variance of the monitoring data for each physical field is calculated based on the absolute difference sequence of monitoring data. Based on the variance of the monitoring data for each physical field, the abruptness score of each physical field is determined; for example, assuming the first... Monitoring data of various physical fields Then, the absolute difference sequence of its monitoring data is the sum of the difference between the previous and subsequent data, and the absolute value of the difference. The absolute difference sequence of monitoring data for various physical fields is Then, the variance of the absolute difference sequence of the monitoring data is calculated as the variance of the monitoring data. For example, the variance of its monitoring data is The variance of the monitoring data is used as the first... The abrupt change fraction of a physical field.

[0047] This embodiment achieves the screening of monitoring data from multiple different physical fields through joint analysis of correlation scores and abrupt change scores, avoiding slope monitoring errors caused by the use of massive physical field data in traditional methods. Correlation scores are calculated based on covariance and standard deviation to measure the strength of the association between two physical fields and identify the coupling relationship between different physical fields. Abrupt change scores are obtained based on absolute difference sequences and variance differences to measure the degree of abnormal fluctuation in the monitoring data of each physical field, focusing on the degree of change in the monitoring data of the physical field itself. The combination of correlation scores and abrupt change scores forms the basis for subsequent data screening, thereby improving the accuracy of physical field characteristic data and thus improving the accuracy of subsequent data coupling and slope monitoring.

[0048] In this embodiment, the step of filtering the monitoring data of the physical field based on the correlation score and the abrupt change score to obtain the physical field characteristic data of the slope includes: The mean aberration score is calculated based on the aberration score of each of the physical fields; Based on the mean of the mutation score and the mutation score of each physical field, the physical fields are screened to obtain a number of mutated physical fields; Using each of the aforementioned aberrant physical fields as nodes and the correlation score between any two of the aforementioned aberrant physical fields as edge weights, an undirected weighted complete graph is constructed; The optimal path length is determined based on the correlation score between each of the two described abrupt physical fields; Based on the preset state compression dynamic programming algorithm and the optimal path length, the undirected weighted complete graph is solved with the goal of maximizing the sum of relevance scores to obtain the optimal path; The abrupt change physical field corresponding to each node in the optimal path is marked as the slope monitoring physical field, and the physical field characteristic data of the slope is obtained based on the monitoring data corresponding to the slope monitoring physical field.

[0049] In one optional embodiment, the mean of mutation scores is calculated based on the mutation scores of all physical fields. Physical fields with mutation scores greater than the mean mutation score are selected and marked as mutation physical fields. Using mutation physical fields as nodes, and the correlation score between any two mutation physical fields as the weight of the edge of the node (i.e., edge weight), an undirected weighted complete graph is constructed. The correlation scores between any two mutation physical fields are accumulated, that is, all edge weights in the undirected weighted complete graph are added together to obtain the sum of edge weights. The sum of edge weights is divided by the number of nodes, rounded down, and incremented by 1 to obtain the optimal path length. For example, assuming the sum of edge weights is... The number of nodes (the number of abrupt physical fields) is Since the range of values ​​for the correlation score has already been described above... Therefore, the total edge weight The range of values ​​is Therefore, the optimal path length The calculation formula is: ;in, Indicates to The value is rounded down; after obtaining the optimal path length, the optimal path is obtained by solving the problem based on the state compression dynamic programming algorithm with the goal of maximizing the sum of relevance scores (i.e., maximizing the sum of edge weights between nodes in the optimal path); Then, the abrupt physical fields corresponding to the nodes in the optimal path are marked as slope monitoring physical fields, and their corresponding monitoring data are used as the physical field characteristic data of the slope.

[0050] It is important to note that in the extreme case where only one abrupt change physical field is selected, the optimal path length, calculated using the formula for optimal path length, is 1, meaning this abrupt change physical field is a slope monitoring physical field. Specifically, since the selection is based on the mean, there will be no situation where no abrupt change physical field exists. In the extreme case where all edge weights of the undirected weighted complete graph are 1, the sum of the edge weights is... The value is Optimal path length The value is In an undirected weighted complete graph, if all edge weights are 0, then the total edge weights are... The value is Optimal path length The value is .

[0051] It should be noted that State Compression Dynamic Programming is a dynamic programming technique used to solve problems with exponentially large state spaces where the state representation can be compressed.

[0052] This embodiment dynamically filters physical fields by using the mean of mutation scores to obtain mutable physical fields. The degree of mutation in the monitoring data of a physical field can characterize the degree of change of the slope in the corresponding physical field, and the degree of change can affect the monitoring results of the slope. By using the mean of mutation scores to filter out physical fields with a higher degree of mutation, physical fields with high correlation to slope monitoring can be retained. Then, an undirected weighted complete graph is constructed based on the correlation scores. Through the state compression dynamic programming algorithm, highly correlated slope monitoring physical fields can be filtered out from the highly correlated physical fields (i.e., mutable physical fields) of slope monitoring. This balances the abnormal fluctuations of a single physical field with the synergistic effect between multiple physical fields, ensuring that the slope monitoring physical fields have both high individual mutation and high overall correlation. This not only enables the final physical field characteristic data to comprehensively and accurately reflect the changes in the slope state, but also further simplifies the data, avoids redundant data, and significantly improves the accuracy of subsequent data coupling and slope monitoring.

[0053] Step S104: Based on the preset multi-source heterogeneous data coupling method, and combined with the mutation score, perform data coupling on the physical field feature data and surface feature data to obtain slope monitoring feature data.

[0054] In this embodiment, the step of coupling the physical field feature data and surface feature data according to the preset multi-source heterogeneous data coupling method and in combination with the abrupt change score to obtain slope monitoring feature data includes: Based on the abrupt change score of each of the slope monitoring physical fields, the abrupt change weight of each of the slope monitoring physical fields is determined; The monitoring data of each of the slope monitoring physical fields are normalized to obtain normalized monitoring data of each of the slope monitoring physical fields; Based on the data sampling time of the normalized monitoring data of each slope monitoring physical field, data extraction is performed on the normalized monitoring data of each slope monitoring physical field to generate the initial physical field monitoring vector of the slope at each data sampling time. Based on the abrupt change weight of each of the slope monitoring physical fields, the initial physical field monitoring vector is weighted to obtain the first physical field monitoring vector of the slope at each data sampling time. Based on the data sampling time of the surface feature data, the surface feature data is concatenated with the first physical field monitoring vector to obtain the monitoring feature vector of the slope at each data sampling time; Based on the monitoring feature vector of the slope at each data sampling time, slope monitoring feature data are obtained.

[0055] In one optional embodiment, the abrupt change score of the slope monitoring physical field is divided by the sum of the abrupt change scores of all slope monitoring physical fields to obtain the abrupt change weight of the slope monitoring physical field; then, the monitoring data of the slope monitoring physical field is normalized using the Min-Max Normalization method. The range is determined to obtain normalized monitoring data of the slope monitoring physical field; then, based on the data sampling time, an initial physical field monitoring vector for the slope is constructed at each data sampling time; for example, assuming the first... Type of physical field, the first Type of physical field, the first The data sampling time and the first The data sampling time; the first data sampling time; The physical field in the first The data sampling time and the first The monitoring data from each sampling time, after normalization, are as follows: and ;No. The physical field in the first The data sampling time and the first The monitoring data from each sampling time, after normalization, are as follows: and Therefore, the slope in the first... The initial physical field monitoring vector for each data sampling time is In the The initial physical field monitoring vector for each data sampling time is Then, the abrupt change weights of the slope monitoring physical field are multiplied by the corresponding data in the initial physical field monitoring vector to obtain the first physical field monitoring vector; for example, assume the first... The mutation weight of the physical field is , No. The mutation weight of the physical field is The slope in the first The first physical field monitoring vector for each data sampling time is ; slope in the first The first physical field monitoring vector for each data sampling time is Then, based on the data sampling time, the surface feature data (i.e., the coverage rate of the protective engineering) is concatenated with the first physical field monitoring vector to obtain the monitoring feature vector of the slope at each data sampling time; for example, assuming that at the first data sampling time... The surface feature data for each data sampling time is In the The surface feature data for each data sampling time is ; then in the first The monitoring feature vector for each data sampling time is ; in the The monitoring feature vector for each data sampling time is .

[0056] In this embodiment, normalization processing eliminates dimensional differences in monitoring data from different physical fields during data coupling, improving overall data stability and consistency. By converting abrupt change scores into abrupt change weights, physical fields with significant abrupt changes occupy a higher proportion in data coupling, thus highlighting the potential hazards of slopes. By concatenating surface feature data with the first physical field monitoring vector, a unified-dimensional monitoring feature vector is formed, achieving spatiotemporal correlation. This effectively integrates the dynamic relationship between the slope surface protection engineering and the internal physical field monitoring data in the time dimension, avoiding information loss and improving the comprehensiveness and accuracy of slope monitoring feature data, thereby enhancing the accuracy of subsequent slope monitoring.

[0057] Step S105: Construct a slope monitoring neural network model based on the slope monitoring feature data, and input the slope monitoring feature data into the slope monitoring neural network model to obtain the slope monitoring results.

[0058] In this embodiment, the step of constructing a slope monitoring neural network model based on the slope monitoring feature data, and inputting the slope monitoring feature data into the slope monitoring neural network model to obtain slope monitoring results includes: Based on the slope monitoring feature data, a preset training dataset is selected to generate a slope monitoring training set; Based on the slope monitoring training set and the preset slope label set, the preset neural network model is trained to obtain the slope monitoring neural network model; The slope monitoring feature data is input into the slope monitoring neural network model to obtain the slope monitoring results.

[0059] In an optional embodiment, generating a slope monitoring training set by filtering a preset training dataset based on the slope monitoring feature data means selecting data from the preset training dataset that has the same data type as the slope monitoring feature data as the slope monitoring training set; for example, assuming the data type of the slope monitoring feature data includes the first... Type of physical field, the first The coverage of physical fields and protective engineering is then selected from the preset training dataset. Type of physical field, the first Training data on the physical field and the coverage of protective engineering are used to improve the training effect and robustness of the preset neural network model. The preset slope label set essentially represents the slope monitoring results. In this embodiment, it includes three types of labels: the first label indicates a landslide risk requiring immediate maintenance and early warning; the second label indicates a defect requiring maintenance; and the third label indicates no landslide risk or defect, requiring no maintenance or early warning. The preset neural network model is trained based on the slope label set, so that the final output of the slope neural network model is one of the slope label sets, which is then used as the slope monitoring result. A pre-defined neural network model is a computational model inspired by biological nervous systems. It processes complex data by simulating the connections and signal transmission between neurons. Its types can include feedforward neural networks (FNN), convolutional neural networks (CNN), recurrent neural networks (RNN), etc. Since there are already relatively mature neural network models in this field, this embodiment will not elaborate on them further.

[0060] This embodiment uses slope monitoring feature data to select a slope monitoring training set that better matches the actual situation of the slope. This reduces the interference of irrelevant training data on model training, improves the accuracy of the slope monitoring neural network model, and thus improves the accuracy of the slope monitoring results.

[0061] This embodiment utilizes satellite remote sensing images for contour recognition, enabling the acquisition of precise effective slope areas within a large-scale satellite remote sensing imagery. Subsequently, feature recognition is performed on these effective slope areas to identify surface feature data. By employing physical field feature analysis methods and monitoring data, the correlation between physical fields and the abrupt changes within the physical fields themselves are assessed. This allows for the selection of more relevant and concise physical field feature data for slope monitoring, avoiding misjudgments caused by massive amounts of physical field feature data. Furthermore, by coupling physical field feature data with surface feature data, the slope monitoring feature data achieves comprehensive coverage from macroscopic contour recognition to microscopic physical field analysis, ensuring that the slope monitoring neural network model can output more accurate slope monitoring results, thus improving the accuracy of slope monitoring.

[0062] Example 2 Please refer to Figure 2 , Figure 2 A schematic diagram of a slope monitoring system based on UAV technology and multi-physics field coupling provided for an embodiment of the present invention includes: a multi-source heterogeneous data acquisition module 201, an image feature recognition module 202, a physical field feature recognition module 203, a multi-source heterogeneous data coupling module 204, and a slope monitoring module 205. The multi-source heterogeneous data acquisition module 201 is used to collect monitoring data of several physical fields of the slope based on preset UAV technology, and to acquire several satellite remote sensing images of the slope.

[0063] The image feature recognition module 202 is used to perform contour recognition on the satellite remote sensing image according to a preset feature recognition algorithm to obtain the effective area of ​​the slope in each satellite remote sensing image, and to perform feature recognition on the effective area of ​​the slope to obtain the surface feature data of the slope.

[0064] In this embodiment, the image feature recognition module 202 includes: an image feature recognition unit; the image feature recognition unit is used to acquire the color channel value of each pixel in each satellite remote sensing image, and convert each satellite remote sensing image into a satellite remote sensing grayscale image based on the color channel value; based on a preset sliding window and a preset convolution kernel, identify the first slope contour and the second slope contour of each satellite remote sensing grayscale image, and determine the effective slope area of ​​each satellite remote sensing grayscale image based on the first slope contour and the second slope contour; perform feature recognition on the effective slope area of ​​each satellite remote sensing grayscale image to obtain the surface feature data of the slope.

[0065] In this embodiment, the image feature recognition unit includes: a slope effective area acquisition subunit; the slope effective area acquisition subunit is used to slide a preset sliding window sequentially across each of the satellite remote sensing grayscale images based on a preset sliding step frequency; after each slide, the average grayscale value of all pixels within the sliding window and the grayscale value of the pixel corresponding to the center of the sliding window are calculated; if the grayscale value of the pixel corresponding to the center of the sliding window is greater than the average grayscale value of all pixels within the sliding window, the pixel corresponding to the center of the sliding window is marked as the first pixel, until the sliding window has completed its slide across each of the satellite remote sensing grayscale images, thus obtaining the first pixel set of each of the satellite remote sensing grayscale images; based on each of the images... The first set of pixels in each satellite remote sensing grayscale image is used to obtain the first slope contour of each image. A preset convolution kernel is then used to convolve each pixel in each image to obtain the response value of each pixel. If the response value of a pixel is greater than a preset response threshold, the pixel is marked as a second pixel, resulting in the second set of pixels in each image. Based on the second set of pixels, the second slope contour of each image is obtained. Based on the first and second slope contours, the effective slope area of ​​each image is determined.

[0066] In this embodiment, the image feature recognition unit further includes: a surface feature data acquisition subunit; the surface feature data acquisition subunit is used to construct a grayscale histogram of the effective slope area of ​​each satellite remote sensing grayscale image based on the grayscale values ​​of all pixels in the effective slope area of ​​each satellite remote sensing grayscale image, and calculate the total number of pixels and the sum of grayscale values ​​of the effective slope area of ​​each satellite remote sensing grayscale image; based on the grayscale histogram, determine several grayscale levels and several feature separation candidate thresholds of the effective slope area of ​​each satellite remote sensing grayscale image, wherein one grayscale level corresponds to one feature separation candidate threshold; initial For each satellite remote sensing grayscale image, the feature separation threshold, feature classification variance, cumulative pixel count, and cumulative grayscale value of the effective slope area are calculated. The feature separation candidate thresholds for the effective slope area of ​​each satellite remote sensing grayscale image are then iterated sequentially. In each iteration, based on the grayscale level corresponding to the currently iterated feature separation candidate threshold, the cumulative pixel count and the cumulative grayscale value are updated. Based on the updated cumulative grayscale value, the updated cumulative pixel count, the total number of pixels, and the cumulative grayscale value, the target proportion, target average grayscale value, non-target proportion, and non-target grayscale value of the currently iterated feature separation candidate threshold are calculated. The average grayscale value is calculated; based on the target ratio, target average grayscale value, non-target ratio, and non-target average grayscale value, the separation variance of the currently traversed feature separation candidate thresholds is calculated; the separation variance of the currently traversed feature separation candidate thresholds is compared with the current feature classification variance. If the separation variance of the currently traversed feature separation candidate thresholds is greater than the current feature classification variance, the feature separation threshold is updated based on the currently traversed feature separation candidate thresholds, and the feature classification variance is updated based on the separation variance of the currently traversed feature separation candidate thresholds, until all feature separation candidate thresholds have been traversed; the effective slope area is determined based on each of the satellite remote sensing grayscale images. The feature separation threshold obtained from the last update of the domain is used as the optimal feature separation threshold for the effective slope area of ​​each satellite remote sensing grayscale image. Based on the optimal feature separation threshold for the effective slope area of ​​each satellite remote sensing grayscale image, the pixels of the effective slope area are classified to obtain the protective engineering category pixels of the effective slope area. Based on the number of protective engineering category pixels, the protective engineering integrity rate of the effective slope area of ​​each satellite remote sensing grayscale image is calculated. Based on the data sampling time of each satellite remote sensing grayscale image, the protective engineering integrity rate is sorted to generate the surface feature data of the slope.

[0067] The physical field feature identification module 203 is used to obtain the correlation score between each pair of physical fields and the abrupt change score of each physical field according to the preset physical field feature analysis method and the monitoring data, and to filter the monitoring data of the physical fields based on the correlation score and the abrupt change score to obtain the physical field feature data of the slope.

[0068] In this embodiment, the physical field feature identification module 203 includes: a physical field feature analysis unit; the physical field feature analysis unit is used to calculate the average value of monitoring data for each physical field based on the monitoring data for each physical field; calculate the average difference of monitoring data for each physical field based on the average value of monitoring data and the monitoring data; calculate the covariance of monitoring data between two physical fields based on the average difference of monitoring data for each physical field, and calculate the standard deviation of monitoring data for each physical field; determine the correlation score between two physical fields based on the covariance of monitoring data between two physical fields and the standard deviation of monitoring data for each physical field; construct an absolute difference sequence of monitoring data for each physical field based on the monitoring data for each physical field, and calculate the variance of monitoring data for each physical field based on the absolute difference sequence of monitoring data; and determine the abrupt change score of each physical field based on the variance of monitoring data for each physical field.

[0069] In this embodiment, the physical field feature identification module 203 includes: a physical field feature data acquisition unit; the physical field feature data acquisition unit is used to calculate the mean of mutation scores based on the mutation scores of each physical field; based on the mean of mutation scores and the mutation scores of each physical field, the physical fields are filtered to obtain several mutated physical fields; an undirected weighted complete graph is constructed with each mutated physical field as a node and the correlation score between two mutated physical fields as edge weights; the optimal path length is determined based on the correlation score between two mutated physical fields; based on a preset state compression dynamic programming algorithm and the optimal path length, the undirected weighted complete graph is solved with the goal of maximizing the sum of correlation scores to obtain the optimal path; the mutated physical fields corresponding to each node in the optimal path are marked as slope monitoring physical fields, and the physical field feature data of the slope is obtained based on the monitoring data corresponding to the slope monitoring physical fields.

[0070] The multi-source heterogeneous data coupling module 204 is used to perform data coupling on the physical field feature data and surface feature data according to the preset multi-source heterogeneous data coupling method and in combination with the mutation score, so as to obtain slope monitoring feature data.

[0071] In this embodiment, the multi-source heterogeneous data coupling module 204 includes: a multi-source heterogeneous data coupling unit; the multi-source heterogeneous data coupling unit is used to determine the mutation weight of each slope monitoring physical field based on the mutation score of each slope monitoring physical field; normalize the monitoring data of each slope monitoring physical field to obtain normalized monitoring data of each slope monitoring physical field; extract data from the normalized monitoring data of each slope monitoring physical field based on the data sampling time of the normalized monitoring data of each slope monitoring physical field to generate an initial physical field monitoring vector of the slope at each data sampling time; weight the initial physical field monitoring vector based on the mutation weight of each slope monitoring physical field to obtain a first physical field monitoring vector of the slope at each data sampling time; concatenate the surface feature data with the first physical field monitoring vector based on the data sampling time of the surface feature data to obtain a monitoring feature vector of the slope at each data sampling time; and obtain slope monitoring feature data based on the monitoring feature vector of the slope at each data sampling time.

[0072] The slope monitoring module 205 is used to construct a slope monitoring neural network model based on the slope monitoring feature data, and input the slope monitoring feature data into the slope monitoring neural network model to obtain the slope monitoring results.

[0073] In this embodiment, the slope monitoring module 205 includes: a slope monitoring unit; the slope monitoring unit is used to filter a preset training dataset based on the slope monitoring feature data to generate a slope monitoring training set; to train a preset neural network model based on the slope monitoring training set and a preset slope label set to obtain a slope monitoring neural network model; and to input the slope monitoring feature data into the slope monitoring neural network model to obtain the slope monitoring result.

[0074] This embodiment uses contour recognition on satellite remote sensing images to obtain precise effective slope areas within a large range of images. Then, feature recognition is performed on these effective slope areas to identify surface feature data. By using physical field feature analysis methods and physical field monitoring data, the correlation between physical fields and the abrupt changes within the physical fields are determined. This allows for the selection of more relevant and concise physical field feature data for slope monitoring, avoiding misjudgments caused by massive amounts of physical field feature data. By coupling physical field feature data and surface feature data, the slope monitoring feature data achieves comprehensive coverage from macroscopic contour recognition to microscopic physical field analysis, ensuring that the slope monitoring neural network model can output more accurate slope monitoring results, thus improving the accuracy of slope monitoring.

[0075] In summary, this invention, through contour recognition of satellite remote sensing images, can obtain precise effective slope areas within a large range of satellite remote sensing images. Subsequently, feature recognition is performed on the effective slope areas to identify surface feature data of the slope. By using physical field feature analysis methods and physical field monitoring data, the correlation between physical fields and the abrupt changes within the physical fields themselves are determined, thereby filtering out more relevant and concise physical field feature data for slope monitoring, avoiding misjudgments caused by massive amounts of physical field feature data. By coupling physical field feature data and surface feature data, slope monitoring feature data achieves comprehensive coverage from macroscopic contour recognition to microscopic physical field analysis, ensuring that the slope monitoring neural network model can output more accurate slope monitoring results; thus improving the accuracy of slope monitoring.

[0076] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A slope monitoring method based on multiphysics coupling using unmanned aerial vehicle (UAV) technology, characterized in that, include: Based on the pre-set UAV technology, monitoring data of several physical fields of the slope are collected, and several satellite remote sensing images of the slope are obtained. The satellite remote sensing images are contoured according to a preset feature recognition algorithm to obtain the effective slope area of ​​each satellite remote sensing image, and the effective slope area is feature-recognized to obtain the surface feature data of the slope. Based on the preset physical field feature analysis method and the monitoring data, the correlation score between each pair of physical fields and the abrupt change score of each physical field are obtained. Based on the correlation score and the abrupt change score, the monitoring data of the physical fields are filtered to obtain the physical field feature data of the slope. Based on the preset multi-source heterogeneous data coupling method, and combined with the mutation score, the physical field feature data and surface feature data are coupled to obtain slope monitoring feature data. A slope monitoring neural network model is constructed based on the slope monitoring feature data, and the slope monitoring feature data is input into the slope monitoring neural network model to obtain the slope monitoring results.

2. The slope monitoring method based on multiphysics coupling of UAV technology as described in claim 1, characterized in that, The process involves performing contour recognition on the satellite remote sensing images according to a preset feature recognition algorithm to obtain the effective slope area for each satellite remote sensing image, and then performing feature recognition on the effective slope area to obtain surface feature data of the slope, including: Obtain the color channel value of each pixel in each of the satellite remote sensing images, and convert each of the satellite remote sensing images into a satellite remote sensing grayscale image based on the color channel value; Based on a preset sliding window and a preset convolution kernel, the first slope contour and the second slope contour of each satellite remote sensing grayscale image are identified, and the effective slope area of ​​each satellite remote sensing grayscale image is determined based on the first slope contour and the second slope contour. Feature identification is performed on the effective area of ​​the slope in each of the aforementioned satellite remote sensing grayscale images to obtain the surface feature data of the slope.

3. The slope monitoring method based on multiphysics coupling of unmanned aerial vehicle (UAV) technology as described in claim 2, characterized in that, The process of identifying the first and second slope contours of each satellite remote sensing grayscale image based on a preset sliding window and a preset convolution kernel, and determining the effective slope region of each satellite remote sensing grayscale image based on the first and second slope contours, includes: Based on a preset sliding step frequency, the preset sliding window is sequentially slid across each of the aforementioned satellite remote sensing grayscale images; After each slide, the average gray value of all pixels in the sliding window and the gray value of the pixel corresponding to the center of the sliding window are calculated. If the gray value of the pixel corresponding to the center of the sliding window is greater than the average gray value of all pixels in the sliding window, the pixel corresponding to the center of the sliding window is marked as the first pixel. This process continues until the sliding window is completed on each satellite remote sensing grayscale image, resulting in the first set of pixels for each satellite remote sensing grayscale image. Based on the first set of pixels in each of the satellite remote sensing grayscale images, the first slope profile of each of the satellite remote sensing grayscale images is obtained; A preset convolution kernel is convolved with each pixel of each of the satellite remote sensing grayscale images to obtain the response value of each pixel in each of the satellite remote sensing grayscale images; If the response value of the pixel is greater than the preset response threshold, the pixel is marked as the second pixel, thus obtaining the second pixel set for each of the satellite remote sensing grayscale images; Based on the second pixel set of each of the satellite remote sensing grayscale images, the second slope profile of each of the satellite remote sensing grayscale images is obtained; Based on the first and second slope contours of each of the satellite remote sensing grayscale images, the effective slope area of ​​each of the satellite remote sensing grayscale images is determined.

4. A slope monitoring method based on multiphysics coupling using unmanned aerial vehicle (UAV) technology as described in claim 2 or 3, characterized in that, The step of identifying the effective area of ​​the slope in each of the satellite remote sensing grayscale images to obtain the surface feature data of the slope includes: Based on the gray values ​​of all pixels in the effective area of ​​the slope in each of the satellite remote sensing grayscale images, a grayscale histogram of the effective area of ​​the slope in each of the satellite remote sensing grayscale images is constructed, and the total number of pixels and the sum of gray values ​​in the effective area of ​​the slope in each of the satellite remote sensing grayscale images are calculated. Based on the grayscale histogram, several grayscale levels and several feature separation candidate thresholds are determined for the effective area of ​​the slope in each satellite remote sensing grayscale image, wherein one grayscale level corresponds to one feature separation candidate threshold. Initialize the feature separation threshold, feature classification variance, cumulative pixel count, and cumulative grayscale value of the effective area of ​​the slope in each of the aforementioned satellite remote sensing grayscale images; The feature separation candidate thresholds for the effective area of ​​the slope in each of the aforementioned satellite remote sensing grayscale images are sequentially traversed; In each traversal, based on the gray level corresponding to the feature separation candidate threshold of the current traversal, the cumulative number of pixels and the total cumulative gray value are updated, and based on the updated total cumulative gray value, the updated cumulative number of pixels, the total number of pixels and the total gray value, the target proportion, target average gray value, non-target proportion and non-target average gray value of the feature separation candidate threshold of the current traversal are calculated. The separation variance of the current feature separation candidate threshold is calculated based on the target ratio, target average gray value, non-target ratio, and non-target average gray value. Compare the separation variance of the currently traversed feature separation candidate thresholds with the current feature classification variance. If the separation variance of the currently traversed feature separation candidate thresholds is greater than the current feature classification variance, then update the feature separation threshold based on the currently traversed feature separation candidate thresholds, and update the feature classification variance based on the separation variance of the currently traversed feature separation candidate thresholds, until all feature separation candidate thresholds have been traversed. The feature separation threshold obtained from the last update of the effective slope area in each of the satellite remote sensing grayscale images is used as the optimal feature separation threshold for the effective slope area in each of the satellite remote sensing grayscale images. Based on the optimal feature separation threshold of the effective slope area in each satellite remote sensing grayscale image, the pixels of the effective slope area are classified to obtain the protective engineering type pixels of the effective slope area, and the protective engineering integrity rate of the effective slope area in each satellite remote sensing grayscale image is calculated based on the number of the protective engineering type pixels. Based on the data sampling time of each satellite remote sensing grayscale image, the integrity rate of the protection project is sorted to generate the surface feature data of the slope.

5. The slope monitoring method based on multiphysics coupling of unmanned aerial vehicle (UAV) technology as described in claim 1, characterized in that, The step of obtaining the correlation score between each pair of physical fields and the abrupt change score of each physical field based on the preset physical field characteristic analysis method and the monitoring data includes: Based on the monitoring data for each of the physical fields, calculate the average value of the monitoring data for each of the physical fields; Based on the average value of the monitoring data and the monitoring data, calculate the average difference of the monitoring data for each physical field; Based on the average difference of the monitoring data for each of the physical fields, the covariance of the monitoring data between each pair of physical fields is calculated, and the standard deviation of the monitoring data for each of the physical fields is calculated. Based on the covariance of the monitoring data between each pair of physical fields and the standard deviation of the monitoring data for each physical field, a correlation score is determined between each pair of physical fields. Based on the monitoring data of each physical field, an absolute difference sequence of monitoring data for each physical field is constructed, and the variance of the monitoring data for each physical field is calculated based on the absolute difference sequence of monitoring data. Based on the variance of the monitoring data for each physical field, a mutation score for each physical field is determined.

6. A slope monitoring method based on multiphysics coupling using unmanned aerial vehicle (UAV) technology as described in claim 1 or 5, characterized in that, The process of filtering the monitoring data of the physical field based on the correlation score and the abrupt change score to obtain the physical field characteristic data of the slope includes: The mean aberration score is calculated based on the aberration score of each of the physical fields; Based on the mean of the mutation score and the mutation score of each physical field, the physical fields are screened to obtain a number of mutated physical fields; Using each of the aforementioned aberrant physical fields as nodes and the correlation score between any two of the aforementioned aberrant physical fields as edge weights, an undirected weighted complete graph is constructed; The optimal path length is determined based on the correlation score between each of the two described abrupt physical fields; Based on the preset state compression dynamic programming algorithm and the optimal path length, the undirected weighted complete graph is solved with the goal of maximizing the sum of relevance scores to obtain the optimal path; The abrupt change physical field corresponding to each node in the optimal path is marked as the slope monitoring physical field, and the physical field characteristic data of the slope is obtained based on the monitoring data corresponding to the slope monitoring physical field.

7. The slope monitoring method based on multiphysics coupling of unmanned aerial vehicle (UAV) technology as described in claim 6, characterized in that, The process of coupling the physical field feature data and surface feature data according to a preset multi-source heterogeneous data coupling method, combined with the abrupt change score, to obtain slope monitoring feature data includes: Based on the abrupt change score of each of the slope monitoring physical fields, the abrupt change weight of each of the slope monitoring physical fields is determined; The monitoring data of each of the slope monitoring physical fields are normalized to obtain normalized monitoring data of each of the slope monitoring physical fields; Based on the data sampling time of the normalized monitoring data of each slope monitoring physical field, data extraction is performed on the normalized monitoring data of each slope monitoring physical field to generate the initial physical field monitoring vector of the slope at each data sampling time. Based on the abrupt change weight of each of the slope monitoring physical fields, the initial physical field monitoring vector is weighted to obtain the first physical field monitoring vector of the slope at each data sampling time. Based on the data sampling time of the surface feature data, the surface feature data is concatenated with the first physical field monitoring vector to obtain the monitoring feature vector of the slope at each data sampling time; Based on the monitoring feature vector of the slope at each data sampling time, slope monitoring feature data are obtained.

8. The slope monitoring method based on multiphysics coupling of unmanned aerial vehicle (UAV) technology as described in claim 1, characterized in that, The process involves constructing a slope monitoring neural network model based on the slope monitoring feature data, and inputting the slope monitoring feature data into the slope monitoring neural network model to obtain slope monitoring results, including: Based on the slope monitoring feature data, a preset training dataset is selected to generate a slope monitoring training set; Based on the slope monitoring training set and the preset slope label set, the preset neural network model is trained to obtain the slope monitoring neural network model; The slope monitoring feature data is input into the slope monitoring neural network model to obtain the slope monitoring results.

9. A slope monitoring system based on multiphysics coupling using unmanned aerial vehicle (UAV) technology, characterized in that, include: The system includes a multi-source heterogeneous data acquisition module, an image feature recognition module, a physical field feature recognition module, a multi-source heterogeneous data coupling module, and a slope monitoring module. The multi-source heterogeneous data acquisition module is used to collect monitoring data of several physical fields of the slope based on preset UAV technology, and to acquire several satellite remote sensing images of the slope. The image feature recognition module is used to perform contour recognition on the satellite remote sensing image according to a preset feature recognition algorithm to obtain the effective area of ​​the slope in each satellite remote sensing image, and to perform feature recognition on the effective area of ​​the slope to obtain the surface feature data of the slope. The physical field feature identification module is used to obtain the correlation score between each pair of physical fields and the abrupt change score of each physical field according to the preset physical field feature analysis method and the monitoring data, and to filter the monitoring data of the physical fields based on the correlation score and the abrupt change score to obtain the physical field feature data of the slope. The multi-source heterogeneous data coupling module is used to perform data coupling on the physical field feature data and surface feature data according to the preset multi-source heterogeneous data coupling method and in combination with the mutation score, so as to obtain slope monitoring feature data. The slope monitoring module is used to construct a slope monitoring neural network model based on the slope monitoring feature data, and input the slope monitoring feature data into the slope monitoring neural network model to obtain the slope monitoring results.

10. A slope monitoring system based on multiphysics coupling of unmanned aerial vehicle (UAV) technology as described in claim 9, characterized in that, The image feature recognition module includes: an image feature recognition unit; The image feature recognition unit is used to obtain the color channel value of each pixel in each satellite remote sensing image, and convert each satellite remote sensing image into a satellite remote sensing grayscale image based on the color channel value; Based on a preset sliding window and a preset convolution kernel, the first slope contour and the second slope contour of each satellite remote sensing grayscale image are identified, and the effective slope area of ​​each satellite remote sensing grayscale image is determined based on the first slope contour and the second slope contour. Feature identification is performed on the effective area of ​​the slope in each of the aforementioned satellite remote sensing grayscale images to obtain the surface feature data of the slope.

Images