Roadbed slope risk monitoring and early warning system based on artificial intelligence

Abstract

The invention discloses a roadbed slope risk monitoring and early warning system based on artificial intelligence, and relates to the technical field of risk monitoring. The method comprises the following steps: acquiring three-dimensional topographic data of a roadbed slope, constructing a three-dimensional topographic model, acquiring a ground surface vegetation image and ground surface vegetation parameters of each monitoring sub-region, and constructing a first evaluation model according to the three-dimensional topographic data, the ground surface vegetation parameters and the monitoring data of different monitoring sub-regions in the same monitoring scene, constructing a second evaluation model according to the three-dimensional topographic data, the earth surface vegetation parameters and the historical monitoring data in different historical accident sets, judging whether a risk area exists or not by using the second evaluation model and generating early warning information, and obtaining recommended vegetation parameters by using the first evaluation model and generating a vegetation optimization scheme; a targeted vegetation optimization scheme can be generated for each monitoring sub-region, and whether a safety accident occurs in a corresponding roadbed slope can be predicted.

Description

Technical Field

[0001] The present invention relates to the technical field of risk monitoring, and specifically to a subgrade slope risk monitoring and early warning system based on artificial intelligence. Background Art

[0002] It is a mature existing technology to use drones equipped with different devices to inspect subgrade slopes and judge whether there are safety accidents. By collecting point cloud data and surface images of subgrade slopes, and using big data analysis technology combined with monitoring data, the current state of subgrade slopes can be obtained in real time. When a safety accident occurs, relevant personnel are notified for maintenance, which can significantly reduce the cost of manual inspection; Most of the existing monitoring methods rely on single sensors, lack the fusion analysis of multi-modal data, and cannot ensure the accuracy of risk monitoring. Moreover, the influence of surface vegetation on the stability of subgrade slopes is often ignored in the existing technology. Too much or too little surface vegetation will indirectly affect the stability of subgrade slopes. How to adjust the surface vegetation of subgrade slopes has become a problem to be solved. In view of the deficiencies of the existing technology, the present invention provides a subgrade slope risk monitoring and early warning system based on artificial intelligence. Summary of the Invention

[0003] The purpose of the present invention is to provide a subgrade slope risk monitoring and early warning system based on artificial intelligence.

[0004] The purpose of the present invention can be achieved through the following technical solutions: A subgrade slope risk monitoring and early warning system based on artificial intelligence includes the following modules: A data acquisition module, which is used to obtain the three-dimensional terrain data of the subgrade slope, construct a three-dimensional terrain model, divide the subgrade slope into several monitoring sub-areas, obtain the surface vegetation images of each monitoring sub-area, and obtain the surface vegetation parameters of each monitoring sub-area according to the surface vegetation images; A data monitoring module, which is used to set different monitoring terminals in each monitoring sub-area respectively, obtain monitoring data respectively, and construct a first evaluation model of the subgrade slope according to the three-dimensional terrain data, surface vegetation parameters and monitoring data of different monitoring sub-areas under the same monitoring scenario; A data analysis module, which is used to obtain the historical accident set of the subgrade slope, and construct a second evaluation model of the subgrade slope according to the three-dimensional terrain data, surface vegetation parameters and historical monitoring data in different historical accident sets; A data evaluation module, which is used to input the three-dimensional terrain data, surface vegetation parameters and real-time monitoring data of each monitoring sub-area into the second evaluation model to judge whether there are risk areas and generate early warning information; A data optimization module, which is used to utilize the first evaluation model to combine the three-dimensional terrain data and expected monitoring data of each risk area to obtain recommended vegetation parameters and generate a vegetation optimization plan.

[0005] Further, the process of obtaining the three-dimensional terrain data of the subgrade slope and constructing a three-dimensional terrain model, and dividing the subgrade slope into several monitoring sub-areas includes: Mount a high-precision lidar on the unmanned aerial vehicle to collect the three-dimensional terrain data of the subgrade slope. The three-dimensional terrain data includes three-dimensional point cloud data, slope and aspect, slope length and height, geographical location, texture information, and color information; Use three-dimensional modeling technology to construct a three-dimensional terrain model of the subgrade slope according to the three-dimensional terrain data. In the three-dimensional terrain model, divide the subgrade slope to obtain several monitoring sub-areas.

[0006] Further, the process of obtaining the surface vegetation images of each monitoring sub-area and obtaining the surface vegetation parameters of each monitoring sub-area according to the surface vegetation images includes: Mount a multi-spectral camera on the unmanned aerial vehicle to collect the surface vegetation images of each monitoring sub-area, and obtain the surface vegetation parameters according to the surface vegetation image and three-dimensional point cloud data of a single monitoring sub-area, including vegetation coverage rate S a 、leaf area index S b 、vegetation canopy volume density S c ;

[0007] n is the number of vegetation areas divided in the surface vegetation image of the single monitoring sub-area by using the semantic segmentation algorithm, i is the number of the divided vegetation areas, i = 1, 2,..., n; is the vegetation existence coefficient of the i-th vegetation area, represents vegetation coverage, represents no vegetation coverage; A p,i is the pixel area of the i-th vegetation area, which is converted into the actual area by combining the ground resolution, A t is the total area of the single monitoring sub-area;

[0008] is a preset fixed parameter, are the near-infrared band reflectance and infrared band reflectance in the surface vegetation image of the single monitoring sub-area respectively, is the solar zenith angle;

[0009] Let \(m\) be the number of vegetation points in the vegetation area of the single monitoring sub-region in the three-dimensional point cloud data, and \(j\) be the number of each vegetation point, where \(j = 1, 2, \ldots, m\). Z m,j is the elevation value of the \(j\)-th vegetation point, and \(Z\) g,j is the ground elevation value of the vertical point of the \(j\)-th vegetation point on the ground. is the elevation layer thickness.

[0010] Furthermore, the process of respectively setting different monitoring terminals in each monitoring sub-region and respectively obtaining monitoring data includes: The monitoring terminal includes a rain gauge, a light intensity meter, a fiber Bragg grating displacement meter, a pore water pressure gauge, an earth pressure cell, and a temperature and humidity sensor, and the monitoring data includes the rainfall, light intensity, displacement value, pore water pressure, stress value, temperature and humidity of each monitoring sub-region.

[0011] Furthermore, the process of constructing the first evaluation model of the subgrade slope according to the three-dimensional terrain data, surface vegetation parameters and their monitoring data of different monitoring sub-regions under the same monitoring scenario includes: The same monitoring scenario means that the rainfall and light intensity of each monitoring sub-region are the same. When the rainfall and light intensity of different monitoring sub-regions are the same, they are regarded as the same monitoring scenario. Generate a first evaluation set according to the rainfall, light intensity, temperature and humidity, pore water pressure, three-dimensional terrain data, surface vegetation parameters, displacement value and stress value of different monitoring sub-regions under the same monitoring scenario, and divide the first evaluation set into a first training set and a first test set. Construct a first convolutional neural network, use the rainfall, light intensity, temperature and humidity, pore water pressure, three-dimensional terrain data, and surface vegetation parameters in the first training set as the input data of the first convolutional neural network, and use the displacement value and stress value in the first training set as the output data of the first convolutional neural network. Train the first convolutional neural network to obtain an initial first convolutional neural network, use the first test set to verify the model of the initial first convolutional neural network, and output the initial first convolutional neural network with an error less than or equal to the preset first test error threshold as the first evaluation model.

[0012] Furthermore, the process of obtaining the historical accident set of the subgrade slope and constructing the second evaluation model of the subgrade slope according to the three-dimensional terrain data, surface vegetation parameters and historical monitoring data in different historical accident sets includes: The historical accident set refers to the three-dimensional terrain data, surface vegetation parameters and historical monitoring data corresponding to when a safety accident has occurred on the subgrade slope, and the historical monitoring data refers to the monitoring data obtained when a safety accident occurs. Generate a second evaluation set based on the three-dimensional terrain data, surface vegetation parameters, and historical monitoring data within different historical accident sets, and divide the second evaluation set into a second training set and a second test set; Construct a second convolutional neural network, use the three-dimensional terrain data, surface vegetation parameters, and historical monitoring data in the second training set as the input data of the second convolutional neural network, and use whether a safety accident will occur on the roadbed slope as the output data of the second convolutional neural network; Train the second convolutional neural network to obtain an initial second convolutional neural network, use the second test set to verify the model of the initial second convolutional neural network, and output the initial second convolutional neural network with a second test error threshold less than or equal to the preset value as the second evaluation model.

[0013] Furthermore, the process of inputting the three-dimensional terrain data, surface vegetation parameters, and real-time monitoring data of each monitoring sub-region into the second evaluation model to determine whether there are risk regions and generating early warning information includes: Use the current monitoring data of a single monitoring sub-region as the real-time monitoring data, combine the three-dimensional terrain data and surface vegetation parameters of this single monitoring sub-region, and input them into the second evaluation model to use the second evaluation model to determine whether a safety accident will occur in this single monitoring sub-region; If so, mark it as a risk region, generate corresponding early warning information, and feedback the early warning information to relevant personnel. If not, mark it as a non-risk region.

[0014] Furthermore, the process of using the first evaluation model to combine the three-dimensional terrain data and expected monitoring data of each risk region to obtain recommended vegetation parameters and generate a vegetation optimization plan includes: Use the mean value of the monitoring data corresponding to each non-risk region in the same monitoring scenario as the single risk region as the expected monitoring data of this single risk region; Use the rainfall, light intensity, temperature and humidity, and pore water pressure in the expected monitoring data of this single risk region, together with its three-dimensional terrain data and surface vegetation parameters, as the input data of the first evaluation model; Continuously adjust the input surface vegetation parameters until the output data of the first evaluation model is equal to the displacement value and stress value in the expected monitoring data of this single risk region, and use the surface vegetation parameters at this time as the recommended vegetation parameters; Optimize the surface vegetation of this single risk region, including replacing the vegetation type, changing the vegetation spacing, and pruning the vegetation branches and leaves. Use the optimization plan that meets the recommended vegetation parameters as the vegetation optimization plan and feedback it to relevant personnel.

[0015] Compared with the prior art, the beneficial effects of the present invention are: By dividing the roadbed slope into several monitoring sub - regions and obtaining the surface vegetation parameters of each monitoring sub - region, the present invention can incorporate the influence of surface vegetation on the stability of the roadbed slope into the monitoring scope. By setting different monitoring terminals and obtaining different monitoring data, and then constructing the first evaluation model of the roadbed slope, it can obtain the influence degree of different surface vegetation parameters on the monitoring data under the same monitoring scenario, and combine the expected monitoring data to obtain the recommended vegetation parameters for each monitoring sub - region, and can generate a targeted vegetation optimization plan for each monitoring sub - region; By obtaining the historical accident set of the roadbed slope and constructing the second evaluation model of the roadbed slope according to the three - dimensional terrain data, surface vegetation parameters, and historical monitoring data in different historical accident sets, and inputting the current data into the second evaluation model, it can predict whether the corresponding roadbed slope will have a safety accident and generate corresponding early warning information for feedback, which can timely remind relevant personnel to carry out maintenance on it. Brief Description of the Drawings

[0016] Figure 1 It is a schematic diagram of the present invention. Detailed Embodiment

[0017] As Figure 1 shown, an artificial - intelligence - based roadbed slope risk monitoring and early warning system includes the following modules: A data acquisition module, which is used to obtain the three - dimensional terrain data of the roadbed slope, construct a three - dimensional terrain model, divide the roadbed slope into several monitoring sub - regions, obtain the surface vegetation images of each monitoring sub - region, and obtain the surface vegetation parameters of each monitoring sub - region according to the surface vegetation images; A data monitoring module, which is used to set different monitoring terminals in each monitoring sub - region respectively, obtain monitoring data respectively, and construct the first evaluation model of the roadbed slope according to the three - dimensional terrain data, surface vegetation parameters, and their monitoring data of different monitoring sub - regions under the same monitoring scenario; A data analysis module, which is used to obtain the historical accident set of the roadbed slope and construct the second evaluation model of the roadbed slope according to the three - dimensional terrain data, surface vegetation parameters, and historical monitoring data in different historical accident sets; A data evaluation module, which is used to input the three - dimensional terrain data, surface vegetation parameters, and real - time monitoring data of each monitoring sub - region into the second evaluation model to judge whether there are risk areas and generate early warning information; A data optimization module, which is used to use the first evaluation model to combine the three - dimensional terrain data and expected monitoring data of each risk area to obtain recommended vegetation parameters and generate a vegetation optimization plan.

[0018] It should be further noted that in the specific implementation process, the process of obtaining the three-dimensional terrain data of the subgrade slope and constructing a three-dimensional terrain model, and dividing the subgrade slope into several monitoring sub-areas includes: Mount a high-precision lidar on the unmanned aerial vehicle (UAV), and use the high-precision lidar to collect the three-dimensional terrain data of the subgrade slope. The three-dimensional terrain data includes three-dimensional point cloud data, slope and aspect, slope length and height, geographical location, texture information, color information, etc.; Use three-dimensional modeling technology to construct a three-dimensional terrain model of the subgrade slope based on the collected three-dimensional terrain data. In the constructed three-dimensional terrain model, divide the subgrade slope to obtain several sub-areas with approximately equal areas, and mark the obtained sub-areas as monitoring sub-areas.

[0019] It should be further noted that in the specific implementation process, the process of obtaining the surface vegetation images of each monitoring sub-area and obtaining the surface vegetation parameters of each monitoring sub-area based on the surface vegetation images includes: Mount a multispectral camera on the UAV, and use the multispectral camera to collect the surface vegetation images of each monitoring sub-area. The surface vegetation images can reflect vegetation type, vegetation coverage, vegetation health status, vegetation height and density; Taking a single monitoring sub-area as an example, obtain the surface vegetation parameters based on the surface vegetation image and its three-dimensional point cloud data of the single monitoring sub-area. The surface vegetation parameters include vegetation coverage rate, leaf area index, and vegetation canopy volume density, which are respectively denoted as S a 、S b 、S c ;

[0020] where n is the number of vegetation areas divided in the surface vegetation image of the single monitoring sub-area using the semantic segmentation algorithm, i is the number of the divided vegetation areas, and i = 1, 2,..., n; is the vegetation existence coefficient of the i-th vegetation area, represents vegetation coverage, represents no vegetation coverage; A p,i is the pixel area of the i-th vegetation area, which is converted into the actual area in combination with the ground resolution, and A t is the total area of the single monitoring sub-area;

[0021] where, is a preset fixed parameter, They are the near-infrared band reflectance and the infrared band reflectance in the surface vegetation image of the single monitoring sub-region respectively, is the solar zenith angle when the UAV conducts image acquisition;

[0022] where m is the number of vegetation points in the vegetation area of the single monitoring sub-region in the three-dimensional point cloud data, j is the number of each vegetation point, and j = 1, 2,..., m; Z m,j is the elevation value of the j-th vegetation point, and Z g,j is the ground elevation value of the vertical point of the j-th vegetation point on the ground, is the elevation layer thickness, with a value of 0.1 meter.

[0023] It should be further noted that in the specific implementation process, the process of setting different monitoring terminals in each monitoring sub-region respectively and obtaining monitoring data respectively includes: The monitoring terminal includes a rain gauge, a illuminance meter, a fiber Bragg grating displacement meter, a pore water pressure gauge, an earth pressure cell, and a temperature and humidity sensor. The monitoring data includes the rainfall, illuminance, displacement value, pore water pressure, stress value, temperature and humidity of each monitoring sub-region; The rainfall of the corresponding monitoring sub-region is obtained in real time through the rain gauge, the illuminance of the corresponding monitoring sub-region is obtained in real time through the illuminance meter, and the displacement value of the corresponding monitoring sub-region is obtained in real time through the fiber Bragg grating displacement meter; The pore water pressure of the corresponding monitoring sub-region is obtained in real time through the pore water pressure gauge, the stress value of the corresponding monitoring sub-region is obtained in real time through the earth pressure cell, and the temperature and humidity of the corresponding monitoring sub-region are obtained in real time through the temperature and humidity sensor.

[0024] It should be further noted that in the specific implementation process, the process of constructing the first evaluation model of the subgrade slope according to the three-dimensional terrain data, surface vegetation parameters and their monitoring data of different monitoring sub-regions under the same monitoring scenario includes: The same monitoring scenario means that the weather conditions of each monitoring sub-region are the same, which is reflected in the rainfall and illuminance. When the rainfall and illuminance of different monitoring sub-regions are the same, they are regarded as the same monitoring scenario; Generate a first evaluation set according to the rainfall, illuminance, temperature and humidity, pore water pressure, three-dimensional terrain data, surface vegetation parameters and their corresponding displacement values and stress values of different monitoring sub-regions under the same monitoring scenario, and divide the first evaluation set into a first training set and a first test set; Construct a first convolutional neural network, take different rainfall amounts, light intensities, temperature and humidity, pore water pressures, three-dimensional terrain data, and surface vegetation parameters in the first training set as the input data of the first convolutional neural network, and take the corresponding displacement values and stress values in the first training set as the output data of the first convolutional neural network; Train the first convolutional neural network to obtain an initial first convolutional neural network, use the first test set to verify the model of the initial first convolutional neural network, and output the initial first convolutional neural network with an error less than or equal to the preset first test error threshold as the corresponding first evaluation model.

[0025] It should be further noted that in the specific implementation process, the process of obtaining the historical accident set of the subgrade slope and constructing the second evaluation model of the subgrade slope according to the three-dimensional terrain data, surface vegetation parameters, and historical monitoring data in different historical accident sets includes: The historical accident set refers to the three-dimensional terrain data, surface vegetation parameters, and historical monitoring data corresponding to when a safety accident has occurred on the subgrade slope. The safety accidents include slope collapse, slope landslide, slope debris flow, slope surface erosion, etc. The historical monitoring data refers to the monitoring data obtained when a safety accident occurs; Generate a second evaluation set according to the three-dimensional terrain data, surface vegetation parameters, and historical monitoring data in different historical accident sets, and divide the second evaluation set into a second training set and a second test set; Construct a second convolutional neural network, take different three-dimensional terrain data, surface vegetation parameters, and historical monitoring data in the second training set as the input data of the second convolutional neural network, and take whether a safety accident will occur on the subgrade slope as the output data of the second convolutional neural network; Train the second convolutional neural network to obtain an initial second convolutional neural network, use the second test set to verify the model of the initial second convolutional neural network, and output the initial second convolutional neural network with an error less than or equal to the preset second test error threshold as the corresponding second evaluation model.

[0026] It should be further noted that in the specific implementation process, the process of inputting the three-dimensional terrain data, surface vegetation parameters, and real-time monitoring data of each monitoring sub-region into the second evaluation model to determine whether there is a risk area and generating a warning message includes: Taking a single monitoring sub-region as an example, take the current monitoring data of the single monitoring sub-region as the real-time monitoring data, and input it into the second evaluation model in combination with the three-dimensional terrain data and surface vegetation parameters of the single monitoring sub-region; Use the second evaluation model to determine whether a safety accident will occur in the single monitoring sub-region based on the input data. If so, mark it as a risk area, generate corresponding warning information, and feedback the generated warning information to relevant personnel to prompt them to perform maintenance on it in a timely manner. If not, mark it as a non-risk area and do not perform any other operations on it.

[0027] It should be further noted that in the specific implementation process, the process of using the first evaluation model to combine the three-dimensional terrain data and expected monitoring data of each risk area to obtain recommended vegetation parameters and generate a vegetation optimization plan includes: In the embodiment of the present invention, the average value of the monitoring data corresponding to each non-risk area in the same monitoring scenario as the single risk area is used as the expected monitoring data of the single risk area. The rainfall, light intensity, temperature and humidity, pore water pressure in the expected monitoring data of the single risk area and their corresponding three-dimensional terrain data and surface vegetation parameters are used as the input data of the first evaluation model; Continuously adjust the input surface vegetation parameters until the output data of the first evaluation model is equal to the displacement value and stress value in the expected monitoring data of the single risk area, and use the surface vegetation parameters at this time as the recommended vegetation parameters of the single risk area; Perform vegetation optimization on the surface vegetation of the single risk area, including replacing the vegetation type, changing the vegetation spacing, pruning the vegetation branches and leaves, etc. Use the optimization plan that meets the recommended vegetation parameters as the corresponding vegetation optimization plan, and feedback it to relevant personnel to prompt them how to optimize it.

[0028] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A roadbed slope risk monitoring and early warning system based on artificial intelligence, characterized in that: Includes the following modules: The data acquisition module is used to obtain the three-dimensional terrain data of the roadbed slope and construct a three-dimensional terrain model, divide the roadbed slope into several monitoring sub-areas, obtain the surface vegetation image of each monitoring sub-area, and obtain the surface vegetation parameters of each monitoring sub-area based on the surface vegetation image; The data monitoring module is used to respectively set different monitoring terminals in each monitoring sub-area and obtain monitoring data respectively, and to build a first evaluation model of the roadbed slope according to the three-dimensional terrain data, surface vegetation parameters and monitoring data of different monitoring sub-areas in the same monitoring scene; A data analysis module is used to obtain a historical accident set of the roadbed slope, and to construct a second assessment model of the roadbed slope based on the three-dimensional terrain data, surface vegetation parameters, and historical monitoring data in different historical accident sets; A data evaluation module is used to input the three-dimensional terrain data, surface vegetation parameters, and real-time monitoring data of each monitoring sub-area into the second evaluation model to determine whether there is a risk area and generate early warning information; The data optimization module is used to use the first assessment model to combine the three-dimensional terrain data of each risk area and the expected monitoring data to obtain recommended vegetation parameters and generate a vegetation optimization plan.

2. The artificial intelligence-based roadbed slope risk monitoring and early warning system according to claim 1 is characterized in that: The process of building a 3D terrain model and dividing the monitoring sub-areas includes: A high-precision laser radar is mounted on the drone to collect three-dimensional terrain data of the roadbed slope, wherein the three-dimensional terrain data includes three-dimensional point cloud data, slope and slope direction, slope length and slope height, geographic location, texture information, and color information; A three-dimensional terrain model of the roadbed slope is constructed based on various three-dimensional terrain data using three-dimensional modeling technology. In the three-dimensional terrain model, the roadbed slope is divided into several monitoring sub-areas.

3. The artificial intelligence-based roadbed slope risk monitoring and early warning system according to claim 2 is characterized in that: The process of obtaining surface vegetation images and their surface vegetation parameters includes: The multispectral camera is mounted on the UAV to collect the surface vegetation images of each monitoring sub-area. The surface vegetation parameters, including the vegetation coverage rate S, are obtained based on the surface vegetation images and three-dimensional point cloud data of a single monitoring sub-area. a , leaf area index S b , vegetation canopy volume density S c ; n is the number of vegetation areas divided in the surface vegetation image of the single monitoring sub-area using the semantic segmentation algorithm, i is the number of the divided vegetation areas, i=1, 2, ..., n; is the vegetation existence coefficient of the i-th vegetation area, represents vegetation cover, represents no vegetation cover; A p,i is the pixel area of ​​the ith vegetation area, which is converted into the actual area based on the ground resolution. t is the total area of ​​the single monitoring sub-area; are preset fixed parameters. are the near-infrared band reflectance and infrared band reflectance in the surface vegetation image of the single monitoring sub-area, is the solar zenith angle; m is the number of vegetation points in the vegetation area of ​​the single monitoring sub-area in the three-dimensional point cloud data, j is the number of each vegetation point, j=1, 2, ..., m; Z m,j is the elevation value of the jth vegetation point, Z g,j is the ground elevation of the vertical point of the jth vegetation point on the ground, is the elevation layer thickness.

4. The artificial intelligence-based roadbed slope risk monitoring and early warning system according to claim 3 is characterized in that: The process of setting up the monitoring terminal and obtaining monitoring data includes: The monitoring end includes a rain gauge, a light meter, a fiber Bragg grating displacement meter, a pore water pressure gauge, an earth pressure box, and a temperature and humidity sensor. The monitoring data includes rainfall, light intensity, displacement value, pore water pressure, stress value, temperature and humidity in each monitoring sub-area.

5. The artificial intelligence-based roadbed slope risk monitoring and early warning system according to claim 4 is characterized in that: The process of constructing the first assessment model of the embankment slope includes: The same monitoring scene refers to the same rainfall and light intensity in each monitoring sub-area. When the rainfall and light intensity in different monitoring sub-areas are the same, they are regarded as the same monitoring scene; Generate a first evaluation set according to rainfall, light intensity, temperature and humidity, pore water pressure, three-dimensional terrain data, surface vegetation parameters and displacement values ​​and stress values ​​of different monitoring sub-areas under the same monitoring scenario, and divide the first evaluation set into a first training set and a first test set; Constructing a first convolutional neural network, using the rainfall, light intensity, temperature and humidity, pore water pressure, three-dimensional terrain data, and surface vegetation parameters in the first training set as input data of the first convolutional neural network, and using the displacement value and stress value in the first training set as output data of the first convolutional neural network; The first convolutional neural network is trained to obtain an initial first convolutional neural network, the initial first convolutional neural network is model verified using a first test set, and the initial first convolutional neural network that is less than or equal to a preset first test error threshold is output as a first evaluation model.

6. The artificial intelligence-based roadbed slope risk monitoring and early warning system according to claim 5 is characterized in that: The process of constructing the second assessment model of the embankment slope includes: The historical accident set refers to the three-dimensional terrain data, surface vegetation parameters, and historical monitoring data corresponding to the safety accidents that occurred on the roadbed slope. The historical monitoring data refers to the monitoring data obtained when the safety accidents occurred. generating a second evaluation set according to the three-dimensional terrain data, surface vegetation parameters, and historical monitoring data in different historical accident sets, and dividing the second evaluation set into a second training set and a second test set; Constructing a second convolutional neural network, taking the three-dimensional terrain data, surface vegetation parameters, and historical monitoring data in the second training set as input data of the second convolutional neural network, and taking whether a safety accident will occur on the roadbed slope as output data of the second convolutional neural network; The second convolutional neural network is trained to obtain an initial second convolutional neural network, the initial second convolutional neural network is model verified using a second test set, and an initial second convolutional neural network that is less than or equal to a preset second test error threshold is output as a second evaluation model.

7. The artificial intelligence-based roadbed slope risk monitoring and early warning system according to claim 6 is characterized in that: The process of determining whether there is a risk area and generating early warning information includes: The current monitoring data of a single monitoring sub-area is used as real-time monitoring data, combined with the three-dimensional terrain data and surface vegetation parameters of the single monitoring sub-area, and input into the second evaluation model, and the second evaluation model is used to determine whether a safety accident will occur in the single monitoring sub-area; If so, it is marked as a risk area, and the corresponding warning information is generated and fed back to the relevant personnel. If not, it is marked as a non-risk area.

8. The artificial intelligence-based roadbed slope risk monitoring and early warning system according to claim 7 is characterized in that: The process of obtaining recommended vegetation parameters and generating vegetation optimization solutions includes: The average of the monitoring data corresponding to each non-risk area in the same monitoring scenario as the single risk area is used as the expected monitoring data of the single risk area; The rainfall, light intensity, temperature and humidity, pore water pressure, three-dimensional terrain data and surface vegetation parameters in the expected monitoring data of the single risk area are used as input data of the first assessment model; The inputted surface vegetation parameters are continuously adjusted until the output data of the first assessment model is equal to the displacement value and stress value in the expected monitoring data of the single risk area, and the surface vegetation parameters at this time are used as the recommended vegetation parameters; Vegetation optimization is carried out on the surface vegetation of this single risk area, including changing vegetation type, changing vegetation spacing, and pruning vegetation branches and leaves. The optimization plan that meets the recommended vegetation parameters is used as the vegetation optimization plan and fed back to relevant personnel.

Images