Unmanned aerial vehicle ai adaptive side slope intelligent monitoring method and system

Abstract

This invention discloses an AI-adaptive intelligent slope monitoring method and system based on unmanned aerial vehicles (UAVs), relating to the field of geological disaster monitoring technology. The method involves acquiring initial data of a target slope to construct a digital twin benchmark model of the target slope, building a UAV swarm to perform slope monitoring tasks, and saving the slope monitoring data to the UAV's onboard processing unit for edge computing. This yields incremental data on areas where changes have occurred and routine monitoring data on areas where no changes have occurred. The incremental data is processed into valid information data and transmitted to a pre-set cloud platform. The valid information data is compared and analyzed with the digital twin benchmark model to obtain several slope risk points and corresponding risk information, which is then pushed to relevant management personnel. A learning library is built by integrating the risk information from all slope risk points. The digital twin benchmark model and the onboard processing unit undergo AI adaptive learning, and subsequent slope monitoring is conducted based on the new learning data.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring technology, specifically to an intelligent slope monitoring method based on UAV AI adaptive technology. Background Technology

[0002] Slope stability monitoring is a crucial aspect of geological disaster prevention and control. Currently, it primarily employs a combination of fixed sensor deployment and regular manual inspections, which suffers from limitations such as limited monitoring points, long data update cycles, and an inability to comprehensively reflect the overall condition of the slope. While drone inspection technology has improved data acquisition efficiency in recent years, it still has the following shortcomings: First, there is a lack of deep integration with the digital model of the slope, resulting in a weak correlation between monitoring data and the actual condition. Second, the data processing capability is limited, making it impossible to achieve intelligent filtering and efficient transmission of monitoring data. Third, the system lacks self-learning ability and is difficult to adapt to complex and ever-changing geological conditions, resulting in insufficient accuracy of risk identification and timely warning. In particular, when facing large and complex slopes, existing technologies are unable to achieve the organic unity of full coverage, real-time analysis and accurate early warning. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for intelligent slope monitoring based on UAV AI adaptation, in order to solve the problems in the background technology.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an AI-adaptive slope monitoring method based on unmanned aerial vehicles (UAVs), comprising the following steps: Step S1: Obtain the initial data of the target slope, construct the digital twin benchmark model corresponding to the target slope, build a drone cluster to perform slope monitoring tasks, and save the slope monitoring data to the drone's own onboard processing unit. Step S2: The airborne processing unit performs edge calculations on the slope monitoring data to obtain the incremental change data corresponding to the changed area and the regular monitoring data of the unchanged area. The incremental change data is processed into effective information data and transmitted to the preset cloud platform. Step S3: The cloud platform compares and analyzes the effective information data with the digital twin benchmark model to obtain several slope risk points, and pushes the risk information of all slope risk points to relevant management personnel. Step S4: Integrate risk information from all slope risk points to build a learning library. Based on the learning library, perform AI adaptive learning on the digital twin benchmark model and the airborne processing unit, and conduct subsequent slope monitoring based on the new learning data.

[0005] In a preferred embodiment, the process of acquiring initial data of the target slope and constructing a digital twin benchmark model corresponding to the target slope includes: The design drawings of the target slope are obtained and the initial data corresponding to the target slope are analyzed. The initial data is used to record the geometric structure information and attribute characteristic information of the target slope. A three-dimensional twin model of the target slope is constructed based on digital twin technology. Several model regions are set on the three-dimensional twin model. The model regions and slope sub-regions are associated based on the same location. After the association is completed, the initial data of the corresponding part of the slope sub-region is mapped to the corresponding model region to construct the digital twin benchmark model of the target slope. The digital twin baseline model is used to characterize the initial state of the target slope in various regions, serving as a data reference object for subsequent target slopes. The data reference object is used to determine whether the target slope has experienced any anomalies.

[0006] In a preferred embodiment, the process of building a drone swarm to perform slope monitoring tasks and saving the slope monitoring data to the drone's onboard processing unit includes: The digital twin baseline model is simultaneously loaded into several drones to serve as the reference data for each drone to perform slope monitoring tasks. The drones compare the data obtained through real-time monitoring with the reference data to initially identify areas with anomalies. The system monitors the positions of several drones, and drones in adjacent positions communicate with each other. When any drone is performing a slope monitoring task in a certain slope area, it informs other drones to avoid the area through inter-group communication. The drone is equipped with a regular inspection mode and a key monitoring mode. When the drone performs slope monitoring tasks, it first conducts regular inspections in the corresponding slope sub-area according to the pre-set route. When an anomaly is found in a specific point area on the route, the corresponding point area is marked as a changed area, and other specific point areas are marked as unchanged areas. In the changed area, the drone's regular inspection mode is switched to key monitoring mode. In the key monitoring mode, the drone is stationed in the area of ​​change to perform 3D real-scene modeling, and all model-related data of the 3D real-scene modeling is saved as slope monitoring data to the airborne processing unit. In the regular inspection mode, the UAV monitors all unchanging areas along its flight path and obtains slope monitoring data for these unchanging areas. All slope monitoring data is then integrated and saved to the onboard processing unit.

[0007] In a preferred embodiment, the process by which an airborne processing unit performs edge calculations on slope monitoring data to obtain incremental change data for areas where changes have occurred and conventional monitoring data for areas where no changes have occurred includes: The airborne processing unit is equipped with several edge computing nodes, which decompose the slope monitoring data into several data point sets, assign an edge computing node to each data point set, and perform edge computing on their respective data point sets through the edge computing nodes. The edge computing node obtains the data point set and determines whether its own computing resources can complete the processing. If so, it directly processes the data point set; otherwise, the edge computing node is connected to the cloud, and the cloud obtains the data point set and allocates cloud resources for processing. Through processing by edge computing nodes, the slope monitoring data stored in the airborne processing unit is divided into incremental change data and regular monitoring data. The incremental change data corresponds to the area where changes have occurred, while the regular monitoring data corresponds to the area where changes have not occurred.

[0008] In a preferred embodiment, the process of processing the incremental change data into valid information data and transmitting it to a preset cloud platform includes: An information processing window is set up, which consists of several frame window slices. Based on the number of frames of the window slice, the incremental data of change is extracted into several sub-thread data streams, and the sub-thread data streams are extracted based on preset effective data elements to obtain the corresponding sub-thread effective information. Establish a data transmission channel between the airborne processing unit and the cloud platform, set several monitoring viewpoints on the data transmission channel, establish data communication between the several monitoring viewpoints, select a monitoring viewpoint, and determine whether there are any abnormalities in the external environment of the valid information of the sub-thread corresponding to the selected monitoring viewpoint. If yes, the anomaly type will be marked on the monitoring viewpoint, and a decision script for handling the current anomaly will be generated. The anomaly details of the current anomaly type will be broadcast synchronously to other monitoring viewpoints that have established data communication. If no, no operation will be performed. When an anomaly occurs in other monitoring viewpoints, the anomaly details data corresponding to the monitoring viewpoints that have previously had anomalies are compared to determine if they are the same anomaly. If they are, the decision script of the monitoring viewpoint that has previously had anomalies is selected and executed. If not, the current anomaly type is marked to the monitoring viewpoint, a decision script is constructed, and the anomaly details data corresponding to the anomaly type are broadcast synchronously to other monitoring viewpoints with data communication. Each supervisory viewpoint summarizes and integrates the valid information of the sub-thread after the exception is handled, constructs the corresponding valid information data, and transmits the valid information data to the cloud platform for storage through the data transmission channel.

[0009] In a preferred embodiment, the process of comparing and analyzing effective information data with a digital twin benchmark model using a cloud platform to obtain several slope risk points, and then pushing the risk information of all slope risk points to relevant management personnel includes: After the cloud platform receives valid information data, it breaks down the valid information data into several subsets of valid information data based on different slope sub-regions based on the regional coordinate set. The subsets of valid information data are then sent to the model area of ​​the digital twin benchmark model according to the distribution of the slope sub-regions on the target slope. The initial data of the slope sub-regions in the model area are compared and analyzed. When the initial data of a certain slope sub-region is inconsistent with the results of the comparison and analysis of the effective information data subset, the corresponding slope sub-region is marked as a slope risk sub-region. The video images collected by the drone monitoring the corresponding slope risk sub-region are retrieved for image feature extraction, and several slope risk points in the slope risk sub-region are located. Obtain the risk information recorded at each slope risk point, and push the risk information of all slope risk points under the same slope risk sub-area to the relevant management personnel who are in an idle state.

[0010] In a preferred embodiment, the process of integrating risk information from all slope risk points to build a learning library, performing AI adaptive learning on the digital twin benchmark model and airborne processing unit based on the learning library, and conducting subsequent slope monitoring based on the new learning data includes: The risk information of all slope risk points is integrated, and a corresponding risk status label is marked for each risk information to form a structured sample set. Then, a slope risk learning library is constructed. The risk information in the learning library is used as input, and the actual response status of the target slope is obtained as output. Machine learning algorithms are used to iteratively train and calibrate the key calculation parameters and evolution rules in the digital twin benchmark model. Feature dimensionality reduction and sample optimization are performed on the data in the learning library to generate a lightweight dataset suitable for edge computing. Based on the lightweight dataset, the artificial intelligence recognition algorithm embedded in the airborne processing unit is trained and updated through transfer learning and online learning techniques. During slope monitoring, continuously updated slope monitoring data is collected and input into a trained digital twin baseline model and an airborne processing unit. The digital twin baseline model performs real-time simulation and stability prediction based on the updated slope monitoring data, while the airborne processing unit performs real-time analysis and anomaly detection based on the updated slope monitoring data. The identification results of the airborne processing unit are fused and compared with the prediction results of the digital twin benchmark model. New risk cases and data verified on-site are fed back to the learning library to expand the learning library and trigger a new round of adaptive learning cycle, thereby completing the iterative optimization of the digital twin benchmark model and the artificial intelligence algorithm of the airborne processing unit.

[0011] This invention also provides an AI-adaptive slope monitoring system based on unmanned aerial vehicles (UAVs), the system comprising: The UAV module is used to acquire initial data of the target slope, construct a digital twin benchmark model corresponding to the target slope, build a UAV cluster to perform slope monitoring tasks, and save the slope monitoring data to the UAV's own onboard processing unit. The data analysis module, through the airborne processing unit, performs edge calculations on the slope monitoring data to obtain the incremental change data corresponding to the changed areas and the regular monitoring data of the unchanged areas. The incremental change data is then processed into effective information data and transmitted to the preset cloud platform. The risk location module uses the cloud platform to compare and analyze effective information data with the digital twin benchmark model to obtain several slope risk points, and pushes the risk information of all slope risk points to relevant management personnel. The AI ​​adaptive module integrates risk information from all slope risk points to build a learning library. Based on this learning library, it performs AI adaptive learning on the digital twin benchmark model and the airborne processing unit, and conducts subsequent slope monitoring based on the new learning data.

[0012] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention achieves full-domain visualization and real-time dynamic mapping of slope status by constructing a collaborative operation system of a digital twin benchmark model and a drone swarm. The drone swarm performs targeted monitoring tasks based on the digital twin model, effectively eliminating monitoring blind spots and significantly improving the spatial resolution and acquisition efficiency of monitoring data. By establishing a hierarchical data processing mechanism of edge computing and cloud processing, it realizes localized intelligent filtering and compressed transmission of monitoring data. The airborne processing unit performs edge computing on the monitoring data and uploads only the effective information data to the cloud platform, which greatly reduces the data transmission load and improves the system response speed.

[0013] 2. This invention adopts an AI adaptive learning architecture, constructing a dual learning mechanism that includes digital twin model optimization and airborne algorithm updates. By continuously integrating new monitoring data and risk cases, the system continuously improves the prediction accuracy of the digital twin benchmark model and the recognition capability of the airborne processing unit, forming an intelligent monitoring closed loop with self-evolution capabilities. Through the comparison and analysis of effective information data and the digital twin benchmark model on the cloud platform, the system achieves accurate positioning and graded early warning of slope risk points. The system can automatically identify abnormal areas of the slope, accurately locate risk points, and intelligently push risk information to relevant management personnel, significantly improving the accuracy of slope monitoring and the timeliness of early warning. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0015] Figure 1 This is a flowchart of the method of the present invention.

[0016] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0018] Example 1, please refer to Figure 1 As shown in this embodiment, the slope intelligent monitoring method based on UAV AI adaptation includes the following steps: Step S1: Obtain the initial data of the target slope, construct the digital twin benchmark model corresponding to the target slope, build a drone cluster to perform slope monitoring tasks, and save the slope monitoring data to the drone's own onboard processing unit. Step S2: The airborne processing unit performs edge calculations on the slope monitoring data to obtain the incremental change data corresponding to the changed area and the regular monitoring data of the unchanged area. The incremental change data is processed into effective information data and transmitted to the preset cloud platform. Step S3: The cloud platform compares and analyzes the effective information data with the digital twin benchmark model to obtain several slope risk points, and pushes the risk information of all slope risk points to relevant management personnel. Step S4: Integrate risk information from all slope risk points to build a learning library. Based on the learning library, perform AI adaptive learning on the digital twin benchmark model and the airborne processing unit, and conduct subsequent slope monitoring based on the new learning data.

[0019] It should be further explained that, in the specific implementation process, the process of obtaining the initial data of the target slope and constructing the corresponding digital twin benchmark model of the target slope includes: Obtain the design drawings corresponding to the target slope, and analyze the design drawings to obtain the initial data corresponding to the target slope. The initial data is used to record the geometric structure information and attribute characteristic information of the target slope. The geometric structure information includes the regional coordinate set corresponding to each of the several slope sub-regions. The regional coordinate set is used to record the coordinate values ​​of several key points on the corresponding slope sub-region, including the normal vector, curvature, elevation and slope change rate corresponding to each of the several slope sub-regions. The attribute feature information includes the slope spectral information, slope radar reflection intensity, slope cover information, and slope infrared information for each of the several slope sub-regions. A three-dimensional twin model of the target slope is constructed based on digital twin technology. Several model regions are set on the three-dimensional twin model. The model regions and slope sub-regions are associated based on the same location. After the association is completed, the initial data of the corresponding part of the slope sub-region is mapped to the corresponding model region, thereby constructing a digital twin benchmark model of the target slope.

[0020] The digital twin baseline model is used to characterize the initial state of the target slope in each region, and serves as a data reference object for subsequent target slopes. The data reference object is used to determine whether the target slope has structural anomalies.

[0021] It should be noted that the geometric structure information of the target slope is used to inform the information elements of the slope at the external level, such as the presence of protrusions or cracks in a certain slope area. The attribute characteristic information of the target slope is used to inform the information elements of the slope at the internal level, such as water accumulation or seepage in cracks in a certain slope area, or rock mass fracturing or deformation in a certain slope area.

[0022] It should be further explained that, in the specific implementation process, the process of building a drone swarm to perform slope monitoring tasks and saving the slope monitoring data to the drone's own onboard processing unit includes: The digital twin baseline model is synchronously loaded into several drones to serve as the reference data for each drone when performing slope monitoring tasks. The drones compare the data obtained through real-time monitoring with the reference data to initially identify areas with anomalies. Several drones were labeled, and the labels were recorded as follows: , =1, 2, 3, ..., n, where, For each drone, a laser detection radar, an infrared imager, a high-definition camera, and a communication module are deployed, and the number is a natural number greater than 0. The system monitors the positions of several drones and enables inter-group communication between drones in adjacent locations. When any drone is performing a slope monitoring task in a certain slope area, it can notify other drones to avoid the area through inter-group communication.

[0023] Each drone is responsible for performing slope monitoring tasks within a specific slope sub-area. The drone is equipped with a regular inspection mode and a key monitoring mode. When the drone performs slope monitoring tasks, it first performs regular inspections in the corresponding slope sub-area according to a pre-set route. When an anomaly is found in a specific location area on the flight path during inspection, the corresponding location area is marked as the changed area, and other specific location areas without anomalies are marked as unchanged areas. In the changed area, the drone's regular inspection mode is switched to the key monitoring mode.

[0024] In the key monitoring mode, the drone will park in the area where the change has occurred and start the equipment deployed on the drone to perform a three-dimensional real-scene model of the area where the change has occurred. All model-related data corresponding to the three-dimensional real-scene model will be saved as slope monitoring data to the airborne processing unit. In the regular inspection mode, the UAV monitors all unchanging areas along its flight path and obtains slope monitoring data for the corresponding unchanging areas. All the obtained slope monitoring data is then integrated and saved to the UAV's onboard processing unit.

[0025] It should be further explained that, in the specific implementation process, the airborne processing unit performs edge computing on the slope monitoring data to obtain the incremental change data corresponding to the changed areas and the regular monitoring data of the unchanged areas. The process of processing the incremental change data into valid information data and transmitting it to the preset cloud platform includes: The airborne processing unit is equipped with several edge computing nodes; The slope monitoring data is broken down into several data point sets with equal data volume, and an edge computing node is assigned to each data point set. Edge computing is then performed on the corresponding data point set through the corresponding edge computing node. The edge computing node obtains the data point set and determines whether its own computing resources can complete the processing. If so, it directly processes the data point set; otherwise, the edge computing node is connected to the cloud, and the cloud obtains the data point set and allocates cloud resources for processing. Through processing by edge computing nodes, the slope monitoring data stored in the airborne processing unit is divided into incremental change data and regular monitoring data. The incremental change data corresponds to the area where changes have occurred, while the regular monitoring data corresponds to the area where changes have not occurred.

[0026] An information processing window is set up, which consists of several frame window slices. The incremental change data is passed through the information processing window, and the incremental change data is extracted into several sub-thread data streams based on the number of frames of the window slice. The sub-thread data streams are then extracted based on preset effective data elements to obtain the corresponding sub-thread effective information. The effective data elements are the feature element information corresponding to several data points of the sub-thread data stream. The feature element information is the effective information part corresponding to the corresponding sub-thread data stream. Several feature element information obtained through effective data elements are integrated to obtain the sub-thread effective information. Establish a data transmission channel between the airborne processing unit and the cloud platform. Set several monitoring points on the data transmission channel. The number of monitoring points should be the same as the number of valid information in the sub-thread. Establish data communication between the several monitoring points.

[0027] Select a supervisory viewpoint and determine whether there is an anomaly in the external environment of the valid information of the sub-thread corresponding to the selected supervisory viewpoint. If so, mark the anomaly type to the supervisory viewpoint, generate a decision script to handle the current anomaly, and synchronously broadcast the anomaly details data corresponding to the current anomaly type to other supervisory viewpoints that have established data communication. If not, do not perform any operation. When an anomaly occurs in other supervisory viewpoints, the anomaly details data corresponding to supervisory viewpoints that have previously had anomalies are prioritized for comparison to determine if they are the same anomaly. If they are, the decision script of the supervisory viewpoint that has previously had anomalies is selected and executed to handle the current anomaly. If not, the current anomaly type is marked to the supervisory viewpoint, a decision script is constructed, and the anomaly details data corresponding to the anomaly type are broadcast synchronously to other supervisory viewpoints with data communication. The above operation is repeated until all supervisory viewpoints have completed the processing of the corresponding sub-thread's valid information. Each supervisory viewpoint summarizes and integrates the valid information of the sub-thread after the exception is handled, constructs the corresponding valid information data, and transmits the valid information data to the cloud platform for storage through the data transmission channel.

[0028] It should be further explained that, in the specific implementation process, the cloud platform compares and analyzes the effective information data with the digital twin benchmark model to obtain several slope risk points, and pushes the risk information of all slope risk points to relevant management personnel. This process includes: After the cloud platform receives valid information data, it breaks down the valid information data into several subsets of valid information data based on different slope sub-regions based on the regional coordinate set. These subsets of valid information data are then sent to the model area corresponding to the digital twin benchmark model according to the distribution location of the slope sub-regions on the target slope. The data is then compared and analyzed with the initial data of the corresponding slope sub-regions stored in the model area.

[0029] When the initial data corresponding to a certain slope sub-region is inconsistent with the results of the comparison and analysis of the effective information data subset, the corresponding slope sub-region is marked as a slope risk sub-region. The video images collected by the drone monitoring the corresponding slope risk sub-region are retrieved, the image features are extracted from the video images, and several slope risk points corresponding to the slope risk sub-region are located. Obtain the risk information recorded for each slope risk point, retrieve the personnel schedule, select the relevant management personnel who are in an idle state from the personnel schedule, and push the risk information of all slope risk points under the same slope risk sub-area to the relevant management personnel.

[0030] It should be further explained that, in the specific implementation process, the process of integrating risk information from all slope risk points to build a learning database, performing AI adaptive learning on the digital twin benchmark model and airborne processing unit based on the learning database, and conducting subsequent slope monitoring based on the new learning data includes: The risk information of all slope risk points is integrated, including historical displacement data, internal stress and strain data, microseismic monitoring data, groundwater level data, rainfall data, physical and mechanical parameters of soil and rock, and on-site inspection records and image data. Each risk information is labeled with a corresponding risk status label to form a structured sample set, and then a slope risk learning library is constructed. Using risk information from the learning library as input and the actual response state of the target slope as output, machine learning algorithms are used to iteratively train and calibrate the key computational parameters and evolution rules in the digital twin benchmark model. Deep learning technology is used to mine the nonlinear mapping relationship between risk information and slope instability mode.

[0031] Feature dimensionality reduction and sample optimization are performed on the data in the learning library to generate a lightweight dataset suitable for edge computing; based on the lightweight dataset, the artificial intelligence recognition algorithm embedded in the airborne processing unit is trained and updated through transfer learning and online learning techniques; During slope monitoring, updated slope monitoring data is continuously collected and input into a trained digital twin baseline model and an airborne processing unit. The digital twin baseline model performs real-time simulation and stability prediction based on the updated slope monitoring data, and the airborne processing unit performs real-time analysis and anomaly detection based on the updated slope monitoring data. The identification results of the airborne processing unit are fused and compared with the prediction results of the digital twin benchmark model. New risk cases and data verified on-site are fed back to the learning library to expand the learning library and trigger a new round of adaptive learning cycle, thereby completing the iterative optimization of the digital twin benchmark model and the artificial intelligence algorithm of the airborne processing unit.

[0032] Example 2, please refer to Figure 2 As shown in this embodiment, the slope intelligent monitoring system based on UAV AI adaptive technology includes: The UAV module is used to acquire initial data of the target slope, construct a digital twin benchmark model corresponding to the target slope, build a UAV cluster to perform slope monitoring tasks, and save the slope monitoring data to the UAV's own onboard processing unit. The data analysis module, through the airborne processing unit, performs edge calculations on the slope monitoring data to obtain the incremental change data corresponding to the changed areas and the regular monitoring data of the unchanged areas. The incremental change data is then processed into effective information data and transmitted to the preset cloud platform. The risk location module uses the cloud platform to compare and analyze effective information data with the digital twin benchmark model to obtain several slope risk points, and pushes the risk information of all slope risk points to relevant management personnel. The AI ​​adaptive module integrates risk information from all slope risk points to build a learning library. Based on this learning library, it performs AI adaptive learning on the digital twin benchmark model and the airborne processing unit, and conducts subsequent slope monitoring based on the new learning data.

[0033] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligent slope monitoring based on UAV AI adaptation, characterized in that, Includes the following steps: Step S1: Obtain the initial data of the target slope, construct the digital twin benchmark model corresponding to the target slope, build a drone cluster to perform slope monitoring tasks, and save the slope monitoring data to the drone's own onboard processing unit. Step S2: The airborne processing unit performs edge calculations on the slope monitoring data to obtain the incremental change data corresponding to the changed area and the regular monitoring data of the unchanged area. The incremental change data is processed into effective information data and transmitted to the preset cloud platform. Step S3: The cloud platform compares and analyzes the effective information data with the digital twin benchmark model to obtain several slope risk points, and pushes the risk information of all slope risk points to relevant management personnel. Step S4: Integrate risk information from all slope risk points to build a learning library. Based on the learning library, perform AI adaptive learning on the digital twin benchmark model and the airborne processing unit, and conduct subsequent slope monitoring based on the new learning data.

2. The intelligent slope monitoring method based on UAV AI adaptive technology according to claim 1, characterized in that, The process of acquiring initial data for the target slope and constructing a digital twin benchmark model corresponding to the target slope includes: The design drawings of the target slope are obtained and the initial data corresponding to the target slope are analyzed. The initial data is used to record the geometric structure information and attribute characteristic information of the target slope. A three-dimensional twin model of the target slope is constructed based on digital twin technology. Several model regions are set on the three-dimensional twin model. The model regions and slope sub-regions are associated based on the same location. After the association is completed, the initial data of the corresponding part of the slope sub-region is mapped to the corresponding model region to construct the digital twin benchmark model of the target slope. The digital twin baseline model is used to characterize the initial state of the target slope in various regions, serving as a data reference object for subsequent target slopes. The data reference object is used to determine whether the target slope has experienced any anomalies.

3. The intelligent slope monitoring method based on UAV AI adaptive technology according to claim 2, characterized in that, The process of building a drone swarm to perform slope monitoring tasks and saving the slope monitoring data to the drone's onboard processing unit includes: The digital twin baseline model is simultaneously loaded into several drones to serve as the reference data for each drone to perform slope monitoring tasks. The drones compare the data obtained through real-time monitoring with the reference data to initially identify areas with anomalies. The system monitors the positions of several drones, and drones in adjacent positions communicate with each other. When any drone is performing a slope monitoring task in a certain slope area, it informs other drones to avoid the area through inter-group communication. The drone is equipped with a regular inspection mode and a key monitoring mode. When the drone performs slope monitoring tasks, it first conducts regular inspections in the corresponding slope sub-area according to the pre-set route. When an anomaly is found in a specific point area on the route, the corresponding point area is marked as a changed area, and other specific point areas are marked as unchanged areas. In the changed area, the drone's regular inspection mode is switched to key monitoring mode. In the key monitoring mode, the drone is stationed in the area of ​​change to perform 3D real-scene modeling, and all model-related data of the 3D real-scene modeling is saved as slope monitoring data to the airborne processing unit. In the regular inspection mode, the UAV monitors all unchanging areas along its flight path and obtains slope monitoring data for these unchanging areas. All slope monitoring data is then integrated and saved to the onboard processing unit.

4. The intelligent slope monitoring method based on UAV AI adaptive technology according to claim 3, characterized in that, The process by which the airborne processing unit performs edge calculations on the slope monitoring data to obtain the incremental change data corresponding to the changed areas and the regular monitoring data of the unchanged areas includes: The airborne processing unit is equipped with several edge computing nodes, which decompose the slope monitoring data into several data point sets, assign an edge computing node to each data point set, and perform edge computing on their respective data point sets through the edge computing nodes. The edge computing node obtains the data point set and determines whether its own computing resources are sufficient to complete the processing. If so, it processes the data point set directly; otherwise, it connects the edge computing node to the cloud, where the cloud obtains the data point set and allocates cloud resources for processing. Through processing by edge computing nodes, the slope monitoring data stored in the airborne processing unit is divided into incremental change data and regular monitoring data. The incremental change data corresponds to the area where changes have occurred, while the regular monitoring data corresponds to the area where changes have not occurred.

5. The intelligent slope monitoring method based on UAV AI adaptive technology according to claim 4, characterized in that, The process of processing incremental data into valid information data and transmitting it to a pre-defined cloud platform includes: An information processing window is set up, which consists of several frame window slices. Based on the number of frames of the window slice, the incremental data of change is extracted into several sub-thread data streams, and the sub-thread data streams are extracted based on preset effective data elements to obtain the corresponding sub-thread effective information. Establish a data transmission channel between the airborne processing unit and the cloud platform, set several monitoring viewpoints on the data transmission channel, establish data communication between the several monitoring viewpoints, select a monitoring viewpoint, and determine whether there are any abnormalities in the external environment of the valid information of the sub-thread corresponding to the selected monitoring viewpoint. If yes, the anomaly type will be marked on the monitoring viewpoint, and a decision script for handling the current anomaly will be generated. The anomaly details of the current anomaly type will be broadcast synchronously to other monitoring viewpoints that have established data communication. If no, no operation will be performed. When an anomaly occurs in other monitoring viewpoints, the anomaly details data corresponding to the monitoring viewpoints that have previously had anomalies are compared to determine if they are the same anomaly. If they are, the decision script of the monitoring viewpoint that has previously had anomalies is selected and executed. If not, the current anomaly type is marked to the monitoring viewpoint, a decision script is constructed, and the anomaly details data corresponding to the anomaly type are broadcast synchronously to other monitoring viewpoints with data communication. Each supervisory viewpoint summarizes and integrates the valid information of the sub-thread after the exception is handled, constructs the corresponding valid information data, and transmits the valid information data to the cloud platform for storage through the data transmission channel.

6. The intelligent slope monitoring method based on UAV AI adaptive technology according to claim 5, characterized in that, The process of comparing and analyzing effective information data with a digital twin benchmark model through a cloud platform to obtain several slope risk points, and then pushing the risk information of all slope risk points to relevant management personnel includes: After the cloud platform receives valid information data, it breaks down the valid information data into several subsets of valid information data based on different slope sub-regions based on the regional coordinate set. The subsets of valid information data are then sent to the model area of ​​the digital twin benchmark model according to the distribution of the slope sub-regions on the target slope. The initial data of the slope sub-regions in the model area are compared and analyzed. When the initial data of a certain slope sub-region is inconsistent with the results of the comparison and analysis of the effective information data subset, the corresponding slope sub-region is marked as a slope risk sub-region. The video images collected by the drone monitoring the corresponding slope risk sub-region are retrieved for image feature extraction, and several slope risk points in the slope risk sub-region are located. Obtain the risk information recorded at each slope risk point, and push the risk information of all slope risk points under the same slope risk sub-area to the relevant management personnel who are in an idle state.

7. The intelligent slope monitoring method based on UAV AI adaptive technology according to claim 6, characterized in that, The process of integrating risk information from all slope risk points to build a learning library, performing AI adaptive learning on the digital twin benchmark model and airborne processing unit based on the learning library, and conducting subsequent slope monitoring based on the new learning data includes: The risk information of all slope risk points is integrated, and a corresponding risk status label is marked for each risk information to form a structured sample set. Then, a slope risk learning library is constructed. The risk information in the learning library is used as input, and the actual response status of the target slope is obtained as output. Machine learning algorithms are used to iteratively train and calibrate the key calculation parameters and evolution rules in the digital twin benchmark model. Feature dimensionality reduction and sample optimization are performed on the data in the learning library to generate a lightweight dataset suitable for edge computing. Based on the lightweight dataset, the artificial intelligence recognition algorithm embedded in the airborne processing unit is trained and updated through transfer learning and online learning techniques. During slope monitoring, continuously updated slope monitoring data is collected and input into a trained digital twin baseline model and an airborne processing unit. The digital twin baseline model performs real-time simulation and stability prediction based on the updated slope monitoring data, while the airborne processing unit performs real-time analysis and anomaly detection based on the updated slope monitoring data. The identification results of the airborne processing unit are fused and compared with the prediction results of the digital twin benchmark model. New risk cases and data verified on-site are fed back to the learning library to expand the learning library and trigger a new round of adaptive learning cycle, thereby completing the iterative optimization of the digital twin benchmark model and the artificial intelligence algorithm of the airborne processing unit.

8. A slope intelligent monitoring system based on UAV AI adaptive technology, used to implement the slope intelligent monitoring method according to any one of claims 1 to 7, characterized in that, The system includes: The UAV module is used to acquire initial data of the target slope, construct a digital twin benchmark model corresponding to the target slope, build a UAV cluster to perform slope monitoring tasks, and save the slope monitoring data to the UAV's own onboard processing unit. The data analysis module, through the airborne processing unit, performs edge calculations on the slope monitoring data to obtain the incremental change data corresponding to the changed areas and the regular monitoring data of the unchanged areas. The incremental change data is then processed into effective information data and transmitted to the preset cloud platform. The risk location module uses the cloud platform to compare and analyze effective information data with the digital twin benchmark model to obtain several slope risk points, and pushes the risk information of all slope risk points to relevant management personnel. The AI ​​adaptive module integrates risk information from all slope risk points to build a learning library. Based on this learning library, it performs AI adaptive learning on the digital twin benchmark model and the airborne processing unit, and conducts subsequent slope monitoring based on the new learning data.

Images