A method for low-altitude monitoring of unmanned aerial vehicles

Abstract

The present application relates to a method for low-altitude monitoring by unmanned aerial vehicles. The method comprises: scanning an archaeological site by means of a camera device carried by an unmanned aerial vehicle and constructing a preliminary two-dimensional plane map of the archaeological site based on the scanning results, obtaining and constructing a preliminary three-dimensional model based on the elevation data of the archaeological site and the preliminary two-dimensional plane map, formulating a flight strategy for instructing the unmanned aerial vehicle to perform a second flight based on a supplementary flight path obtained based on the divided fuzzy area, obtaining and dynamically adjusting the flight strategy based on the sound pressure distribution information and airflow field information within the archaeological site, predicting the risk area in the preliminary three-dimensional model to optimize the flight path again, identifying and marking the archaeological site defects and key areas with historical value of the corresponding archaeological site in the supplementary three-dimensional model, and delegating logistics tasks based on the supplementary three-dimensional model after marking, which can effectively solve the problem of low intelligence of unmanned aerial vehicles in the current large-scale excavation of archaeological sites.

Description

Technical Field

[0001] The present application relates to the technical field of drone monitoring, and in particular to a method for low-altitude drone monitoring. Background Art

[0002] In the process of excavating archaeological sites, traditional manual processing methods are inefficient; with the large-scale and multi-field application of drones, the excavation process of archaeological sites has ushered in a new turning point in improving efficiency.

[0003] However, in the practical process of using drones for large-scale archaeological site excavation, the role of drones is still limited to the early stages of manual control for archaeological site detection, and the degree of intelligence is relatively low. Summary of the invention

[0004] Based on this, it is necessary to provide a UAV low-altitude monitoring method to address the above technical issues.

[0005] The present application provides a method for low-altitude monitoring by unmanned aerial vehicles, which is applied to the excavation of archaeological sites. The method includes:

[0006] In response to the drone being launched, edge detection of the archaeological site is performed using a camera device carried by the drone, and the drone is instructed to perform a first flight according to the result of the edge detection;

[0007] When the drone makes its first flight, the archaeological site is scanned and a preliminary two-dimensional plan of the archaeological site is constructed based on the scanning results; the preliminary two-dimensional plan includes the terrain contour and terrain texture in the archaeological site; the terrain contour includes the contours of buildings, roads and walls; the terrain texture includes the roughness of ground stones, the arrangement of bricks and tiles, the looseness of the soil and the texture of vegetation; the preliminary two-dimensional plan is used to indicate the orientation and area of ​​the archaeological site;

[0008] Acquire and construct a preliminary three-dimensional model based on the elevation data of the archaeological site and the preliminary two-dimensional plan; the elevation data includes the undulation of the terrain of the archaeological site, the height of the buildings of the archaeological site, and the three-dimensional structure information; divide the fuzzy area in the preliminary three-dimensional model according to the flight path of the first flight and the preset accuracy requirements, and generate a supplementary flight path passing through the fuzzy area, so as to formulate a flight strategy for instructing the UAV to perform a second flight according to the supplementary flight path;

[0009] Obtain and dynamically adjust the flight strategy based on the sound pressure distribution information and airflow field information within the archaeological site, and generate recommended specifications of the UAV that meet the dynamic adjustment results; the dynamic adjustment includes adjusting the flight path, flight altitude and flight speed in the UAV's flight strategy;

[0010] Use the time series analysis prediction model to predict the risk areas in the preliminary three-dimensional model and re-optimize the flight path in the flight strategy to avoid the flight path in the flight strategy including risk areas; risk areas include natural collapse areas and areas affected by the first flight of the drone;

[0011] Supplementary data was obtained through the second flight, and the preliminary three-dimensional model was registered and aligned with the supplementary data to obtain a supplementary three-dimensional model. The archaeological site defects and key areas of historical value of the corresponding archaeological sites in the supplementary three-dimensional model were identified and marked through a deep learning model; archaeological site defects include surface cracks and wall damage.

[0012] The above-mentioned unmanned aerial vehicle low-altitude monitoring method can detect archaeological sites through the camera equipment carried by the unmanned aerial vehicle, and instruct the unmanned aerial vehicle to smoothly enter the archaeological site according to the detection results. After entering, the archaeological site is scanned, and a preliminary two-dimensional plane model of the terrain contour and terrain texture is first obtained. Then, on the basis of the preliminary two-dimensional plane model, the preliminary three-dimensional model of the archaeological site is obtained in combination with the elevation data. Then, according to the supplementary flight path, the fuzzy area in the preliminary three-dimensional model is filled to obtain the flight measurement that instructs the unmanned aerial vehicle to re-enter the archaeological site, and combined with the influence of sound pressure and airflow on the unmanned aerial vehicle, the flight path, flight altitude and flight speed in the flight strategy of the unmanned aerial vehicle are adjusted. Then, through the time series analysis prediction model, the risk area in the preliminary three-dimensional model is predicted, and the flight path is further adjusted. Finally, when the unmanned aerial vehicle is able to fly effectively, the acquisition of supplementary data and the execution of subsequent logistics tasks are considered to formulate a task plan for the unmanned aerial vehicle, thereby forming a complete set of unmanned aerial vehicle control system, which effectively solves the problem of low intelligence level of unmanned aerial vehicles in the current large-scale excavation of archaeological sites. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 The figure is a flow chart of a method for low-altitude monitoring of unmanned aerial vehicles in one embodiment. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0016] In the description of the present application, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0017] In this application, unless otherwise clearly specified and limited, the terms "initial", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0018] It should be noted that when an element is considered to be "connected" to another element, it can be directly connected to the other element, or connected to the other element through an intermediate element. In addition, the "connection" in the following embodiments should be understood as "electrical connection", "communication connection", etc. if there is transmission of electrical signals or data between the connected objects.

[0019] In the process of using drones to collect large-scale data on ancient monument monitoring, there are a series of complex technical contradictions and challenges. First, drones need to collect data efficiently while minimizing interference and damage to ancient monuments, which requires drones to have precise flight control capabilities and highly sensitive sensors. Secondly, after obtaining massive monitoring data, how to realize automatic data processing and key feature extraction is another urgent problem to be solved. Traditional manual processing methods are inefficient, and the use of machine learning algorithms for automatic processing faces difficulties such as algorithm selection, parameter tuning, and computing resource optimization. Furthermore, when extracting key feature information of ancient monuments, how to balance the comprehensiveness and redundancy of features and find the optimal feature combination is also a problem worthy of in-depth study. In addition, in addition to performing monitoring missions, drones also need to undertake the heavy responsibility of logistics, such as delivering monitoring equipment to sensitive areas that are difficult to reach, providing necessary tools for the maintenance of ancient buildings, or delivering archaeological site restoration materials; how to effectively plan the logistics and distribution of drones carrying maintenance tools, monitoring equipment, and even restoration materials while ensuring the quality of monitoring has become a new challenge. This requires an unprecedented fine balance in drone design, power configuration and task scheduling, and ensuring the coordinated optimization of monitoring and logistics tasks under limited payload and battery life to form an efficient working mode.

[0020] In an exemplary embodiment, Figure 1 As shown, a method for low-altitude monitoring of unmanned aerial vehicles is provided, including the following steps 202 to 218. Among them:

[0021] Step 202, in response to the drone being launched, edge detection of the archaeological site is performed using the camera equipment carried by the drone, and the drone is instructed to perform a first flight based on the result of the edge detection.

[0022] Specifically, after the drone takes off, it uses the onboard camera equipment to conduct an all-round, multi-angle scan of the target archaeological site area to obtain complete archaeological site image data; the camera equipment can be a high-definition camera equipment.

[0023] Specifically, the outer boundary of the archaeological site is extracted from the image data of the archaeological site, and the center of gravity of the polygon formed by the outer boundary is calculated based on the outer boundary of the archaeological site using a geometric algorithm. Then, the perpendicular bisectors of each connecting line are calculated by connecting the center of gravity point with the connecting points of different straight lines on the outer boundary. The longest perpendicular bisector is selected as the path reference for the first flight of the UAV inside the archaeological site. The calculated perpendicular bisector path reference data is sent to the UAV, instructing the UAV to fly from the outer boundary of the archaeological site to the inner area along the path.

[0024] In an exemplary embodiment, the steps of performing edge detection on the archaeological site by using a camera device carried by a drone and instructing the drone to perform a first flight according to the result of the edge detection include:

[0025] The image data of the archaeological site is obtained by performing edge detection on the archaeological site through a camera device;

[0026] Calling an image processing algorithm to perform a first preprocessing operation on the archaeological site image data; the first preprocessing operation includes noise removal and image enhancement;

[0027] Using a deep learning-based edge detection algorithm, the image data of the archaeological site after the first preprocessing operation is analyzed to extract the outer boundary contour of the archaeological site;

[0028] Calculate the perpendicular bisector of the outer boundary of the archaeological site based on the outer boundary contour, which will serve as a reference for the path of the drone to enter the interior of the archaeological site for the first flight;

[0029] The path reference of the first flight is transmitted to the UAV to instruct the UAV to start the first flight.

[0030] Specifically, the acquired archaeological site image data can be transmitted to the ground control center, and the archaeological site image data can be subjected to a first preprocessing operation through image processing algorithms such as median filtering and histogram equalization to remove noise interference and improve image quality. An edge detection algorithm based on deep learning, such as the Canny algorithm (Canny Edge Detection, canny edge detection operator) or the Sobel algorithm (Sobel operator, Sobel operator), is used to analyze the archaeological site image data after the first preprocessing operation to identify and extract the outer boundary contour of the archaeological site.

[0031] Furthermore, while the drone is flying along the perpendicular bisector, it continuously uses the onboard camera to perform high-definition scanning of the interior of the archaeological site to obtain detailed internal archaeological site image data. The internal archaeological site image data obtained by the drone is transmitted to the ground control center in real time, and the multi-angle and multi-directional archaeological site image data are registered and fused through image stitching algorithms such as SIFT feature point matching and RANSAC algorithm. Then, the SFM (Structure from Motion) algorithm is used to estimate the camera pose and three-dimensional point cloud based on the stitched archaeological site image data, and finally a complete internal three-dimensional model of the archaeological site is generated through algorithms such as Poisson Surface Reconstruction, providing data support for subsequent archaeological research on the archaeological site.

[0032] Furthermore, when the drone scans the archaeological site, the flight altitude can be set to 50 meters and the flight speed can be set to 5 meters per second. The camera equipment uses a high-definition camera with a resolution of 4000×3000 pixels, and cooperates with a 3-axis gimbal to ensure shooting stability. By setting the flight path points, the archaeological site is collected in all directions and at multiple angles with an overlap of 30%, and more than 200 archaeological site image data are obtained. After the archaeological site image data is transmitted to the ground control center, the image is denoised using a 5×5 median filter algorithm, and then the image contrast is enhanced by a 256-level histogram equalization algorithm to improve the accuracy of edge detection. The Canny algorithm is used for edge detection, and the high threshold is set to 200 and the low threshold is set to 50 to obtain a clear outer boundary contour of the archaeological site. The Harris Corner Detection algorithm is used to extract feature points on the boundary contour, and the Graham scan method is used to calculate the centroid coordinates of the boundary polygon. Taking the centroid as the starting point, lines are connected with points on the boundary, and the perpendicular bisectors of each connecting line are calculated. The longest bisector is selected as the reference for the UAV's flight path.

[0033] Furthermore, the drone flew along the perpendicular bisector at a speed of 10 m / s and an altitude of 50 m according to the reference path, scanning the interior of the archaeological site, taking a high-definition photo every 10 m, and obtaining a total of 500 photos of the internal archaeological site as archaeological site image data. The acquired archaeological site image data was uploaded to the control center, and feature points were extracted by the SIFT algorithm (Scale-invariant feature transformation), and the RANSAC algorithm (Random sample consensus) was used to eliminate the wrong matching points. Then, the homography matrix between the archaeological site image data was solved by the least squares method to complete the image registration. The SFM algorithm was used to calculate the camera pose and three-dimensional point cloud, and the point cloud density was set to 100 points per square meter. The Poisson surface reconstruction algorithm was used to generate a three-dimensional mesh model of the interior of the archaeological site, with the number of triangles reaching more than 5 million, providing high-precision digital model support for archaeological research.

[0034] Step 204, when the UAV is performing its first flight, the archaeological site is scanned by a camera device, and a preliminary two-dimensional plan view of the archaeological site is constructed based on the scanning results; the preliminary two-dimensional plan view includes the terrain contour and terrain texture in the archaeological site; the terrain contour includes the contours of buildings, roads and walls; the terrain texture includes the roughness of ground stones, the arrangement of bricks and tiles, the looseness of the soil and the texture of vegetation; the preliminary two-dimensional plan view is used to indicate the orientation and area of ​​the archaeological site.

[0035] In an exemplary embodiment, during the first flight of the drone toward the interior of the site along the perpendicular bisector of the outer boundary, the upper and lower altitude limits can also be calculated based on the scanning results of the site, so that the flight altitude of the drone during the first flight is not lower than the lower altitude limit and not higher than the upper altitude limit.

[0036] In an exemplary embodiment, the steps of scanning the archaeological site by a camera device and constructing a preliminary two-dimensional plan view of the archaeological site based on the scanning result include:

[0037] Performing a second preprocessing operation on the scanned result using a computer vision algorithm; the second preprocessing operation includes noise removal and image enhancement;

[0038] Identifying terrain contours in the scanned result after the second preprocessing operation by edge detection and contour extraction algorithms; the edge detection and contour extraction algorithms include Canny algorithm and Sobel algorithm;

[0039] Using texture analysis technology to identify terrain texture in the scanned result after the second preprocessing operation; the texture analysis technology includes gray level co-occurrence matrix and local binary pattern;

[0040] A preliminary two-dimensional plan of terrain contours and terrain texture is constructed using computer graphics techniques; computer graphics techniques include AutoCAD and ArcGIS software techniques.

[0041] Specifically, during the first flight, the drone used high-definition camera equipment to conduct a detailed scan of the archaeological site area to obtain high-resolution image data to ensure that the image quality meets the needs of subsequent analysis. The acquired image data was transmitted to the ground control center, and the computer vision algorithm was used to perform a second preprocessing operation on the image, including image enhancement and denoising, to improve the image quality and the accuracy of feature extraction. Edge detection was performed by applying the Canny algorithm, which includes Gaussian filtering, calculating gradient amplitude and direction, non-maximum suppression, double threshold processing and edge connection. Then, a contour extraction algorithm based on the region growing method was used to connect the connected edge pixels into a complete contour line to identify the terrain contours of buildings, roads and walls in the image. Texture analysis techniques such as gray-level co-occurrence matrix (GLCM) and local binary pattern (LBP) were used to identify and classify the surface materials in the image.

[0042] Furthermore, for the gray-level co-occurrence matrix, appropriate distance and direction parameters are selected, the gray-level co-occurrence matrix is ​​calculated, and texture features such as energy, entropy, and contrast are extracted; for the local binary pattern, an appropriate window size and threshold are set, and the LBP code is calculated for each pixel. The histogram of the LBP code is statistically used as the texture feature to extract the surface texture features of different materials such as stones, bricks and tiles, soil, and vegetation to obtain the terrain texture. According to the texture analysis results, detailed features such as the roughness of stones, the arrangement of bricks and tiles, and the looseness of soil are further extracted to provide richer information for subsequent archaeological site analysis.

[0043] Furthermore, the terrain contour obtained by the comprehensive contour line extraction and the terrain texture obtained by the texture analysis are used to calculate the approximate location and scope of the archaeological site using the circumscribed rectangle or minimum circumscribed polygon of the contour line, and the actual geographic coordinates and area are converted into pixel scale information to obtain the accurate location coordinates and coverage area of ​​the archaeological site, providing basic data support for subsequent archaeological work. Based on the location, area, and main structure and material distribution information of the archaeological site obtained, computer drawing technology such as AutoCAD or ArcGIS is used to import contour line and material distribution data, create layers, draw vector graphics of buildings, roads, and walls, add annotations and legends, etc., to construct a preliminary two-dimensional plan of the archaeological site, intuitively present the overall layout and spatial relationship of the archaeological site, and lay the foundation for subsequent archaeological excavation and research work.

[0044] Furthermore, the drone scanned the archaeological site area at a height of 50 meters at a speed of 3 meters per second, and used a 40-megapixel high-definition camera to obtain 0.5-cm resolution image data of the archaeological site. The image data of the archaeological site was transmitted to the ground control center, and the image was denoised using a Gaussian filter, and the image contrast was enhanced using an adaptive histogram equalization algorithm. Edge detection was performed using the Canny algorithm, with the standard deviation of the Gaussian filter set to 1.5, the low threshold set to 50, and the high threshold set to 150, to extract clear edges. The contour extraction algorithm based on the region growing method was used, with the growth similarity threshold set to 20 and the connectivity set to 8 neighborhoods, to extract the complete contour lines of buildings, roads, and walls. The GLCM and LBP algorithms were used for texture analysis. For GLCM, a 45-degree direction and a distance of 2 were selected to calculate the co-occurrence matrix of 16 gray levels and extract five texture features such as energy, entropy, and contrast. For LBP, a 3x3 window and a threshold of 8 were selected to extract 59 LBP-encoded histogram features. The texture features were classified using a support vector machine (SVM) to identify the materials of stone, bricks, soil, and vegetation with an accuracy of more than 95%. According to the material classification results, detailed features such as the roughness of stones, the regularity of brick and tile arrangement, and the looseness of soil were further calculated. The contour line and material distribution information were superimposed, and the location coordinates and coverage area of ​​the archaeological site were calculated by the minimum circumscribed rectangle of the contour line with an accuracy of centimeters. The contour line and material distribution data were imported using AutoCAD, and a two-dimensional plan was drawn at a scale of 1:500. Elements such as legends, directions, and scales were added to generate a preliminary two-dimensional plan of the archaeological site that met the archaeological surveying and mapping specifications.

[0045] Step 206, obtaining and constructing a preliminary three-dimensional model based on the elevation data of the archaeological site and the preliminary two-dimensional plan; the preliminary three-dimensional model is generated by a three-dimensional module of multi-view geometry or volume reconstruction; the elevation data includes the undulation of the terrain of the archaeological site, the height of the buildings of the archaeological site and the three-dimensional structural information; the elevation data is obtained by the stereo measurement equipment and structured light scanning equipment carried by the drone.

[0046] Specifically, based on the position information in the preliminary two-dimensional plan, the drone is controlled to conduct a second data collection flight over the archaeological site along a preset second flight path to obtain elevation data; texture data can also be obtained at the same time to make more detailed settings for the preliminary three-dimensional model.

[0047] In an exemplary embodiment, the steps of obtaining and constructing a preliminary three-dimensional model based on the elevation data of the archaeological site and the preliminary two-dimensional plan include:

[0048] In the case of the first flight of the UAV, in response to the completion of the construction of the preliminary two-dimensional plan, a second flight path for collecting elevation data is formulated according to the orientation of the archaeological site indicated by the preliminary two-dimensional plan;

[0049] When the UAV flies along the second flight path, the surface of the archaeological site is three-dimensionally scanned by a stereo measurement device to obtain point cloud data of a first precision, and the surface of the archaeological site is three-dimensionally scanned by a structured light scanning device to obtain point cloud data of a second precision; the point cloud density of the point cloud data of the first precision is not less than 100 points per square meter; the second precision is greater than the first precision;

[0050] The point cloud data of the first precision and the point cloud data of the second precision are registered and fused, and a complete three-dimensional point cloud model is obtained by position calibration under a unified coordinate system;

[0051] Perform filtering, denoising and downsampling on the 3D point cloud model, and extract the elevation data of the archaeological site;

[0052] The terrain contours and terrain textures of the archaeological site in the preliminary two-dimensional plan are mapped to the three-dimensional point cloud model, and the surface of the point cloud is reconstructed through a multi-view geometry algorithm to obtain a preliminary three-dimensional model.

[0053] Specifically, the surface of the archaeological site is scanned in three dimensions by using a drone equipped with high-precision stereo measurement equipment, such as a laser radar or a binocular vision system. According to the scale and accuracy requirements of the archaeological site, the point cloud density is set to 500 or 1,000 points per square meter to meet the needs of high-precision modeling and obtain point cloud data with the first accuracy. For complex or delicate architectural structures, such as sculptures and reliefs, a structured light scanning device is used for close-range, high-precision three-dimensional scanning to obtain point cloud data with the second accuracy of 0.5 mm or 0.1 mm resolution to ensure accurate reconstruction of delicate structures. The acquired stereo measurement and structured light scanning data are aligned and fused, and the ground control points are measured by GPS or a total station to establish a unified coordinate system. Then, the ICP (Iterative Closest Point) algorithm is used to accurately align the first accuracy point cloud data with the second accuracy point cloud data, so as to achieve seamless fusion of data collected by different devices. The fused 3D point cloud model is filtered, denoised and down-sampled. The data quality is measured by indicators such as point cloud density and noise level. The point cloud is down-sampled using octree or voxelization methods to improve processing efficiency while ensuring reconstruction accuracy. The elevation information of terrain and buildings in the 3D point cloud model is extracted to generate a DEM (Digital Elevation Model).

[0054] Furthermore, the terrain contours and terrain textures of the archaeological site in the two-dimensional plane map are mapped to the three-dimensional point cloud model, and the surface of the point cloud is reconstructed through the multi-view geometry algorithm SFM and MVS (multi-view stereo). The SFM algorithm estimates the camera pose through feature point matching and motion recovery, and the MVS algorithm generates a dense point cloud through dense matching, and then performs surface reconstruction to generate a preliminary three-dimensional model with high-resolution texture. The reconstructed preliminary three-dimensional model is post-processed, and the missing parts of the model are repaired using a hole filling algorithm based on normal vector diffusion, and the model is simplified using an adaptive mesh simplification algorithm to reduce data redundancy and improve rendering efficiency. The texture of the model is optimized through an optimization method based on texture synthesis, eliminating discontinuities at the texture seams, improving the visual quality of the model, and finally generating a high-precision, high-realism preliminary three-dimensional model.

[0055] Furthermore, at an altitude of 30 meters above the archaeological site, the drone collected data at a speed of 2 meters per second in an S-shaped flight path. The laser radar scanning frequency was set to 100Hz, and the shooting interval of the binocular camera was 0.5 seconds, obtaining high-precision three-dimensional data with a point cloud density of 800 points per square meter. For fine structures such as sculptures, a structured light scanner was used to scan at a resolution of 0.2 mm, a scanning distance of 1 meter, and a scanning time of 10 seconds to obtain a high-precision point cloud of 20 million points. Using 5 GPS control points and 10 ground landmarks measured by the total station, a unified coordinate system was established by the least squares method, quaternions were used to represent rotations, and translation vectors were used to represent displacements. After 20 iterations, a point cloud fusion result with a registration error of less than 1 cm was obtained. A statistical filter was applied to remove noise, and downsampling was performed with a voxel size of 0.5 meters, reducing the number of point clouds to 1 / 4 of the original, and topographic points were extracted to generate a DEM with a resolution of 1 meter. 100,000 feature points were extracted through the SIFT algorithm, 1,000 key frames were selected using the incremental reconstruction method of SFM, 20 million dense matching points were generated using the CMVS algorithm, and a high-precision mesh model of 5 million triangular facets was obtained through Poisson reconstruction. 200 holes were filled with the normal diffusion algorithm, the Edge Collapse algorithm was used to simplify the number of mesh facets to 1 million, and the GraphCut algorithm based on texture fusion optimized 6,000 texture images to generate a seamless texture map with a resolution of 8192x8192, and finally a high-precision preliminary 3D model with an error of less than 5 cm from the actual archaeological site was obtained.

[0056] Step 208, based on the flight path of the first flight and the preset accuracy requirement, divide the fuzzy area in the preliminary three-dimensional model, and generate a supplementary flight path passing through the fuzzy area, so as to formulate a flight strategy for instructing the UAV to perform a second flight based on the supplementary flight path.

[0057] Specifically, the preliminary 3D model is analyzed, and the accuracy and integrity of the model are evaluated by calculating indicators such as point cloud density and surface normal vector continuity. The fuzzy areas in the model can be determined, and the parts with missing information can also be determined. The calculation method of surface normal vector continuity includes calculating the normal vector of each point through PCA (principal components analysis), and then comparing the angles between the normal vectors of adjacent points. The larger the angle, the worse the continuity of the normal vector. If it exceeds a certain threshold, it can be determined as a fuzzy area.

[0058] In an exemplary embodiment, according to the flight path of the first flight and the preset accuracy requirement, the steps of dividing the fuzzy area in the preliminary three-dimensional model, generating a supplementary flight path passing through the fuzzy area, and formulating a flight strategy for instructing the drone to perform a second flight according to the supplementary flight path include:

[0059] The integrity of the preliminary 3D model is analyzed by integrity indicators to determine whether there are fuzzy areas in the preliminary 3D model; the integrity indicators include point cloud density and surface normal vector continuity;

[0060] If so, the octree space partitioning algorithm is used to divide the archaeological site into several voxel grids, and the point cloud density and surface normal vector continuity in each voxel grid are evaluated, and the voxel grids whose point cloud density or surface normal vector continuity does not meet the preset threshold are marked as fuzzy voxel grids;

[0061] Perform cluster analysis on all fuzzy voxel grids, merge adjacent fuzzy voxel grids to form several connected data collection areas to be supplemented, and calculate the position and size of the data collection area to be supplemented according to the central coordinates and boundary range of each data collection area to be supplemented;

[0062] According to the location and size of the area to be supplemented for data collection, a supplementary flight path covering the area to be supplemented for data collection is generated, and the flight altitude, flight speed, angle of the camera equipment and sampling frequency of the UAV are set based on the supplementary flight path; according to the supplementary flight path, the flight altitude, flight speed, angle of the camera equipment and sampling frequency of the UAV, the flight strategy for the second flight is obtained.

[0063] Specifically, the preliminary three-dimensional model is superimposed and analyzed with the path data of the first flight. By comparing the sampling interval of the point cloud data with the preset accuracy requirements, the fuzzy areas caused by omissions in the flight path or insufficient sampling frequency are identified. According to the model analysis results, the octree space partitioning algorithm is used to divide the archaeological site area into several voxel grids. The octree space partitioning includes determining the initial size of the octree according to the spatial range of the point cloud data, recursively partitioning each node until the number of point clouds or voxel size in the node reaches the preset threshold, and then performing density and normal vector continuity analysis on the point cloud in each leaf node. If the preset threshold is not met, the voxel is marked as a fuzzy voxel grid. All fuzzy voxel grids are clustered and analyzed using density-based clustering algorithms, including DBSCAN (Density-Based Spatial Clustering of Applications with Noise). Fuzzy voxel grids are clustered according to the spatial distance and size similarity between them. Fuzzy voxel grids with close distance and similar size are merged into a region for supplementary data collection, and the center coordinates and boundary range of each region are calculated. According to the location and size of the region for supplementary data collection, a supplementary flight path covering all areas is generated.

[0064] Furthermore, heuristic search algorithms, including the A* algorithm, are used to optimize the flight path of the UAV. The objective functions include covering all areas to be supplemented with data, minimizing the flight distance, ensuring a certain overlap rate, etc., while considering factors such as the UAV's dynamic constraints and obstacle avoidance. The optimized supplementary flight path is converted into UAV control instructions, and the waypoint information, flight mode, speed and other parameters are transmitted to the UAV's flight control system using the MAVLink protocol, and uploaded to the UAV through the ground control station, instructing the UAV to perform a second data collection flight on the fuzzy area according to the specified path.

[0065] In one embodiment, in the second flight, the drone performs low-altitude hovering and multi-angle data collection on the blurred area according to the supplementary flight path to obtain high-precision point cloud and image data. The ICP algorithm is used for point cloud registration, and the TSDF (Truncated Signed Distance Function) or Poisson surface reconstruction algorithm is used for point cloud fusion and gridding. The supplementary collected data is fused with the preliminary 3D model to improve the overall accuracy and integrity of the model, and finally a high-quality 3D reconstruction result of the archaeological site is obtained.

[0066] Specifically, the preliminary three-dimensional model was analyzed, and the point cloud data was loaded using Cloud Compare software. The radius for point cloud density calculation was set to 0.5 meters, and the number of neighboring points within 0.5 meters around each point was counted. If the point cloud density was less than 100 points / square meter, it was determined to be a fuzzy area.

[0067] Furthermore, the Normal Estimation class in the PCL library is used to calculate the normal vector of each point, and the search radius is set to 1 meter. The normal vector is calculated by the eigenvalue and eigenvector of the covariance matrix, and then the cosine value of the angle between the normal vectors of adjacent points is calculated. If it is less than 0.8, the normal vector is considered discontinuous and marked as a fuzzy area. The point cloud data is superimposed and displayed with the route trajectory of the first flight. If the route spacing is greater than 10 meters or the sampling interval is greater than 0.5 seconds, the data collection in this area is considered insufficient and marked as a fuzzy area. The octree module is used to divide the point cloud into octrees, setting the minimum voxel size to 1 meter and the maximum number of points to 1000. The octree is recursively divided until the stop condition is reached. The density and normal vector continuity of the point cloud in each leaf node are evaluated, and the nodes that do not meet the threshold are marked as fuzzy areas. The density-based DBSCAN clustering algorithm was applied to all fuzzy area point clouds, with the search radius set to 5 meters and the minimum number of points set to 100. The connected areas in the clustering results were extracted, and the areas with a distance of less than 10 meters and comparable sizes were merged to generate the areas to be supplemented with data collection, and the centroid coordinates and minimum enclosing rectangle of each area were calculated.

[0068] Furthermore, the A* algorithm is used for flight path planning. The center point of the area to be supplemented is used as the waypoint. The horizontal and vertical safety distances are set to 5 meters, the search space grid size is 1 meter, the g(n) function is the flight distance, and the h(n) function is the Euclidean distance to the target point to generate the optimal flight path of the drone. The waypoint coordinates, speed and other parameters are encapsulated into instructions according to the MAVLink protocol and transmitted to the drone through the ground station. The drone is controlled to fly autonomously according to the planned path and to collect supplementary data for the fuzzy area. The flight speed is 1 meter / second, the altitude is 30 meters, the heading angle overlap is 80%, and the side overlap is 60%. The drone is equipped with a Velodyne VLP-16 laser radar and a Sony A7RII full-frame camera to obtain point cloud data with a resolution of 1 cm and image data with a resolution of 0.5 cm. The ICP algorithm is used to align the supplementary collected point cloud with the preliminary model, setting the maximum number of iterations to 100, the convergence error to 1 cm, the number of matching point pairs to 1000, and the density of the registered point cloud is increased to 500 points / square meter. The registered point cloud is meshed using the Screened Poisson surface reconstruction algorithm. The octree depth is set to 10 and the minimum sampling density is set to 5. This generates a noise-free, hole-free, detail-rich, high-precision 3D mesh model with 10 million triangles and a geometric error of less than 5 cm with the actual archaeological site.

[0069] Step 210, obtain and dynamically adjust the flight strategy based on the sound pressure distribution information and airflow field information in the archaeological site, and generate recommended specifications for the UAV that meet the dynamic adjustment results; the sound pressure distribution information is obtained based on the sound data in the archaeological site collected historically; the airflow field information is obtained based on the airflow data in the archaeological site collected historically; the dynamic adjustment includes adjusting the flight path, flight altitude and flight speed in the UAV's flight strategy; the recommended specifications include wing design, structural weight and sound-absorbing materials.

[0070] Specifically, when the drone flies for the first time, the ambient sound data inside the archaeological site can be collected through the onboard microphone array, and the airflow data inside the archaeological site can be collected through the porous probe and the air pressure sensor, and the airflow data includes airflow velocity and air pressure data.

[0071] In an exemplary embodiment, the steps of obtaining and dynamically adjusting the flight strategy based on the sound pressure distribution information and the airflow field information in the archaeological site, and generating the recommended specifications of the drone that meet the dynamic adjustment results include:

[0072] When the drone makes its first flight, the ambient sound data inside the archaeological site is collected through the drone’s onboard microphone array, and the airflow data inside the archaeological site is collected through the drone’s onboard porous probe and air pressure sensor; the airflow data includes airflow velocity and air pressure data;

[0073] Perform time domain analysis on environmental sound data, convert time domain signals into frequency domain signals through the fast Fourier transform algorithm, obtain sound pressure level spectra at different frequencies, and mark areas with higher sound pressure levels as acoustically sensitive areas to obtain sound pressure distribution information;

[0074] The airflow data is vector-field analyzed, and the airflow velocity, pressure, and air density of each area in the archaeological site are calculated through numerical simulation algorithms. The areas with turbulent airflow and high airflow velocity are identified, and the areas with turbulent airflow and high airflow velocity are marked as aerodynamic sensitive areas to obtain airflow field information.

[0075] Based on the distribution of acoustically sensitive areas and aerodynamically sensitive areas in archaeological sites, a flight strategy optimization model is trained through machine learning algorithms to predict the flight status of drones in archaeological sites with different flight parameters.

[0076] According to the prediction results of the flight strategy optimization model, the flight path in the flight strategy is dynamically adjusted, and the flight altitude and flight speed are reduced in the acoustically sensitive area, and the flight altitude and flight speed are increased in the aerodynamically sensitive area;

[0077] Based on the flight path after adjusting the flight strategy, the flight altitude and flight speed that need to be lowered, and the flight altitude and flight speed that need to be increased, the recommended specifications of the UAV are generated; the recommended specifications include wingspan, aspect ratio, airfoil selection, structural weight control, material selection and surface sound absorption structure.

[0078] Specifically, the collected environmental sound data was subjected to time-frequency analysis, and the time domain signal was converted into a frequency domain signal through the fast Fourier transform (FFT) algorithm. The sampling frequency was set to 44.1kHz, the window function was the Hamming window, and the spectrum resolution was 1Hz. The sound pressure level spectra at different frequencies were obtained, and the areas with sound pressure levels exceeding 70dB were identified and marked as acoustic sensitive areas. The airflow data was numerically simulated and analyzed, and CFD (Computational Fluid Dynamics) software such as ANSYS Fluent was used to establish a three-dimensional model of the archaeological site, set boundary conditions and turbulence models, and the grid size was 0.1 meters. The Reynolds-averaged Navier-Stokes equation was solved to obtain the spatial distribution of airflow velocity and pressure, and the areas with airflow velocities exceeding 5m / s or pressure gradients exceeding 10Pa / m were identified and marked as aerodynamic sensitive areas.

[0079] Furthermore, a flight strategy optimization model is constructed by integrating the distribution of acoustic sensitive areas and aerodynamic sensitive areas, as well as terrain undulation, obstacle density and other features. The support vector regression (SVR) algorithm is selected, and the input features include sound pressure level, airflow velocity, terrain undulation, etc., and the output features include flight altitude, speed, path curvature, etc. The radial basis kernel function is used to optimize the regularization parameters and kernel function parameters through grid search. The performance of the flight strategy optimization model is evaluated using 5-fold cross validation, and the average root mean square error (RMSE) is less than 1 meter and 0.5m / s. According to the prediction results of the flight strategy optimization model, the parameters of the second flight of the drone are dynamically adjusted.

[0080] For example, when the predicted flight altitude is less than 15 meters or the speed is less than 2m / s, the obstacle avoidance mechanism is triggered to automatically increase the flight altitude and speed; when the predicted flight altitude exceeds 40 meters or the speed exceeds 4m / s, the deceleration mechanism is triggered to automatically reduce the flight altitude and speed; in other cases, the flight parameters predicted by the flight strategy optimization model are maintained. Considering the power response time of the drone, a buffer zone is set when adjusting the parameters to smoothly transition to the target value.

[0081] Furthermore, the recommended specifications of the drone are automatically generated for the actual needs of the second flight. The aspect ratio of the wing is optimized according to the sound pressure level distribution, and a low Reynolds number airfoil such as SD7062 is selected, with an aspect ratio of 6, and the average lift-to-drag ratio is increased by 10%; the fuselage length is optimized according to the airflow velocity distribution, and a streamlined design is selected, the length is shortened by 15%, and the drag coefficient is reduced by 5%; composite materials are selected according to the ambient temperature and humidity conditions, and carbon fiber reinforced epoxy resin is used, with a volume fraction of 60% and a density of 1.8g / cm³, which increases the strength by 50% and the toughness by 20%; sound-absorbing materials are covered on the surface of the wing and fuselage, and microporous polyurethane foam is selected with a thickness of 5 mm and an acoustic absorption coefficient of 0.9. The optimized flight strategy parameters and drone specification parameters are uploaded to the ground control station, and the mission instructions for the second flight are generated and sent to the drone through a digital telex link. The ISM2.4GHz frequency band is selected, OFDM modulation and LDPC coding are used, the transmission rate is 10Mbps, the delay is less than 50 milliseconds, and the packet loss rate is less than 0.1%. A flight monitoring interface is set up at the ground control station to display the drone's position, speed, attitude and other status information in real time, as well as the progress and quality assessment results of data collection. When there is a deviation from the route or abnormal speed, an alarm is automatically triggered, and the operator determines whether manual intervention is required. When there is an emergency such as equipment failure or communication interruption, the return program is automatically triggered, and the drone returns to the take-off point autonomously. Through optimized flight strategies and drone specifications, a three-dimensional reconstruction result of the archaeological site with full coverage and an accuracy better than 5 cm is finally obtained.

[0082] During the first flight, the drone used a tetrahedron array of four microphones with a sampling frequency of 96kHz and a bit depth of 24bit to collect 10 minutes of sound data at an altitude of 20 meters above the ground. It also collected air velocity and air pressure data at a frequency of 10Hz through a four-hole probe and an air pressure sensor. MATLAB was used to perform FFT (Fast Fourier Transform) analysis on the sound data, with a Hamming window length of 1024, an overlap rate of 50%, and a spectral resolution of 93.75Hz. The 1 / 3 octave sound pressure level of the 20Hz-20kHz frequency band was extracted, and a sound pressure level distribution cloud map was drawn to identify three areas where the sound pressure level exceeded 80dB. OpenFOAM (Open Source Field Operation and Manipulation) was used to simulate the airflow data using CFD. The grid size was 0.05 meters. LES was selected as the turbulence model. The unsteady compressible Navier-Stokes equations were solved. The simulation time was 60 seconds. A horizontal section 5 meters above the ground was extracted. Velocity vectors and pressure contours were drawn. Two areas with velocities exceeding 10 m / s and pressure gradients exceeding 20 Pa / m were identified. Three features, namely, sound pressure level, airflow velocity, and terrain undulation, were selected as inputs to the flight strategy optimization model. Three parameters, namely, flight altitude, velocity, and path curvature, were selected as outputs. The training set and test set were constructed with a ratio of 8:2. Grid search was used to optimize the regularization parameter C and kernel function parameter gamma. The optimal parameters of the 5-fold cross validation were C=10 and gamma=0.1, with an average RMSE of 0.8 meters and 0.3 m / s.

[0083] Furthermore, according to the flight parameters predicted by the flight strategy optimization model, the adjustment thresholds of height and speed are set to 12 meters and 1.5m / s respectively, and the buffer interval is 20%, that is, the height buffer is 2.4 meters and the speed buffer is 0.3m / s. According to the environmental characteristics of the archaeological site, the drone is recommended to use a wing with a wingspan of 3 meters and an average chord length of 0.5 meters, with a lift-to-drag ratio of 15; the fuselage is 1.5 meters long and the front radius is 0.2 meters; the composite material uses T800-grade carbon fiber and epoxy resin, with a volume ratio of 1:1 and a unidirectional tensile strength of up to 2500MPa; the wing and fuselage are covered with a polyurethane foam sound-absorbing layer with a thickness of 8 mm and a noise reduction of 8dB. Using LoRa (Low-Rank Adaptation) communication in the 433MHz frequency band, the transmission power is 500mW, the receiving sensitivity is -148dBm, the transmission rate is 5kbps, the communication distance can reach 5 kilometers, the end-to-end delay is less than 1 second, and the bit error rate is less than 10-5. The ground station interface is refreshed once a second, showing the drone's 3D trajectory, speed curve, point cloud data volume, etc., and performing quality assessment. When the deviation from the route exceeds 10 meters, the speed exceeds 6m / s, and the attitude angle exceeds 30 degrees, a red alarm is issued to prompt the operator to intervene; when the key sensor is damaged and the data transmission link is interrupted for more than 10 seconds, the drone automatically returns, and the error of the forced landing point is less than 1 meter. The final 3D model of the archaeological site has a point cloud density of 1,000 points / square meter, an accuracy better than 2 cm, and a texture resolution of 1 mm / pixel.

[0084] Step 212, using the time series analysis prediction model to predict the risk area in the preliminary three-dimensional model, and re-optimize the flight path in the flight strategy to avoid the flight path in the flight strategy including the risk area; the risk area includes the natural collapse area and the area affected by the first flight of the drone.

[0085] In an exemplary embodiment, the steps of using a time series analysis prediction model to predict risk areas in a preliminary three-dimensional model and re-optimizing a flight path in a flight strategy include:

[0086] The damage information of the archaeological site was extracted from the preliminary three-dimensional model, and a time series data set of historical data on the damage status was constructed; the damage information included the location, area and degree of damage of the archaeological site; the data granularity of the time series data set was daily and the time span was nearly five years;

[0087] Performing a third preprocessing operation on the time series data set; the third preprocessing operation includes outlier detection, eliminating missing values ​​using a median filter method, and interpolating using a linear interpolation method;

[0088] Call the time series prediction model, fit the model parameters according to the historical data of the damage status, and optimize the model parameters through grid search and other methods to obtain the trained time series prediction model; the time series prediction model includes ARIMA and LSTM;

[0089] Apply the trained time series prediction model to the preliminary 3D model, predict the probability of damage risk in the next month for each grid, and divide each grid into high-risk area and low-risk area according to the preset probability threshold;

[0090] Perform spatial clustering analysis on each grid, connect the grids where the high-risk area is located into risk areas using a density clustering algorithm, and extract polygons formed by the boundaries of the risk areas; the density clustering algorithm includes the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm;

[0091] The polygon formed by the boundary of the risk area is combined with the flight path and flight altitude in the flight strategy, and the optimal flight path and flight altitude that avoid the risk area are planned through the A* path-finding algorithm; the optimal flight path and flight altitude replace the flight path and flight altitude in the flight strategy to obtain the second optimal flight strategy, and the UAV is instructed to fly through the second optimal flight strategy.

[0092] Specifically, the damage information of the archaeological site, including the location, area, and degree of damage, is extracted from the preliminary three-dimensional model to construct a time series data set of damage status, with a data granularity of days and a time span of nearly 5 years. The damage status time series data is subjected to a third preprocessing, including outlier detection and removal, missing value interpolation, etc., and median filtering and linear interpolation methods are used to ensure the integrity and continuity of the data. Suitable time series prediction models are selected, such as ARIMA (Autoregressive Integrated Moving Average Model), LSTM (Long Short-Term Memory), etc., and model parameters are fitted according to historical data of damage status. The model performance is evaluated by cross-validation and other methods within the preset parameter range through grid search or random search, and the parameter combination with the smallest prediction error is selected. Evaluation indicators can be root mean square error (RMSE), mean absolute percentage error (MAPE), etc. The trained time series prediction model is applied to the preliminary three-dimensional model of the archaeological site, and the probability of damage risk within the next month is predicted grid by grid.

[0093] Furthermore, according to the actual probability of damage events in historical data, the optimal threshold is found through methods such as ROC (Receiver Operating Characteristic) or PR curve (Precision-Recall) to strike a balance between risk and efficiency. It is possible to consider using multiple thresholds, such as 30%, 60%, and 90%, to divide the risk area into three levels: low, medium, and high, and formulate different management and control strategies. Spatial clustering analysis is performed on the predicted risk areas, and density clustering algorithms such as DBSCAN are used to merge spatially adjacent high-risk grids into connected risk areas, and the boundary polygons of the areas are extracted. The extracted risk area boundary polygons are combined with the drone's route planning algorithm, and the optimal route that avoids all risk areas is planned through the A* pathfinding algorithm. The Euclidean distance or Manhattan distance is used as a heuristic function to calculate the estimated cost from the current position of the drone to the target position. The cost function needs to comprehensively consider factors such as flight distance, energy consumption, and risk, and set different weight coefficients. At the same time, it is necessary to constrain the kinematic parameters of the drone, such as the maximum speed, acceleration, turning radius, etc., such as the maximum horizontal speed of 8m / s, the maximum climbing speed of 3m / s, and the minimum turning radius of 10 meters, to ensure that the generated route is executable. The waypoint coordinates, flight altitude, speed and other parameters of the optimized route are passed to the drone control system. When the drone performs the second supplementary data collection flight mission, the position and attitude information of the drone are monitored in real time, and the obstacle detection and tracking algorithm based on sensors such as vision or radar is used to build an environmental map within a range of 10 meters around the drone in real time, and the collision risk of the drone is calculated based on the position, speed and trajectory prediction of the obstacle. The safety distance threshold is set to 5 meters. When the distance between the drone and the obstacle is less than the safety distance, the obstacle avoidance control is triggered. According to the position and movement trend of the obstacle, the optimal obstacle avoidance direction and distance are calculated by the artificial potential field method, and a resultant force vector of radial repulsion and tangential gravity is generated. The drone is controlled to move along the direction of the vector, and the speed, heading and height of the drone are adjusted by the PID controller to achieve smooth obstacle avoidance until the distance to the obstacle is greater than the safety distance, leaving the danger zone and resuming normal route flight.

[0094] In one embodiment, the real-time position of the drone can also be compared with the boundary of the risk area. If the drone is about to enter the risk area, the same obstacle avoidance control logic is triggered to automatically adjust the flight altitude and direction of the drone to ensure that the drone is always outside the safe distance until it flies away from the risk area and resumes normal flight. This can minimize the damage risk of the drone while ensuring the completion of the data collection task and improve flight safety. The archaeological site damage information in the past 1825 days is extracted from the preliminary three-dimensional model, including the XYZ coordinates of the damage location, the damage area (square meters), the damage degree (0-1 scale), etc., to construct a time series data set. The Tukey algorithm is used to detect outliers of the damage degree, and outliers exceeding 3 times the interquartile range are eliminated. The missing values ​​are interpolated by the local weighted scatter point smoothing method. The SARIMA seasonal autoregressive integrated moving average model is selected, and the optimal time series prediction model order (p, d, q) (P, D, Q) m is determined to be (1, 1, 2) (0, 1, 1) 7 by the AIC information criterion, and the time series prediction model parameters are fitted. A 10-fold cross validation was performed on 10 damage degree time series, and the RMSE was 0.12. The time series prediction model was applied to 262,144 grids of the three-dimensional model to predict the probability of damage risk in the next 30 days. The optimal threshold was determined to be 0.75 through the ROC curve, and grids with damage risk > 0.75 were defined as high-risk areas. The OPTICS algorithm was used to cluster high-risk grids, with a radius parameter ε of 10 meters and a minimum number of points MinPts of 5, and 6 high-risk area polygons were extracted.

[0095] Furthermore, the A* algorithm is used to plan the route. The state space is the XYZ coordinates and heading angle of the drone. The transfer cost is the flight distance. The heuristic function is the Euclidean distance between the drone and the target point. The constraints are the drone flight speed ≤15m / s, acceleration ≤2m / s², turning radius ≥20m, and the minimum obstacle avoidance distance is 8m. The planned route is 6.5km long and the flight time is 8.6min. The drone is equipped with ZED2i binocular vision and LivoxMID-70 laser radar. It detects dynamic and static obstacles within the range of 10-30m at a frequency of 10Hz, and uses the VoxelNet end-to-end 3D target detection algorithm to generate an obstacle octree map in real time. When the obstacle is within 20m, the velocity potential field method is triggered to avoid obstacles. The resultant force vector weight coefficients Kv=1.2 and Kω=2.5 guide the drone to deviate 45° and decelerate by 50% until it is 5m away from the obstacle. The YOLOv5 model is used to detect risk area signs in the drone video stream in real time. When the IOU>0.85 and lasts for 3 seconds, the same obstacle avoidance logic is triggered, and the data collection task is finally completed, obtaining 2.3GB of three-dimensional data of the archaeological site.

[0096] Step 214, setting a parking area for the drone in the preliminary three-dimensional modeling, and forcing the drone to land in the parking area when the drone detects an abnormality; the detected abnormality includes insufficient power and loss of control signal.

[0097] In an exemplary embodiment, the steps of setting a parking area for the drone in the preliminary three-dimensional modeling and forcing the drone to land in the parking area when the drone detects an abnormality include:

[0098] Extract the elevation data, terrain contours and terrain textures from the preliminary 3D model, use the Kriging spatial interpolation algorithm to analyze the elevation data, and obtain the terrain undulation contour map of the archaeological site. Then, perform a buffer zone analysis based on the terrain undulation contour map, terrain contours and terrain textures to obtain areas with a distance greater than a preset distance from the buildings in the archaeological site, a slope less than a preset slope and a vegetation coverage lower than a preset percentage, and mark them as potential drone parking areas.

[0099] Among them: the preset distance can be 10 meters, the preset slope can be 5 degrees, and the preset percentage can be 30%.

[0100] The potential drone parking area is overlaid with the preliminary 3D model for analysis. The preliminary 3D model is segmented into polygonal areas of 100-500 square meters using the SLIC superpixel segmentation algorithm. The roughness of the polygonal area is determined based on the terrain texture in the preliminary 3D model using a feature matching method, and the area with a high degree of roughness is filtered out to obtain the parking area. The feature matching methods include gray-level co-occurrence matrix and local binary pattern.

[0101] According to the location distribution of the parking areas, the Thiessen polygon algorithm is used to obtain the parking area network, and the minimum spanning tree algorithm is used to construct the shortest flight path connecting all the parking areas to obtain the parking area topology structure for the UAV forced landing;

[0102] The lighting management system on the drone detects the voltage, current, temperature and other parameters of the battery pack on the drone in real time, and uses the Kalman filter algorithm to estimate the remaining available capacity of the battery pack. When the remaining available capacity is lower than the preset value, a low battery warning is triggered;

[0103] The link budget model is used to calculate the received power of the control signal on the drone. When the received power is lower than the sensitivity threshold for more than a preset time, a signal loss warning is triggered.

[0104] When the low battery warning or signal loss warning is triggered, the GPS coordinates of the current drone are obtained, and the nearest preset number of parking areas are obtained according to the parking area topology and GPS coordinates. The optimal obstacle avoidance path from the current GPS coordinates to each of the nearest preset number of parking areas is planned according to the A* heuristic search algorithm, and the optimal parking area is obtained in combination with the remaining available capacity of the battery pack;

[0105] Through the laser radar and onboard computer carried by the drone, a three-dimensional grid map within the preset range of the drone is constructed in real time to detect obstacles above the preset height; when the obstacle is located on the optimal obstacle avoidance path, the dynamic window method is used to quickly re-plan the local path within the preset time, adjust the yaw angle and pitch angle, and avoid the obstacle;

[0106] When the UAV reaches the optimal parking area, the ultrasonic radar on the UAV measures the current height of the UAV from the ground, and controls the descent speed of the UAV according to the current height of the UAV from the ground. The PID control algorithm is used to control the UAV to descend within the preset height range, and the power system of the UAV is shut down after the UAV touches the ground.

[0107] Specifically, in the preliminary three-dimensional model of the archaeological site, by analyzing factors such as the flatness of the terrain, vegetation coverage, and the safe distance from the building, the Kriging spatial interpolation algorithm was used, the Gaussian variogram model was selected, and the interpolation parameters were optimized through cross-validation to generate the terrain undulation contour map of the archaeological site area. Then, through buffer analysis, the minimum turning radius and climbing rate of the drone were comprehensively considered, and the areas with a distance greater than 15 meters from the building, a slope less than 3 degrees, and a vegetation coverage less than 20% were extracted and marked as potential drone parking areas. The potential parking areas were superimposed and analyzed with the orthophoto map of the archaeological site with a ground resolution of 0.15 meters. The SLIC superpixel segmentation algorithm was used to select a superpixel size of 30 pixels and a compactness parameter of 15, and the image was segmented into polygonal areas ranging from 200 to 1000 square meters. Through texture feature matching, including grayscale co-occurrence matrix, Gabor filtering, color histogram, etc., the areas with uneven surface material and high roughness were filtered out to obtain a refined parking area that meets the parking conditions. According to the location distribution of the parking areas, the Thiessen polygon algorithm is used to generate a parking area network covering the entire archaeological site area. The minimum spanning tree algorithm is used to construct the shortest flight path connecting all the parking areas, forming a parking area topology structure for emergency landing of drones.

[0108] Furthermore, when the UAV is performing flight missions at archaeological sites, the voltage, current, temperature and other parameters of the battery pack are monitored in real time through the onboard power management system, and the remaining available capacity of the battery is estimated using the Kalman filter algorithm. When the available capacity is less than 20%, a low battery warning is triggered; the received power of the control signal is calculated through the link budget model, and when the received power is less than -93dBm for more than 5 seconds, a signal loss warning is triggered; the attitude angle is estimated through IMU data filtering, and when the pitch angle or roll angle exceeds 35°, a loss of control warning is triggered; the current wind speed and direction are estimated through the onboard barometer and meteorological data, and when the wind speed exceeds 12m / s, a high wind warning is triggered. When any abnormal warning occurs, the drone control system immediately obtains the current GPS coordinates and flight status information, and finds the nearest five alternative parking areas by querying the parking area topological structure database. Based on the A* heuristic search algorithm, a weighted cost function that comprehensively considers flight distance, energy loss and obstacle density is designed, and the weight combination is optimized using a genetic algorithm to plan the optimal obstacle avoidance flight path from the current position to each alternative parking area, while switching to the backup power supply to ensure flight and control power. The drone begins to execute the optimal forced landing path, and through the laser radar and onboard computer, a three-dimensional grid map within a range of 15 meters is constructed in real time to detect obstacles above 0.3 meters. When an obstacle appears on the flight path, the dynamic window method is used to quickly re-plan the local path within 0.2 seconds, adjust the yaw angle and pitch angle, avoid obstacles, and keep the flight altitude and speed as stable as possible. While performing the forced landing mission, the drone uses the onboard RGB camera and infrared camera to take real-time ground images at a height of 30 meters. It uses the PSPNet (Pyramid Scene Parsing Network) semantic segmentation algorithm to identify different types of land objects such as grass, sand, and water, and evaluate their flatness, hardness, humidity and other characteristics. Combined with the terrain undulation and obstacle distribution, the location of the optimal forced landing point is dynamically updated to guide the drone to adjust its course and altitude in advance. When the drone flies above the optimal forced landing point finally selected, it uses ultrasonic radar to measure the current height from the ground, controls the descent speed, and uses the PID control algorithm to achieve a smooth landing within a range of 0.5 meters. After touching the ground, the motor is immediately turned off, the propeller is locked, and the precise GPS coordinates and track data are sent to the ground station through the LoRa communication module, waiting for manual retrieval to complete the entire emergency landing process.

[0109] In the three-dimensional model of the archaeological site, 1000 random sampling points were selected, and the semivariogram model of the undulation was established using the Kriging interpolation algorithm. Through 5-fold cross-validation, it was determined that the interpolation accuracy was highest when the search radius was 50 meters and the number of samples was 20. Then, with a step length of 30 meters, regular grid points covering the entire archaeological site area were generated, and neighborhood queries and interpolation estimates were performed on each grid point to obtain a 1:500 scale terrain undulation contour map. Then, with the building as the center, a 20-meter buffer zone was generated, and areas with a slope of less than 5 degrees and a vegetation index NDVI of less than 0.3 were extracted to obtain a total area of ​​1.5 square kilometers of potential parking areas. These areas were superimposed with orthophotos with a resolution of 0.1 meters, and the SLIC segmentation algorithm based on graph cuts was used. The region size parameter was set to 40 pixels and the compactness parameter was set to 20 to obtain 500 superpixel areas. For each area, 8 gray-level co-occurrence matrix features, 6 Gabor filter features and 3 HSV color histogram features were extracted, and the support vector machine (SVM) model was trained for material classification. The areas with a roughness greater than 0.8 were eliminated, and finally 250 refined parking areas were obtained. The Douglas-Peucker algorithm was used to simplify the area boundaries, and then the adjacency relationship graph was generated by the Thiessen polygon algorithm. The Kruskal (Kruskal-Wallis one-way analysis of variance by ranks) minimum spanning tree algorithm was used to construct the network topology structure of the forced landing path with the shortest total length, with the area center point as the node and the flight distance as the edge weight.

[0110] Furthermore, during the flight of the drone, the remaining battery capacity is estimated every 1 second through the Kalman filter. When the capacity drops below 30%, and when the received power is calculated by the link budget model to be lower than -90dBm for more than 3 seconds, the low battery and signal loss warnings are triggered respectively. When the IMU (inertial measurement unit) data calculates that the pitch angle or roll angle of the drone exceeds 40 degrees, and the barometer measures that the ambient wind speed is higher than 15m / s, the out-of-control and high wind warnings are triggered respectively. Once an abnormal warning occurs, the drone obtains the current GPS coordinates and searches the path network with the A* algorithm. The flight distance, power consumption and number of obstacles are assigned weights of 0.5, 0.3, and 0.2 respectively. The three paths with the lowest cost are selected, and the particle swarm algorithm is used to optimize the weights to obtain the optimal forced landing route. The forced landing begins, and the laser radar scans at a frequency of 10Hz to detect obstacles within 10 meters. When the obstacle is found to be less than 5 meters away, the artificial potential field method is used to re-plan the obstacle avoidance path with a yaw angle of no more than 30 degrees within 1 second, and the adaptive PID control algorithm is used to make the UAV's flight attitude smoothly transition within 3 seconds. At the same time, the UAV continuously shoots ground images at an altitude of 20 meters, and uses the PSPNet semantic segmentation network to detect 5 types of sites such as grassland and sand in real time. For each type of site, 100 areas are randomly sampled, and 8 features such as the elevation standard deviation and the mean echo intensity are calculated. The decision tree model is input to determine whether the forced landing conditions are met, and the best forced landing point is dynamically selected, guiding the UAV to adjust the heading and descent height 2 seconds in advance. Finally, ultrasonic waves are used to measure the distance at a frequency of 50Hz to control the descent speed and land smoothly. After landing, the longitude and latitude of the forced landing point and 10 seconds of track data are sent through the LoRa module to complete a safe forced landing process.

[0111] Step 216, obtain supplementary data through a second flight, align and register the preliminary three-dimensional model with the supplementary data to obtain a supplementary three-dimensional model, and identify and mark the archaeological site defects and key areas of historical value of the corresponding archaeological site in the supplementary three-dimensional model through a deep learning model; archaeological site defects include surface cracks and wall damage.

[0112] In an exemplary embodiment, the steps of registering and aligning the preliminary three-dimensional model with the supplementary data to obtain a supplementary three-dimensional model, and identifying and marking the archaeological site defects and key areas of historical value of the corresponding archaeological site in the supplementary three-dimensional model through a deep learning model include:

[0113] Obtain data representation of the supplementary data and the preliminary three-dimensional model under the same datum according to a unified coordinate system;

[0114] Performing a fourth preprocessing operation on the supplementary data; the fourth preprocessing operation includes removing noise from the supplementary data to obtain high-quality supplementary data;

[0115] By comparing the spatial resolution of the high-quality supplementary data and the preliminary three-dimensional model, the spatial resolution of the high-quality supplementary data is adjusted by using an interpolation processing method, and the spatial resolution of the high-quality supplementary data and the preliminary three-dimensional model are secondary corrected by using an image registration algorithm to obtain the corrected high-quality supplementary data; the image registration algorithm includes a feature-based registration method;

[0116] The convolutional neural network is used to perform feature learning training on the data set obtained based on the characteristics of the existing archaeological sites to obtain a trained deep learning model;

[0117] Identify archaeological site features at archaeological sites based on trained deep learning models and calibrated high-quality supplementary data;

[0118] By visualizing the identified archaeological site features, the archaeological site defects and key areas with historical value of the archaeological site are obtained, and the obtained archaeological site defects and key areas with historical value are marked in the supplementary three-dimensional model.

[0119] Specifically, the supplementary data obtained by the second flight of the UAV is synchronized with the preliminary 3D model using a unified coordinate system to obtain data representation of the two under the same reference. Assuming that the coordinate system of the preliminary 3D model is WGS84 (World Geodetic System 1984), and the supplementary data uses a local coordinate system, the coordinates of the supplementary data can be converted to WGS84 through coordinate conversion tools, such as the `ogr2ogr` command in GDAL (Geospatial Data Abstraction Library).

[0120] For example, the command is as follows: ogr2ogr -t_srsEPSG:4326 -s_srsEPSG:XXXXoutput.shpinput.shp, where XXXX is the EPSG code of the local coordinate system. By comparing the spatial resolution of the preliminary 3D model and the supplementary data, the data resolution is adjusted using an interpolation processing method so that the supplementary data can match the preliminary 3D model.

[0121] Furthermore, if the resolution of the preliminary 3D model is 1 meter and the resolution of the supplementary data is 5 meters, the resolution of the supplementary data can be adjusted to 1 meter using linear interpolation. This can be achieved using the “raster interpolation” tool in ArcGIS software. Evaluate the noise and integrity of the supplementary data, remove low-quality data, and obtain a high-quality data set for subsequent processing. Use statistical analysis methods to evaluate the standard deviation and mean of the data, and remove data points with abnormally high standard deviations, such as 3 times the standard deviation above the mean. Select obvious and stable feature points such as corners of buildings and landform features, and accurately align the preliminary 3D model and supplementary data using point cloud registration technology. Ten obvious corner points can be selected as control points, and the ICP algorithm can be used for point cloud registration, and the accuracy can be improved to within 1 meter. Use an appropriate image registration algorithm, such as a feature-based registration method, to calibrate the aligned data again to ensure the consistency of all data in space. The SIFT algorithm can be used to extract feature points and calculate the relative transformation matrix to further improve the registration accuracy to 0.5 meters.

[0122] Further, the corrected dataset is trained using a deep learning algorithm to annotate key features such as the edges of the archaeological site buildings, surface cracks, and wall damage. The U-Net convolutional neural network model can be used to process an archaeological site area of ​​1 square kilometer. The training process may require the use of a dataset containing 10,000 manually annotated samples. The convolutional neural network deep learning model is used to learn the features of the training data, thereby improving the model's ability to recognize the characteristic points of the archaeological site. After training, the model can achieve 90% accuracy and 85% recall, which means that the model can efficiently identify most of the key features. The trained deep learning model is applied to an independent test dataset to evaluate the performance of the model on actual data and judge the practicality of the model. On a test set containing 200 independent images, the model showed an overall accuracy of 88%. By visualizing the model annotation results, the location and morphology of the edges, cracks, and damaged areas of the archaeological site buildings are intuitively displayed, providing a reference for historical value assessment and protection. The model's prediction results can be overlaid on high-resolution satellite images using QGIS software, showing clear damaged areas and structural changes.

[0123] Step 218, delegate logistics tasks based on the annotated supplementary three-dimensional model, and generate new drone payloads and flight plans based on the logistics tasks; logistics tasks include setting up and monitoring sensitive areas, maintaining the building structure of the archaeological site, and repairing defects in the archaeological site; drone payloads and flight plans include transporting monitoring equipment, maintenance tools and repair materials by drone.

[0124] In an exemplary embodiment, the steps of generating a new drone payload and flight plan according to a logistics mission include:

[0125] Develop a load list for the drone based on the logistical mission and adjust it to the drone’s maximum payload; the load list includes monitoring equipment, maintenance tools, and materials for restoring archaeological sites;

[0126] The mass of the drone's payload compartment is designed according to the maximum payload, and then the items in the loading list are classified according to the frequency of use and importance, and the load layout in the payload compartment is determined based on the classification results; the payload compartment is a modular quick-detachable payload compartment, which is loaded, unloaded and replaced through an electric telescopic mechanism and a locking device, and is equipped with protective equipment, inventory and status detection equipment; the protective equipment includes partitions, brackets and fixing belts; the status detection equipment includes RFID tags and sensors; the RFID tags are used to conduct real-time inventory of items in the payload compartment, and the sensors are used to detect the status of items in the payload compartment;

[0127] Formulate a flight plan for monitoring archaeological sites, and optimize the flight path of the drone using a heuristic algorithm based on the preset time limit requirements of the flight plan, the flight performance of the drone, and the capacity of the drone's battery pack, so as to obtain a monitoring flight path that can cover the archaeological site with the shortest flight distance and the lowest energy consumption while ensuring detection coverage and data quality; and avoid obstacles based on the structures within the archaeological site shown in the supplementary three-dimensional model and the weather information obtained;

[0128] Through intelligent scheduling algorithms, the path of logistics tasks is inserted based on the monitored flight path, and the logistics task path and time window are optimized to minimize transportation costs and maximize task efficiency. Combined with the urgency and priority of materials, the flight speed and altitude of the drone are dynamically adjusted to ensure timely delivery of materials.

[0129] Using computer vision and machine learning algorithms, a maintenance task list is generated based on the defects of the archaeological sites marked in the supplementary three-dimensional model, and the degree of damage of the defects of the archaeological sites is evaluated. A maintenance task list is generated based on the degree of damage, and then the maintenance task list is matched with the logistics tasks to formulate a task execution plan, guide the drones to autonomously pick up, transport and deliver goods, and assist ground personnel to complete maintenance operations through operating mechanisms such as the robotic arms carried by the drones.

[0130] Specifically, according to the needs of the heritage monitoring mission, a loading list for the drone is drawn up, including monitoring equipment, including high-definition cameras, lidar, infrared thermal imagers, etc., maintenance tools including screwdrivers, pliers, multimeters, etc., and repair materials including waterproof coatings, adhesives, grouting materials, etc. The weight and volume of each item are calculated, and the loading plan is optimized to ensure that the total weight of the items does not exceed the maximum payload of the drone and that the center of gravity is reasonably positioned.

[0131] On the other hand, according to the power system and battery capacity of the drone, the maximum range and endurance are analyzed to ensure that the flight requirements of long time and heavy load are met. According to the properties and uses of the loaded items, a reasonable layout plan of the load compartment is designed, and the items are classified and combined according to the frequency of use and importance. Taking into account the environmental factors such as shockproof, waterproof and high temperature resistance during the flight, lightweight and high-strength materials including carbon fiber, aluminum-magnesium alloy, etc. are selected to make the load compartment, and protective measures such as partitions, brackets and fixing belts are set. A modular and quick-detachable load compartment is installed on the drone body. Through the electric telescopic mechanism and locking device, the load compartment can be quickly loaded and unloaded and replaced, reducing the time the drone stays on the ground. At the same time, RFID tags and sensors are set in the load compartment to realize real-time inventory and status monitoring of items, and the inventory information and status data are uploaded to the ground control station in real time through the wireless communication module. Taking into account the time requirements, flight performance, energy efficiency and other factors of the heritage monitoring task, a heuristic algorithm is used to optimize the route of the drone. Taking genetic algorithm as an example, the route is encoded as chromosomes, and indicators such as flight distance, energy consumption, and risk are used as fitness functions. The optimal route is evolved through operations such as selection, crossover, and mutation.

[0132] Furthermore, under the premise of ensuring monitoring coverage and data quality, the flight distance and energy consumption are minimized, and alternative routes and emergency plans are designed taking into account factors such as bad weather and obstacles. While performing monitoring tasks, the material needs of archaeological site managers are responded to in a timely manner, and material distribution tasks are inserted on the basis of existing routes through intelligent scheduling algorithms.

[0133] Specifically, an event-driven dynamic scheduling algorithm is used to adaptively adjust the routes and task sequences of drones according to the urgency and priority of material needs, optimize the distribution path and time window, and improve transportation efficiency. For large or sophisticated maintenance equipment, drones can work with ground personnel to complete the loading, transportation, and unloading of equipment. Computer vision and machine learning algorithms are used to analyze the archaeological site image data collected by drones in real time. Semantic segmentation and target detection algorithms based on deep learning, such as U-Net (Unity Networking), MaskR-CNN (Mask Region-based Convolutional Neural Network), and mask region-based convolutional neural networks are used to train segmentation and detection models for archaeological site facilities through transfer learning and fine-grained annotation. For newly collected images, the trained model is used for inference to obtain the category label of each pixel and the bounding box of each target, and calculate the attribute information such as the damaged area, location, and type of the facility.

[0134] Furthermore, the detected maintenance tasks are matched with material requirements, and the optimal task execution plan is formulated to guide the drone to carry out maintenance operations autonomously or semi-autonomously. A task management platform based on microservice architecture can be established to divide the functions of drone scheduling, route planning, data processing, etc. into independent microservices, and communicate and integrate through RESTful API (Representational State Transfer). Use SpringBoot to build microservice infrastructure, use Kubernetes (container orchestration platform) to implement containerized deployment and orchestration of microservices, use Kafka distributed messaging system to implement asynchronous communication between microservices, and use MongoDB to store unstructured data. At the same time, develop a user-friendly Web interface to create, submit, monitor and analyze tasks. Through the CI / CD process, continuous integration and deployment of microservices can be achieved to improve the maintainability and scalability of the system. The multi-drone collaborative operation mode can be adopted to improve the parallelism and robustness of task execution through task decomposition and dynamic allocation. Using deep reinforcement learning algorithms such as PPO (Proximal Policy Optimization) and MADDPG (Multi-Agent Deep Deterministic Policy Gradient), the strategy and route of the drone are adaptively adjusted according to historical flight data and environmental feedback, achieving multi-machine collaboration, intelligent obstacle avoidance and dynamic adaptation, and continuously improving the intelligence level and application effect of drones in heritage monitoring and logistics support. The drone is equipped with 10 high-definition cameras, 2 laser radars, 1 infrared thermal imager, 20 sets of maintenance tools and 50 kg of repair materials. The total weight is controlled within 150 kg. Through finite element analysis and load balance algorithm, the center of gravity position is optimized so that it deviates from the geometric center by no more than 5 cm. According to the battery capacity and energy consumption model of the drone, the maximum range under full load conditions is estimated to be 50 kilometers and the endurance time is 2 hours. The payload compartment is made of honeycomb sandwich carbon fiber composite material, with 3 independent cabins, the cabin wall thickness is 5 mm, the interior is filled with damping foam, and 5 acceleration sensors and temperature and humidity sensors are installed to monitor the load status in real time. The payload compartment is connected to the fuselage through 10 electric telescopic rods and a spiral locking mechanism, and the unlocking and ejecting time is less than 10 seconds. Each item is attached with an electronic tag, and 4 RFID antennas are set in the payload compartment. The inventory frequency is 1Hz, and the data is uploaded to the ground station through a 433MHz wireless module. The route planning adopts a multi-objective genetic algorithm with a chromosome length of 500, a crossover probability of 0.8, a mutation probability of 0.1, and 500 iterations. When convergence, 5 alternative routes are generated.Task scheduling uses an event-driven dynamic priority queue algorithm. The urgency of material needs is quantified as a priority of 0-10, and the compatibility with the drone's location and payload is represented as a binary vector of 0-1. Bit operations are used to quickly screen and sort, and the response time is less than 1 second.

[0135] Specifically, the drone can carry an onboard computer with an 8-core 3GHz CPU, 16GB memory, and a graphics card of NVIDIA GTX2080. It runs the MaskR-CNN semantic segmentation model trained with 1.2 million archaeological site images, and can process 4K resolution images at a speed of 25 frames per second, with an average pixel accuracy of 0.95 and a missed detection rate of 2%. Through binarization, connected domain analysis, polygon fitting and other algorithms, the vector contours and attribute information of the damaged area are extracted, and the information is associated with the material demand database to generate a maintenance task sequence. The management platform adopts a microservice architecture with front-end and back-end separation. The back-end uses the Spring Boot framework, is deployed based on Docker containers, and uses Kuber netes cluster orchestration. Services are decoupled through Kafka message queues. Each service has an independent CI / CD pipeline, and the average build time is 10 minutes. The front-end uses the Vue.js framework, adopts a responsive layout, and supports access from multiple terminals such as PCs, tablets, and mobile phones. When multiple UAVs work together, a multi-agent reinforcement learning algorithm based on graph neural networks is used. The state space includes 20 variables such as the UAV's position, speed, heading, and power. The action space includes 10 variables such as route points, task allocation, and obstacle avoidance strategies. The reward function comprehensively considers five indicators such as task completion rate, flight distance, and energy consumption. By combining strategy gradient and value iteration, the agents can learn stable and efficient collaborative strategies. Compared with fixed routes, the task completion time is shortened by 30% and energy consumption is reduced.

[0136] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0137] In an exemplary embodiment, the present application also provides a computer device, including a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps of a drone low-altitude monitoring method provided above in the present application are implemented.

[0138] In an exemplary embodiment, the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for low-altitude monitoring of a drone as provided above in the present application.

[0139] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0140] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0141] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for low-altitude monitoring of unmanned aerial vehicles, characterized in that: Applied to the excavation of archaeological sites, the method comprises: In response to the drone being launched, edge detection of the archaeological site is performed using a camera device carried by the drone, and the drone is instructed to perform a first flight according to a result of the edge detection; In the case where the drone performs the first flight, the archaeological site is scanned and a preliminary two-dimensional plane map of the archaeological site is constructed based on the scanning result; the preliminary two-dimensional plane map includes the terrain contour and terrain texture in the archaeological site; the terrain contour includes the contour lines of buildings, roads and walls; the terrain texture includes the roughness of ground stones, the arrangement of bricks and tiles, the looseness of soil and the texture of vegetation; the preliminary two-dimensional plane map is used to indicate the orientation and area of ​​the archaeological site; Acquire and construct a preliminary three-dimensional model based on the elevation data of the archaeological site and the preliminary two-dimensional plan; the elevation data includes the undulation of the terrain of the archaeological site, the height of the buildings of the archaeological site, and the three-dimensional structure information; divide the fuzzy area in the preliminary three-dimensional model according to the flight path of the first flight and the preset accuracy requirements, and generate a supplementary flight path passing through the fuzzy area, so as to formulate a flight strategy for instructing the UAV to perform a second flight according to the supplementary flight path; The steps of dividing a fuzzy area in the preliminary three-dimensional model according to the flight path of the first flight and the preset accuracy requirement, and generating a supplementary flight path passing through the fuzzy area, and formulating a flight strategy for instructing the drone to perform a second flight according to the supplementary flight path include: performing integrity analysis on the preliminary three-dimensional model through integrity indicators to determine whether there is a fuzzy area in the preliminary three-dimensional model; the integrity indicators include point cloud density and surface normal vector continuity; if so, dividing the archaeological site into a plurality of voxel grids, and evaluating the point cloud density and surface normal vector continuity in each voxel grid, and removing the voxel grids whose point cloud density or surface normal vector continuity does not meet the preset threshold. The grids are marked as fuzzy voxel grids; cluster analysis is performed on all the fuzzy voxel grids, and adjacent fuzzy voxel grids are merged to form a number of connected data collection areas to be supplemented, and the position and size of the data collection area to be supplemented are calculated according to the central coordinates and boundary range of each data collection area to be supplemented; according to the position and size of the data collection area to be supplemented, a supplementary flight path covering the data collection area to be supplemented is generated, and the flight altitude, flight speed, angle and sampling frequency of the UAV are set based on the supplementary flight path; the flight strategy of the second flight is obtained according to the supplementary flight path, the flight altitude, flight speed, angle and sampling frequency of the UAV; Acquiring and dynamically adjusting the flight strategy based on the sound pressure distribution information and airflow field information within the archaeological site, and generating recommended specifications of the UAV that meet the dynamic adjustment results; the dynamic adjustment includes adjusting the flight path, flight altitude and flight speed in the flight strategy of the UAV; The steps of obtaining and dynamically adjusting the flight strategy according to the sound pressure distribution information and airflow field information in the archaeological site, and generating recommended specifications of the UAV that meet the dynamic adjustment results include: when the UAV is performing the first flight, collecting the environmental sound data inside the archaeological site through the onboard microphone array carried by the UAV, and collecting the airflow data inside the archaeological site through the porous probe and air pressure sensor carried by the UAV; the airflow data includes airflow velocity and air pressure data; performing time domain analysis on the environmental sound data, converting the time domain signal into a frequency domain signal, obtaining sound pressure level spectra at different frequencies, and marking areas with higher sound pressure levels as acoustically sensitive areas to obtain the sound pressure distribution information; performing vector field analysis on the airflow data, calculating the airflow velocity, pressure and pressure of each area in the archaeological site and air density, and identify areas with turbulent airflow and high airflow speed, mark the areas with turbulent airflow and high airflow speed as aerodynamic sensitive areas to obtain the airflow field information; based on the distribution of the acoustic sensitive areas and the aerodynamic sensitive areas in the archaeological site, predict the flight state of the UAV in the archaeological site with different flight parameters; according to the prediction results, dynamically adjust the flight path in the flight strategy, reduce the flight altitude and flight speed in the acoustic sensitive areas, and increase the flight altitude and flight speed in the aerodynamic sensitive areas; based on the flight path adjusted by the flight strategy, the flight altitude and flight speed that need to be reduced, and the flight altitude and flight speed that need to be increased, generate the recommended specifications of the UAV; the recommended specifications include wing span, aspect ratio, airfoil selection, structural weight control, material selection and surface sound absorption structure; Using a time series analysis prediction model, predicting the risk area in the preliminary three-dimensional model, and re-optimizing the flight path in the flight strategy to avoid the flight path in the flight strategy including the risk area; the risk area includes a natural collapse area and an area affected by the first flight of the UAV; The step of using the time series analysis prediction model to predict the risk area in the preliminary three-dimensional model and re-optimizing the flight path in the flight strategy includes: extracting damage information of the archaeological site from the preliminary three-dimensional model, and constructing a time series data set of historical data on the damage status; the damage information includes the damage location, damage area and damage degree of the archaeological site; performing a third preprocessing operation on the time series data set; the third preprocessing operation includes outlier detection, eliminating missing values ​​using the median filtering method, and interpolating using the linear interpolation method; calling the time series prediction model, fitting model parameters according to the historical data of the damage status, and optimizing the model parameters to obtain a trained time series prediction model; the time series prediction model includes ARI MA and LSTM; applying the trained time series prediction model to the preliminary three-dimensional model, predicting the damage risk probability within the next month grid by grid, and dividing each grid into a high-risk area and a low-risk area according to a preset probability threshold; performing spatial clustering analysis on each grid, connecting the grids where the high-risk area is located as the risk area, and extracting the polygons formed by the boundaries of the risk area; combining the polygons formed by the boundaries of the risk area with the flight path and flight altitude in the flight strategy, and planning the optimal flight path and flight altitude to avoid the risk area; and replacing the flight path and flight altitude in the flight strategy with the optimal flight path and flight altitude to obtain a second optimal flight strategy, and instructing the UAV to fly through the second optimal flight strategy; Supplementary data is acquired through a second flight, and the preliminary three-dimensional model is registered and aligned with the supplementary data to obtain a supplementary three-dimensional model. The archaeological site defects and key areas of historical value of the archaeological site corresponding to the supplementary three-dimensional model are identified and marked through a deep learning model; the archaeological site defects include surface cracks and wall damage.

2. The method according to claim 1, characterized in that The step of performing edge detection on the archaeological site by using the camera equipment carried by the drone and instructing the drone to perform a first flight according to the result of the edge detection comprises: Performing edge detection on the archaeological site by the camera device to obtain archaeological site image data; Performing a first preprocessing operation on the archaeological site image data; the first preprocessing operation includes noise removal and image enhancement; Analyzing the archaeological site image data after the first preprocessing operation to extract the outer boundary contour of the archaeological site; Calculating a perpendicular bisector of the outer boundary of the archaeological site according to the outer boundary contour as a path reference for the UAV to enter the interior of the archaeological site for a first flight; The path reference of the first flight is transmitted to the drone to instruct the drone to start the first flight.

3. The method according to claim 1, characterized in that The step of scanning the archaeological site and constructing a preliminary two-dimensional plan view of the archaeological site based on the scanning result comprises: Performing a second preprocessing operation on the scanned result; the second preprocessing operation includes noise removal and image enhancement; The terrain contour in the scanned result after the second pre-processing operation and the terrain texture in the scanned result after the second pre-processing operation are identified and the preliminary two-dimensional plan view about the terrain contour and the terrain texture is constructed.

4. The method according to claim 2, characterized in that: The step of obtaining and constructing a preliminary three-dimensional model based on the elevation data of the archaeological site and the preliminary two-dimensional plane map comprises: In the case where the UAV performs the first flight, in response to the completion of the construction of the preliminary two-dimensional plan, a second flight path for collecting elevation data is formulated according to the orientation of the archaeological site indicated by the preliminary two-dimensional plan; When the UAV flies along the second flight path, a three-dimensional scan is performed on the surface of the archaeological site by a stereo measurement device to obtain point cloud data of a first precision, and a structured light scanning device is used to perform a three-dimensional scan on the surface of the archaeological site to obtain point cloud data of a second precision; the second precision is greater than the first precision; Registering and fusing the point cloud data of the first precision and the point cloud data of the second precision, and obtaining a complete three-dimensional point cloud model by position calibration in a unified coordinate system; Performing filtering, denoising and downsampling processing on the three-dimensional point cloud model, and extracting elevation data of the archaeological site; The terrain contour and terrain texture of the archaeological site in the preliminary two-dimensional plan view are mapped to the three-dimensional point cloud model, and the surface of the point cloud is reconstructed to obtain the preliminary three-dimensional model.

5. The method according to claim 1, characterized in that The method further comprises: Extracting the elevation data, the terrain contour and the terrain texture in the preliminary three-dimensional model, analyzing the elevation data using the Kriging spatial interpolation algorithm to obtain a terrain undulation contour map of the archaeological site, and then performing a buffer zone analysis based on the terrain undulation contour map, the terrain contour and the terrain texture, obtaining areas that are a preset distance from buildings in the archaeological site, have a slope less than a preset slope and a vegetation coverage less than a preset percentage and are marked as potential drone parking areas; The potential drone parking area is superimposed and analyzed with the preliminary three-dimensional model, and the superimposed preliminary three-dimensional model is divided into polygonal areas of 100-500 square meters. Then, based on the terrain texture in the preliminary three-dimensional model, a feature matching method is used to determine the roughness of the polygonal area, and the area with a higher degree of roughness is filtered out to obtain the parking area; According to the position distribution of the parking areas, a shortest flight path connecting all the parking areas is constructed to obtain a topological structure of the parking areas for forced landing of the UAV; The lighting management system carried by the drone detects the voltage, current and temperature of the battery pack on the drone in real time, estimates the remaining available capacity of the battery pack, and triggers a low battery warning when the remaining available capacity is lower than a preset value; Calculating the received power of the control signal on the drone, and triggering a signal loss warning when the received power is lower than a sensitivity threshold for more than a preset time; When the low battery warning or signal loss warning is triggered, the current GPS coordinates of the drone are obtained, and according to the parking area topology and the GPS coordinates, the nearest preset number of parking areas are obtained, and the optimal obstacle avoidance path from the current GPS coordinates to each of the nearest preset number of parking areas is planned, and the optimal parking area is obtained in combination with the remaining available capacity of the battery pack; By using the laser radar and onboard computer carried by the drone, a three-dimensional grid map within a preset range of the drone is constructed in real time to detect obstacles above a preset height; when the obstacle is located on the optimal obstacle avoidance path, the local path is quickly replanned within a preset time, and the yaw angle and pitch angle are adjusted to avoid the obstacle; When the drone reaches the optimal parking area, the ultrasonic radar carried by the drone measures the current height of the drone from the ground, and the descent speed of the drone is controlled according to the current height of the drone from the ground, so as to control the drone to descend within a preset height range.

6. The method according to claim 1, characterized in that The steps of registering and aligning the preliminary three-dimensional model with the supplementary data to obtain a supplementary three-dimensional model, and identifying and marking the archaeological site defects and key areas of historical value of the archaeological site corresponding to the supplementary three-dimensional model through a deep learning model include: Obtaining data representation of the supplementary data and the preliminary three-dimensional model under the same reference according to a unified coordinate system; Performing a fourth preprocessing operation on the supplementary data; the fourth preprocessing operation includes removing noise from the supplementary data to obtain high-quality supplementary data; By comparing the spatial resolution of the high-quality supplementary data with that of the preliminary three-dimensional model, adjusting the spatial resolution of the high-quality supplementary data, and performing secondary correction on the spatial resolution of the high-quality supplementary data and the preliminary three-dimensional model, corrected high-quality supplementary data is obtained; performing feature learning training on a data set obtained according to the characteristics of the existing archaeological site through a convolutional neural network, and obtaining a trained deep learning model; identifying archaeological site features in the archaeological site according to the trained deep learning model and based on the corrected high-quality supplementary data; By visualizing the identified features of the archaeological site, the archaeological site defects and key areas with historical value of the archaeological site are obtained, and the obtained archaeological site defects and key areas with historical value are marked in the supplementary three-dimensional model.

7. The method according to claim 1, characterized in that The method further comprises: Developing a loading list for the drone based on the logistical mission and adjusting the loading list based on the maximum payload of the drone; the loading list includes monitoring equipment, maintenance tools and archaeological site restoration materials; The mass of the payload compartment of the UAV is designed according to the maximum payload, and then the items in the loading list are classified according to the frequency of use and importance, and the load layout in the payload compartment is determined according to the classification result; the payload compartment is a modular quick-detachable payload compartment, which is loaded, unloaded and replaced by an electric telescopic mechanism and a locking device, and is internally provided with protection equipment, inventory and status detection equipment; the status detection equipment includes an RFID tag and a sensor; the RFID tag is used to perform a real-time inventory of the items in the payload compartment, and the sensor is used to perform status detection on the items in the payload compartment; Formulate a flight plan for monitoring the archaeological site, optimize the route of the drone according to the preset time limit requirements of the flight plan, the flight performance of the drone and the capacity of the battery pack of the drone, and obtain a monitoring flight path that can cover the archaeological site with the shortest flight distance and the lowest energy consumption while ensuring detection coverage and data quality; and avoid obstacles according to the structure in the archaeological site shown in the supplementary three-dimensional model and the obtained weather information; Insert the path of the logistics task on the basis of the monitoring flight path, and optimize the logistics task path and time window, with the goal of minimizing transportation costs and maximizing task efficiency. Combined with the urgency and priority of the materials, dynamically adjust the flight speed and altitude of the drone to ensure timely delivery of the materials. Using computer vision and machine learning algorithms, a maintenance task list is generated based on the defects of the archaeological site marked in the supplementary three-dimensional model, and the degree of damage of the defects of the archaeological site is evaluated. A maintenance task list is generated based on the degree of damage, and then the maintenance task list is matched with the logistics tasks to develop a task execution plan to guide the drone to autonomously pick up, transport and deliver goods.

Images