Earthwork volume automatic calculation method and system based on unmanned aerial vehicle image

Abstract

The invention discloses an automatic earthwork volume calculation method and system based on an unmanned aerial vehicle image, and relates to the technical field of surveying and mapping and remote sensing, and the method comprises the steps: controlling an unmanned aerial vehicle platform to carry out the automatic aerial photographing, and obtaining a multi-view sequence image and pose data; preprocessing the image and reconstructing an initial three-dimensional point cloud model; dynamically calibrating the three-dimensional point cloud model and an engineering coordinate system; non-surface interference objects are intelligently classified and removed; generating a high-precision digital elevation model according to the terrain complexity; and the total filling amount and the total excavation amount of the construction area are calculated. The system comprises an unmanned aerial vehicle image acquisition module, an image preprocessing and three-dimensional reconstruction module, a coordinate dynamic calibration module, a non-earth surface interferent intelligent identification and elimination module, a self-adaptive digital elevation model generation module, an earthwork amount calculation module and a system control and data integration platform. The centimeter-level registration precision can be realized, the engineering quantity calculation accuracy is improved, and the calculation precision and efficiency are balanced.

Description

Technical Field

[0001] This invention relates to the field of surveying and remote sensing technology, and in particular to a method and system for automatically calculating earthwork quantities based on UAV imagery. Background Technology

[0002] With the accelerated pace of infrastructure construction and urbanization, earthwork engineering, as a crucial component of road, bridge, building, and water conservancy projects, directly impacts project cost control, construction schedule scheduling, and resource allocation decisions through the accuracy and efficiency of its quantity calculations. The core of earthwork quantity calculation lies in the precise modeling of terrain changes and volume difference analysis within the construction area. Traditional methods primarily rely on ground surveying equipment such as total stations and levels for manual point placement and elevation data collection, followed by volume estimation using triangulation or grid methods. These methods are not only time-consuming and labor-intensive but also limited by terrain complexity and visibility conditions, making them difficult to implement in large-scale or high-risk work areas. This results in incomplete data coverage and significant accuracy fluctuations, failing to meet the high timeliness and reliability requirements of modern engineering. Furthermore, with the expansion of project scale and the acceleration of construction pace, traditional methods are severely inadequate in terms of data update frequency and dynamic monitoring capabilities, unable to support the needs of refined construction management and real-time progress feedback.

[0003] Among these advancements, 3D terrain reconstruction technology based on UAV remote sensing imagery offers a new technical approach for calculating earthwork volumes. This technology utilizes UAVs equipped with high-resolution cameras to conduct low-altitude aerial photography of the construction area, acquiring multi-view image data. It then uses photogrammetry principles to generate high-precision digital surface models or digital elevation models, and subsequently calculates the cut and fill volumes through model differentiation. Compared to traditional ground surveying, UAV imagery technology offers significant advantages such as operational flexibility, wide coverage, high data density, and strong security, and has already seen initial applications in some engineering scenarios. However, existing earthwork calculation methods based on UAV imagery still face several key technical bottlenecks, hindering their large-scale promotion and automated deployment in complex engineering environments.

[0004] Existing technologies reveal multiple systemic flaws in practical applications: First, the lack of a high-precision dynamic coupling mechanism between image acquisition and the engineering coordinate system leads to systematic offsets in the generated 3D models, requiring post-processing correction using numerous ground control points, significantly increasing fieldwork and diminishing the advantages of automation. Second, terrain modeling algorithms are sensitive to non-surface disturbances such as vegetation, temporary material stockpiles, and construction machinery, making it difficult to effectively distinguish between the real surface and temporary coverings, resulting in distorted model elevations and consequently, deviations in quantity calculations. Third, volume calculations typically employ fixed-resolution grid differencing, failing to consider variations in local terrain complexity, leading to insufficient accuracy in critical areas such as slopes and gullies, while generating redundant calculations in flat areas. Finally, the entire calculation process lacks end-to-end system integration; from image acquisition and model reconstruction to quantity output, multiple independent software platforms must be manually connected, resulting in low data flow efficiency and susceptibility to human error. Summary of the Invention

[0005] The purpose of this invention is to provide an automatic calculation method and system for earthwork volume based on UAV imagery, in order to solve the problems of low registration accuracy between the 3D model and the engineering coordinate system, large influence of non-surface interference objects, and low calculation efficiency in the prior art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] An automatic calculation method and system for earthwork volume based on UAV imagery includes:

[0008] The UAV image acquisition module is used to execute automatic flight along a preset route and acquire multi-view high-resolution image sequences of the construction area. At the same time, it records the instantaneous position and attitude parameters of each image through an integrated high-precision GNSS receiver and IMU unit.

[0009] The image preprocessing and 3D reconstruction module receives image data and pose parameters output by the UAV image acquisition module, performs image distortion correction and radiometric correction, and generates an initial 3D point cloud of the construction area based on the structure-recovery-motion algorithm and dense matching technology.

[0010] The coordinate dynamic calibration module couples the initial 3D point cloud with the engineering coordinate system. By constructing a coordinate transformation model based on the least squares principle, it uses a small number of ground control points or known engineering benchmarks to perform parameter calculations, thereby achieving high-precision automatic registration between the 3D model and the engineering coordinate system.

[0011] The non-surface disturbance intelligent identification and removal module loads 3D point cloud data after dynamic coordinate calibration, applies a point cloud classification algorithm based on random forest and 3D geometric features, automatically identifies and filters out point cloud clusters corresponding to vegetation, temporary stockpiles and construction machinery, and outputs a clean surface point cloud.

[0012] The adaptive digital elevation model generation module receives the clean surface point cloud, dynamically calculates the optimal grid resolution based on the local point cloud density and terrain undulation, and generates a high-fidelity digital elevation model using an improved inverse distance weighted interpolation algorithm; the earthwork volume calculation module imports the design elevation surface data and the digital elevation model output by the adaptive digital elevation model generation module, performs grid difference operation, and calculates the total volume based on the cut and fill height and area of ​​each grid cell;

[0013] The system control and data integration platform, as the central processing unit, coordinates and controls the timing execution and data flow of all the above modules, and provides a graphical human-computer interaction interface for parameter configuration, task scheduling and result visualization.

[0014] Preferably, the UAV image acquisition module includes a flight control submodule, an image sensing submodule, and a positioning and attitude determination submodule. The flight control submodule generates an optimal flight path based on preset flight altitude, heading overlap, and lateral overlap, and drives the UAV platform to fly autonomously along this path. The image sensing submodule is mounted on the UAV gimbal and uses a global shutter CMOS sensor with a pixel size of 2.4μm, achieving a ground sampling distance better than 2cm at a flight altitude of 100m. The positioning and attitude determination submodule integrates a high-frequency GNSS receiver and a tactical-grade IMU. GNSS positioning accuracy in RTK / PPK mode reaches 1cm+1ppm horizontally and 2cm+1ppm vertically, while the IMU attitude measurement accuracy is better than 0.05°.

[0015] Furthermore, the coordinate transformation model executed by the coordinate dynamic calibration module is a seven-parameter Bursa model, the mathematical expression of which is: ; in, For translation parameter vectors, As a scale factor, For the rotation angle The module constructs an orthogonal rotation matrix. Using at least three evenly distributed ground control points, it employs a least-squares adjustment algorithm to solve for the seven transformation parameters, achieving accurate transformation from model coordinates to engineering coordinates with residual error controlled within 2cm.

[0016] Furthermore, the non-surface disturbance intelligent identification and removal module employs features including the point cloud's normal vector, local curvature, elevation variance, and nearest neighbor distribution statistics based on a KD-tree. During the training phase, the random forest classifier uses a labeled point cloud dataset containing four categories: vegetation, material stockpiles, machinery, and surface. Each decision tree splits nodes based on Gini impurity, and the final category assignment for each point is determined through voting among multiple trees. For point cloud clusters classified as non-surface disturbances, this module performs spatial clustering analysis and removes them after secondary verification based on their geometric contours and size features.

[0017] Preferably, in the adaptive digital elevation model generation module, the optimal grid resolution is determined based on the local terrain complexity index. The calculation formula is as follows: ;

[0018] in, To analyze the standard deviation of elevation of point clouds within the window, This represents the number of valid point clouds within the window. This module has a preset complexity threshold. ,when When a higher resolution (e.g., 0.1m) is used for mesh generation, then... At this time, a lower resolution (e.g., 0.5m) is used for mesh generation. In the improved inverse distance weighted interpolation algorithm, the weight calculation function incorporates an elevation reliability factor, the expression of which is: ,in For the point to be interpolated to the th The horizontal distance between sample points It is an exponential parameter (usually taken as 2). For the point to be interpolated and the first The normalized ratio of the elevation difference of each sample point to the local terrain slope.

[0019] On the other hand, an automatic calculation method for earthwork volume based on UAV imagery is described, with the following specific steps:

[0020] Step S110: Control the UAV platform equipped with high-precision GNSS / IMU and image sensor to automatically take aerial photos of the target construction area according to the preset aerial photography parameters, and simultaneously acquire multi-view sequence images and position and attitude data corresponding to each image;

[0021] Step S120: Preprocess the acquired sequence images, including optical distortion correction and radiometric consistency adjustment, and reconstruct the initial three-dimensional point cloud model of the construction area based on motion recovery structure and multi-view stereo vision algorithm.

[0022] Step S130: Dynamically calibrate the coordinate system of the initial 3D point cloud model with the engineering coordinate system. By establishing a seven-parameter coordinate transformation model and solving the transformation parameters using known ground control points, accurate registration between the 3D model and the engineering coordinate system is achieved.

[0023] Step S140: Perform intelligent classification on the registered 3D point cloud, extract and remove non-surface disturbance point clouds belonging to vegetation, temporary stockpiles and construction machinery, and obtain a clean point cloud dataset representing the real surface.

[0024] Step S150: Based on the clean point cloud dataset, dynamically determine the grid division resolution according to the local terrain complexity, and use the inverse distance weight interpolation algorithm that integrates elevation reliability factors to generate a high-precision digital elevation model.

[0025] Step S160: Perform grid cell differential calculation between the generated digital elevation model and the design elevation surface model. Based on the fill and cut height and area of ​​each cell, accumulate to obtain the total fill volume and total cut volume of the construction area.

[0026] In step S170, the integrated system platform performs task scheduling, process control and data management for steps S110 to S160 above, and outputs the final earthwork volume calculation result report and visualization charts.

[0027] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0028] This invention significantly reduces the systematic offset between the 3D model and the engineering coordinate system by integrating a high-precision GNSS / IMU and a dynamic coordinate calibration model, improving the registration accuracy to the centimeter level, reducing the reliance on a large number of ground control points, and improving the efficiency of field operations.

[0029] This invention employs point cloud intelligent classification technology based on random forest and three-dimensional geometric features, which can effectively identify and filter out non-surface interference, ensuring the purity of the digital elevation model data source and thus significantly improving the accuracy of engineering quantity calculations.

[0030] This invention introduces an adaptive mesh generation mechanism based on terrain complexity, using high resolution in key terrain areas to ensure accuracy and low resolution in flat areas to improve computational efficiency, thus achieving the best balance between computational accuracy and efficiency.

[0031] This invention constructs an end-to-end automated process from image acquisition, preprocessing, 3D reconstruction, coordinate calibration, interference removal, model generation to engineering quantity calculation, eliminating human error introduced by manual operation on multiple platforms, and realizing full automation and high reliability of earthwork engineering quantity calculation. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the overall technical architecture of the automatic earthwork volume calculation system based on UAV imagery proposed in this invention.

[0033] Figure 2 This is a schematic diagram of the core principle framework for intelligent identification and removal of non-surface interference objects in this invention. Detailed Implementation

[0034] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the present invention and not to limit the present invention. For those skilled in the art, the present invention can be practiced without some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present invention by illustrating examples of the invention.

[0035] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0036] In the embodiments of the present invention, the same reference numerals denote the same components, and for the sake of brevity, detailed descriptions of the same components are omitted in different embodiments. It should be understood that the thickness, length, width, and other dimensions of various components in the embodiments of the present invention shown in the accompanying drawings, as well as the overall thickness, length, width, and other dimensions of the integrated device, are merely illustrative and should not constitute any limitation on the present invention; the term "multiple" in the present invention refers to two or more (including two).

[0037] Example 1

[0038] In large-scale earthwork construction scenarios, such as highway subgrade filling projects, an automated earthwork volume calculation system based on UAV imagery begins to execute the complete work process. (See also...) Figure 1This system comprises a UAV image acquisition module, an image preprocessing and 3D reconstruction module, a coordinate dynamic calibration module, a non-surface interference intelligent identification and removal module, an adaptive digital elevation model generation module, an earthwork quantity calculation module, and a system control and data integration platform. The system control and data integration platform, acting as the central processing unit, first sends aerial survey commands to the UAV image acquisition module. Based on preset aerial survey parameters, including a flight altitude of 120 meters, a forward overlap of 80%, and a lateral overlap of 70%, the UAV image acquisition module plans the optimal flight path covering the entire construction area through the flight control submodule. The positioning and attitude determination submodule activates the high-frequency GNSS receiver and tactical-level IMU unit to acquire centimeter-level position information and attitude data better than 0.05° in real time in RTK mode. The image sensing submodule uses a global shutter CMOS sensor with a pixel size of 2.4μm, achieving a ground sampling distance of 1.8cm at a specified flight altitude, ensuring sufficient geometric accuracy and texture detail in the acquired image sequence.

[0039] During autonomous flight along a predetermined route, the image sensing submodule continuously acquires high-resolution images from multiple perspectives at a rate of 2 frames per second. Each image simultaneously records precise timestamps, latitude and longitude coordinates, elevation values, roll angle, pitch angle, and yaw angle parameters. All image data and pose parameters are transmitted back to the ground station in real time via a wireless data transmission link, where they are received and processed by the image preprocessing and 3D reconstruction module. This module first performs optical distortion correction on the original images, eliminating radial and tangential distortion based on camera calibration parameters. Then, radiometric correction is performed, eliminating the effects of uneven illumination through a histogram equalization algorithm. Finally, based on the structure-of-motion (SOMO) algorithm, scale-invariant feature transformation (SEM) points are extracted from each image, and the camera trajectory and sparse 3D point cloud are recovered through multi-view geometric constraints. Next, a semi-global matching algorithm is used for dense matching, generating an initial 3D point cloud model containing millions of 3D coordinate points, with a point cloud density of 200 points per square meter.

[0040] After loading the initial 3D point cloud, the coordinate dynamic calibration module reads the engineering coordinate system reference parameters. This module constructs a seven-parameter Bursa coordinate transformation model, the mathematical expression of which is: ;

[0041] The module automatically identifies four ground control points distributed throughout the construction area. These control points use forced centering markers, and their engineering coordinate system coordinates are obtained through static GNSS measurements, achieving an accuracy of 3mm horizontally and 5mm vertically. Seven transformation parameters are solved using a least-squares adjustment algorithm, including three translation parameters, three rotation parameters, and one scale factor. After the calculation is complete, the module transforms the initial 3D point cloud to the engineering coordinate system, with the transformation residual error controlled within 1.5cm, meeting the accuracy requirements for earthwork engineering calculations.

[0042] After receiving the coordinate-calibrated 3D point cloud data, the non-surface interference intelligent identification and removal module initiates the intelligent filtering process. (See also...) Figure 2 This module first constructs a KD-tree spatial index for the point cloud, calculating 3D geometric features such as normal vector, local curvature, and elevation variance for each point. A pre-trained model, trained on 20,000 labeled point samples, is loaded into a random forest classifier, encompassing four categories: vegetation, temporary material stockpiles, construction machinery, and ground surface. The classifier extracts features and makes classification decisions for each point. For point cloud clusters identified as non-ground disturbances, the module performs Euclidean clustering analysis to remove small-scale noise points with fewer than 50 points. Simultaneously, large-scale point cloud clusters undergo convex hull detection and contour analysis to ensure only true ground surface point clouds are retained. After processing, a clean ground surface point cloud dataset is output, with a vegetation removal rate exceeding 95% and a construction machinery identification accuracy exceeding 98%.

[0043] After receiving clean surface point clouds, the adaptive digital elevation model generation module initiates a dynamic mesh generation algorithm. The module divides the construction area into 100m × 100m analysis windows and calculates the local terrain complexity index for each window. The calculation formula is as follows: ;

[0044] in, To analyze the standard deviation of elevation of point clouds within the window, The number of valid point clouds within the window; the module has a preset complexity threshold. =2.5, when When >2.5, a 0.1m high-resolution grid is used. For distances ≤2.5, a 0.5m standard resolution grid is used. During the grid interpolation stage, an improved inverse distance weighted interpolation algorithm is employed. The weight calculation function incorporates an elevation reliability factor, the expression of which is: ,in p represents the horizontal distance from the point to be interpolated to the j-th sample point, with a value of 2. For the point to be interpolated and the first The normalized ratio of the elevation difference of each sample point to the local terrain slope. Through this adaptive mechanism, high accuracy is maintained in sloping areas with complex terrain, while computational efficiency is improved in flat fill areas.

[0045] The earthwork volume calculation module imports design elevation surface data and the generated digital elevation model. The design elevation surface is stored in a grid format, with the grid size consistent with the digital elevation model. The module performs grid cell difference operations. For each 5m × 5m grid cell, it calculates the difference between the current elevation and the design elevation; positive values ​​represent fill height, and negative values ​​represent cut height. The fill and cut volume calculation uses the prism formula: the earthwork volume of a single grid cell is V = (H1 + H2 + H3 + H4) / 4 × A, where H1~H4 are the fill and cut heights at the four corner points of the grid, and A is the grid area. The total fill and cut volumes are obtained by summing the earthwork volumes of all grid cells, with the results accurate to cubic meters.

[0046] The system control and data integration platform coordinates the sequential execution of the entire process, displaying work progress, processing status, and intermediate results in a graphical human-computer interface. The platform generates a final earthwork volume calculation report, including cut and fill volume statistics, earthwork balance diagrams, and heat maps of volume distribution, among other visual results. The entire process is completed within 2 hours, achieving more than 10 times the efficiency of traditional manual measurement methods, with a calculation accuracy exceeding 97%.

[0047] Example 2

[0048] In the context of earthwork measurement in mining operations, an automatic earthwork volume calculation system based on UAV imagery is optimized for specific terrain conditions. The UAV image acquisition module utilizes a heavy-duty hexacopter platform equipped with a full-frame aerial survey camera, achieving a ground sampling distance of 2.5cm at a flight altitude of 150 meters. The flight control submodule automatically generates variable-altitude flight paths based on the complex terrain of the mining area, lowering the flight altitude to 80 meters in steep slope areas to ensure point cloud density, and raising it to 200 meters in flat areas to improve operational efficiency.

[0049] To address the significant dust interference present in mining areas, the image preprocessing and 3D reconstruction module employs multispectral image fusion technology, enhancing surface feature identification through near-infrared bands. The motion reconstruction algorithm incorporates an inertial measurement unit (IMU) for initialization, effectively resolving the matching difficulties in areas lacking texture in the mining region. The generated 3D point cloud density reaches 150 points per square meter, with the density in key areas increased to 300 points per square meter.

[0050] The coordinate dynamic calibration module deployed six ground control points in the mining area, using specialized mining targets. These control points covered the bottom of the pit, slope platforms, and surface areas. Elevation anomaly correction was incorporated into the seven-parameter Bursa model solution to ensure coordinate transformation accuracy under significant elevation differences. The final coordinate registration accuracy reached 2cm horizontally and 3cm vertically, meeting the accuracy requirements for mine earthwork measurement.

[0051] The intelligent identification and removal module for non-surface disturbances targets disturbances unique to mining areas, such as ore piles and mining equipment. It adds ore texture and geometric features to the random forest classifier. The module employs a multi-scale feature extraction strategy, identifying individual ore point clouds at a small scale and the outlines of large mining equipment at a large scale. After classification and filtering, the purity of the surface point cloud in the mining area reaches 96%, providing a reliable data foundation for subsequent engineering quantity calculations.

[0052] The adaptive digital elevation model (DEM) generation module introduces anisotropy indices into the calculation of terrain complexity in mining areas, calculating the terrain undulations in the east-west and north-south directions respectively. The dynamic range of grid resolution is expanded to 0.05m~1.0m, using an ultra-high resolution of 0.05m at the mining face and a standard resolution of 1.0m in abandoned mining areas. In the improved inverse distance weighted interpolation algorithm, the elevation reliability factor considers the elevation accuracy decay law of the mining area point cloud, assigning higher weights to near points in the deep areas of the mining pit.

[0053] The earthwork volume calculation module is tailored to the characteristics of mining operations, adding specialized measurement functions such as stripping volume, extraction volume, and spoil volume. The module supports multi-period data comparison and analysis, automatically calculating the total amount of ore and rock stripping within a specified time period. Calculation results show that the system achieves 95% accuracy in earthwork volume calculation in mining areas, improving operational efficiency by 8 times compared to traditional measurement methods, providing timely and accurate volume data for mine production management.

[0054] The system control and data integration platform deploys a professional version of the software in the mining area, adding extended functions such as blasting warning zone identification and slope stability monitoring. The platform interfaces with the mine production management system, and the earthwork volume calculation results are directly imported into the production reporting system, realizing automated statistics and management of mining volume.

[0055] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. An automatic earthwork volume calculation method based on unmanned aerial vehicle images, characterized in that, The method comprises the following steps: Step S110, controlling a UAV platform carrying a high-precision GNSS / IMU and an image sensor to automatically take photos of a target construction area according to preset aerial photography parameters, and synchronously acquiring a multi-view sequence of images and position and attitude data corresponding to each image; Step S120, preprocessing the acquired sequence of images, including optical distortion correction and radiation consistency adjustment, and reconstructing an initial three-dimensional point cloud model of the construction area based on a motion recovery structure and a multi-view stereo vision algorithm; Step S130, dynamically calibrating a coordinate system of the initial three-dimensional point cloud model and an engineering coordinate system, establishing a seven-parameter coordinate conversion model, and solving conversion parameters by using known ground control points; Step S140, intelligently classifying the registered three-dimensional point cloud, extracting and removing non-ground interference point cloud belonging to vegetation, temporary stockpiles and construction machinery, and obtaining a pure point cloud data set representing a real ground surface; Step S150, dynamically determining a grid division resolution according to local terrain complexity based on the pure point cloud data set, and generating a high-precision digital elevation model by using a reverse distance weight interpolation algorithm fusing an elevation reliability factor; Step S160, performing grid unit difference calculation on the generated digital elevation model and a design elevation surface model, and accumulating total filling and total excavation quantities of the construction area according to the filling and excavation height and area of each unit; Step S170, performing task scheduling, process control and data management on the above steps S110 to S160 by an integrated system platform, and outputting a final earthwork quantity calculation result report and a visual chart. 2.The method of claim 1, wherein, In the step S120, scale-invariant feature transform feature points of each image are extracted based on a motion recovery structure algorithm, camera motion trajectories and sparse three-dimensional point clouds are recovered through multi-view geometric constraints, and a semi-global matching algorithm is used for dense matching to generate an initial three-dimensional point cloud model. 3.The method of claim 1, wherein, In the step S140, the intelligent classification process comprises the following steps: constructing a KD-tree spatial index of the point cloud, calculating a normal vector, local curvature and elevation variance three-dimensional geometric features for each point, using a pre-trained random forest classifier to extract features and make classification decisions for each point, performing Euclidean cluster analysis on point cloud clusters identified as non-ground interference, and removing small-scale noise points with a point number less than 50, and performing convex hull detection and contour analysis on large-scale point cloud clusters to only retain real ground point clouds. 4.The method of claim 1, wherein, In the step S150, the specific process of dynamically determining the grid division resolution comprises the following steps: dividing the construction area into analysis windows, and calculating a local terrain complexity index for each window.

5. An automatic earthwork volume calculation system based on UAV images, applied to the automatic earthwork volume calculation method based on UAV images in any one of claims 1-4, characterized in that, The system comprises the following components: A UAV image acquisition module for automatically flying along a preset flight path and acquiring a multi-view high-resolution image sequence of the construction area, and recording instantaneous position and attitude parameters of each image by using an integrated high-precision GNSS receiver and an IMU unit; The image preprocessing and three-dimensional reconstruction module receives the image data and pose parameters output by the unmanned aerial vehicle image acquisition module, performs image distortion correction and radiation correction, and generates an initial three-dimensional point cloud of the construction area based on a motion recovery structure algorithm and a dense matching technology; The coordinate dynamic calibration module couples the initial three-dimensional point cloud with an engineering coordinate system, constructs a coordinate conversion model based on the least squares principle, solves parameters by using ground control points or known engineering reference points, and realizes high-precision automatic registration of the three-dimensional model and the engineering coordinate system; The non-ground interference object intelligent identification and removal module loads the three-dimensional point cloud data after coordinate dynamic calibration, applies a point cloud classification algorithm based on random forests and three-dimensional geometric features, automatically identifies and filters out point cloud clusters corresponding to vegetation, temporary stockpiles and construction machinery, and outputs pure ground point clouds; The adaptive digital elevation model generation module receives the pure ground point clouds, dynamically calculates the optimal grid resolution according to the local point cloud density and the terrain undulation, and generates a high-fidelity digital elevation model by using an improved inverse distance weighting interpolation algorithm; The earthwork volume calculation module imports the design elevation surface data and the digital elevation model output by the adaptive digital elevation model generation module, performs grid difference operation, and calculates the total volume based on the fill and cut height and area of each grid element; The system control and data integration platform coordinates and controls the timing execution and data flow of all the above modules, and provides a graphical human-computer interaction interface for parameter configuration, task scheduling and result visualization. 6.The UAV image-based earthwork volume automatic calculation system of claim 5, wherein, The unmanned aerial vehicle image acquisition module includes a flight control sub-module, an image sensing sub-module and a positioning and pose determination sub-module; the flight control sub-module generates an optimal flight path according to a preset flight height, heading overlap and lateral overlap, and drives the unmanned aerial vehicle platform to fly along the path autonomously; the image sensing sub-module is mounted on the unmanned aerial vehicle gimbal and uses a global shutter CMOS sensor; and the positioning and pose determination sub-module integrates a high-frequency GNSS receiver and a tactical IMU. 7.The UAV image-based earthwork volume automatic calculation system of claim 5, wherein, The coordinate conversion model executed by the coordinate dynamic calibration module is a seven-parameter Bursa model; the coordinate dynamic calibration module solves the above seven conversion parameters by using a least squares adjustment algorithm through not less than three uniformly distributed ground control points. 8.The UAV image-based earthwork volume automatic calculation system of claim 5, wherein, The features used by the non-ground interference object intelligent identification and removal module include the normal vector, local curvature, elevation variance of the point cloud, and the statistics of the distribution of neighboring points based on KD-tree; the random forest classifier uses a labeled point cloud data set containing four categories of vegetation, stockpile, machinery and ground in the training stage, each decision tree splits nodes based on Gini impurity, and finally determines the category attribution of each point by voting of multiple trees; for the point cloud clusters classified as non-ground interference objects, the non-ground interference object intelligent identification and removal module performs spatial clustering analysis and removes them after secondary verification according to their geometric profile and size characteristics. 9.The UAV image-based earthwork volume automatic calculation system of claim 5, wherein, The optimal grid resolution in the adaptive digital elevation model generation module is determined according to a local terrain complexity index The adaptive digital elevation model generation module is provided with a complexity threshold value When , a higher resolution is used for grid division; when , a lower resolution is used for grid division; in the improved inverse distance weighted interpolation algorithm, a height reliability factor is introduced into a weight calculation function. 10.The UAV image-based earthwork volume automatic calculation system of claim 5, wherein, The earthwork quantity calculation module calculates the difference between the present elevation and the design elevation for each grid unit when performing the grid unit difference operation, and a positive value indicates the filling height and a negative value indicates the excavation height; the filling and excavation quantity calculation adopts the prism formula, and the total filling quantity and the total excavation quantity are obtained by accumulating the earthwork quantities of all grid units.

Images