Flight mapping control method and system of unmanned aerial vehicle and unmanned aerial vehicle

Abstract

The invention relates to a flight mapping control method and system of an unmanned aerial vehicle and the unmanned aerial vehicle. The method comprises the steps that firstly, point cloud, image and position data collected by an unmanned aerial vehicle sensor are acquired, format standardization and timestamp synchronization are achieved through preprocessing, and a multi-source data set is formed; based on the multi-source data set, determining an optimal data processing scheme and executing parallel data fusion through dynamic resource scheduling adaptive processing time delay and calculation load to obtain a fusion data result; performing real-time quality evaluation on a data fusion result, calculating and judging the coverage integrity and the precision level of a mapping region through a multi-dimensional index, and generating a quality evaluation result; and finally, constructing reference data based on a quality evaluation result, optimizing a flight path after simulation verification, generating a control instruction, and synchronously updating a data acquisition strategy. The precision and efficiency of flight mapping of the unmanned aerial vehicle are effectively improved, and the adaptability to a complex environment is enhanced.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) technology, and specifically relates to a flight mapping control method, system, and UAV for UAVs. Background Technology

[0002] In the current technological development process, drones have shown great application potential in many fields, such as agricultural monitoring, urban planning, and emergency rescue, making their flight mapping and control technology crucial. Existing drone flight mapping and control technologies offer various implementation schemes, but all have certain limitations. Positioning methods that partially rely on the Global Positioning System (GPS) become unreliable or even completely inaccessible areas with severe GPS signal blockage, such as indoor environments or canyons. While vision-based positioning and mapping schemes have become a popular research direction, they also face many challenges. For example, visual measurements have poor distance accuracy at long distances, error shifts are difficult to converge, and they are greatly affected by external conditions such as exposure and shadows. In complex outdoor scenes with alternating direct sunlight and shadows, the positioning and mapping effect is significantly reduced. In laser mapping schemes, common multi-source fusion point cloud matching methods suffer from low overall accuracy and complex algorithm structures. For example, patent CN201810597661.5 uses RTK information and laser point cloud edge feature matching for fusion positioning and mapping, but its accuracy in point cloud feature extraction is low, and it fails to utilize inter-frame correlation information, leading to a decrease in mapping accuracy. Furthermore, some UAV terrain-following algorithms rely on pre-flight elevation maps to calculate the optimal terrain-following route, failing to react in real-time to terrain changes. Alternatively, while high-precision laser radars are used, their large weight and difficulty in motion control cause the laser radar to acquire rearward distance information when the UAV tilts forward, affecting real-time terrain-following flight and mapping. Summary of the Invention

[0003] Therefore, it is necessary to provide a flight mapping control method, system, and UAV that can effectively improve the accuracy and efficiency of UAV flight mapping and enhance its adaptability to complex environments, in order to address the aforementioned technical problems.

[0004] In a first aspect, this application provides a flight mapping control method for an unmanned aerial vehicle (UAV), comprising:

[0005] The system acquires point cloud, image, and location data collected by drone sensors, and performs preprocessing to standardize the format and synchronize the timestamps, resulting in a multi-source data set.

[0006] Based on a multi-source dataset, dynamic resource scheduling is used to adapt to processing latency and computational load, determine the optimal data processing scheme, and perform parallel data fusion to obtain the fused data results.

[0007] Real-time quality assessment of the fused data results is performed, and the quality assessment results are obtained by calculating the coverage integrity and accuracy level of the mapped area through multi-dimensional indicators.

[0008] Based on the quality assessment results, benchmark data is constructed, and after simulation verification, the path is optimized and control commands are generated, while the data acquisition strategy is updated synchronously.

[0009] In one embodiment, based on a multi-source data set, dynamic resource scheduling is used to adapt processing latency and computational load, determine the optimal data processing scheme, and perform parallel data fusion to obtain fused data results, including:

[0010] Based on dynamic resource scheduling information from multi-source datasets, feature parameters of each data source are extracted and the system computing power is decomposed through a load balancing algorithm to generate a resource supply vector.

[0011] The initial scheme space is constructed by performing operations on the feature parameters and the resource supply vector. After filtering by constraints, the stability index is calculated to form an initial processing scheme matrix with priority ranking.

[0012] A multidimensional evaluation was conducted on each candidate scheme in the initial processing scheme matrix to obtain the evaluation results; the multidimensional evaluation included resource utilization, data fusion timeliness and fault tolerance indicators.

[0013] By combining real-time processing environmental parameters, the evaluation results are weighted and optimized using a reinforcement learning model to generate a Pareto optimal solution set; the Pareto optimal solution set is the set of solutions that satisfy the Pareto optimal state under resource and time delay constraints.

[0014] Execution schemes that meet the current computing power constraints are selected from the Pareto optimal solution set, and parallel data fusion operations are performed to generate initial fused data results.

[0015] If the processing latency of the initial fused data results exceeds the preset threshold, the resource allocation strategy will be dynamically adjusted based on the current load distribution.

[0016] Based on the adjusted resource allocation strategy, the resource allocation ratio is dynamically adjusted according to the load prediction algorithm to obtain the final fused data result. The load prediction algorithm adopts a temporal hybrid model, combined with a long short-term memory network to capture periodic load fluctuations, correct sudden changes, update samples through a sliding window, and output future load prediction values.

[0017] In one embodiment, the initial processing scheme matrix is ​​constructed using the following formula:

[0018] M ij =filter(tensor(F) i ,S j ))×λ j

[0019] Among them, M ij This represents the element in the i-th row and j-th column of the initial processing scheme matrix M, corresponding to the quantized value of the i-th feasible scheme in the j-th evaluation dimension. n represents the total number of feasible solutions retained after constraint filtering, k represents the number of dimensions for evaluating the solutions, and F i S represents the feature parameter vector of the i-th data source. j Let F represent the resource supply vector corresponding to the j-th evaluation dimension, and let tensor(F) be the vector. i ,S j ) represents the feature vector F of the data source. i With resource supply vector S j Tensor product operations are performed to generate the original combination of schemes through multidimensional space mapping. `filter(·)` represents the constraint filtering function, and λ... j This represents the stability weight coefficient of the j-th evaluation dimension.

[0020] In one embodiment, the parallel data fusion operation includes:

[0021] Based on the parallel processing rules in the execution scheme, spatiotemporal alignment preprocessing is performed on the multi-source data set, and the multi-source data type is obtained by secondary verification of the data structure characteristics.

[0022] Independent processing threads are allocated according to the multi-source data types, and special processing such as point cloud noise reduction and stitching, image feature extraction and matching, and spatiotemporal calibration of location information are performed synchronously to generate intermediate feature data.

[0023] The intermediate feature data output by each thread is correlated and fused using a multimodal fusion model. Through feature weight allocation and error compensation calculation, a structured initial fusion data result is output. The multimodal fusion model is generated by pre-training on historical multi-source data and adapted to various drone scenarios through transfer learning.

[0024] In one embodiment, the fused data results are evaluated in real time. The evaluation results are obtained by calculating the coverage integrity and accuracy level of the mapped area using multi-dimensional indicators, including:

[0025] Spatial coverage data is extracted from the fused data results, and a coverage integrity assessment matrix is ​​constructed based on the comparison results of spatial coverage with preset thresholds.

[0026] Based on the coverage integrity assessment matrix, point cloud features of non-compliant areas are extracted, and feature matching degree, normal vector consistency and geometric constraint error are calculated to generate a precision deviation tensor.

[0027] A Gaussian mixture model is constructed using the accuracy deviation tensor to determine the regional accuracy confidence level, and the correction process is triggered by comparison through Monte Carlo simulation.

[0028] In the calibration process, non-compliant areas are resampled, and registration parameters are optimized by combining location information to generate calibrated fusion data with enhanced spatiotemporal consistency.

[0029] Multi-scale feature descriptors are extracted from the calibrated and fused data to construct a 3D scene graph structure. Topological similarity is calculated based on graph neural networks to generate the final quality assessment result.

[0030] In one embodiment, the precision deviation tensor is calculated using the following formula:

[0031] T = ω1·H + ω2·N + ω3·G

[0032] Where T represents the precision bias tensor, and H represents the feature matching degree matrix. p i This represents the coordinates of the feature points in the point cloud to be evaluated. The coordinates of the reference feature point are represented by n, the number of matching point pairs is represented by n, and the normal vector consistency matrix is ​​represented by N. n j This represents the normal vector of the point cloud to be evaluated. Let I represent the reference normal vector, m represent the number of sampling points, I represent the 3×3 identity matrix, and G represent the geometric constraint error matrix. d l ω1 represents the distance from the point to the reference plane, k represents the number of constraint points, and ω1, ω2, and ω3 represent the weights of feature matching degree, normal vector consistency, and geometric constraint error, respectively.

[0033] In one embodiment, baseline data is constructed based on the quality assessment results, the path is optimized and control commands are generated after simulation verification, and the data acquisition strategy is updated synchronously, including:

[0034] Based on the quality assessment results, a flight control benchmark dataset is constructed, which integrates the current trajectory parameters of the UAV with the initial configuration of the data acquisition strategy to generate the input benchmark.

[0035] Based on the 3D mapping error field and feature sparsity distribution in the quality assessment results of the input benchmark, a multi-objective decision model for flight path optimization is constructed.

[0036] Based on the global optimal solution space of trajectory parameters obtained by the multi-objective decision model, dynamic control commands including the curvature smoothness of the cover point and the sensor triggering timing are generated.

[0037] The mapping performance of the trajectory parameters adjusted after the execution of dynamic control commands is simulated in a full-scene manner. If the point cloud density and image overlap in key areas do not meet the preset standard, the backup path planning based on graph optimization is initiated, and the globally optimal flight path correction scheme is output.

[0038] Based on the globally optimal flight path correction scheme, a flight control command set is generated through trajectory tracking, and the data acquisition strategy, including sensor exposure delay and data compression ratio parameters, is updated synchronously.

[0039] Secondly, this application also provides a flight mapping control system for an unmanned aerial vehicle (UAV), the system comprising:

[0040] The multi-source data acquisition module is used to acquire point cloud, image and location data collected by UAV sensors. After preprocessing, the data is standardized in format and synchronized with timestamps to obtain a multi-source data set.

[0041] The data fusion processing module is used to adapt processing latency and computing load based on multi-source data sets through dynamic resource scheduling, determine the optimal data processing scheme, and perform parallel data fusion to obtain fused data results.

[0042] The data quality assessment module is used to perform real-time quality assessment of the fused data results. It calculates the coverage integrity and accuracy level of the mapped area through multi-dimensional indicators to obtain the quality assessment results.

[0043] The flight strategy optimization module is used to build benchmark data based on the quality assessment results, optimize the path and generate control commands after simulation verification, and update the data acquisition strategy synchronously.

[0044] Thirdly, this application also provides an unmanned aerial vehicle (UAV) including a memory and a processor, wherein the memory stores a computer program and the processor executes the computer program to implement the method described above.

[0045] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0046] The aforementioned method, system, UAV, and storage medium for flight mapping control of a UAV first acquire point cloud, image, and location data collected by the UAV's sensors. These data are then preprocessed to achieve format standardization and timestamp synchronization, forming a multi-source data set. Based on this multi-source data set, dynamic resource scheduling is used to adapt to processing latency and computational load, determining the optimal data processing scheme and performing parallel data fusion to obtain the fused data result. Subsequently, the fused data result undergoes real-time quality assessment. Multi-dimensional indicators are used to calculate and determine the coverage integrity and accuracy level of the mapping area, generating a quality assessment result. Finally, benchmark data is constructed based on the quality assessment result. After simulation verification, the flight path is optimized, control commands are generated, and the data acquisition strategy is updated synchronously. Standardized preprocessing and timestamp synchronization of multi-source data ensure data consistency; the combination of dynamic resource scheduling and parallel data fusion improves data processing efficiency and can adapt to different processing latency and computational load; real-time quality assessment can promptly grasp the coverage integrity and accuracy level of the mapping area, providing an accurate basis for flight path optimization; optimizing the path and updating the data acquisition strategy based on the quality assessment results forms a closed loop from data acquisition to flight control, effectively improving the accuracy and efficiency of UAV flight mapping and enhancing adaptability to complex environments. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 A flowchart of a flight mapping control method for an unmanned aerial vehicle (UAV) provided in an embodiment of the present invention;

[0049] Figure 2 This is a structural block diagram of a flight mapping control system for an unmanned aerial vehicle (UAV) provided in an embodiment of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0051] In one embodiment, such as Figure 1 As shown, this application provides a flight mapping control method for an unmanned aerial vehicle (UAV), which may include the following steps:

[0052] Step S101: Acquire point cloud, image and location data collected by UAV sensors, and perform preprocessing to achieve format standardization and timestamp synchronization to obtain a multi-source data set.

[0053] Specifically, the drone collects environmental 3D point cloud data, 2D image data, and real-time position and attitude data through sensors such as lidar, camera, and GPS / IMU. In the preprocessing stage, the different types of data are converted into formats and unified into a preset data structure standard. The timestamps of each data are calibrated based on sensor hardware trigger signals and system clock to ensure the consistency of point cloud, image, and position information in the time dimension, ultimately forming a multi-source data set that can be directly used for subsequent processing.

[0054] Step S102: Based on the multi-source data set, adapt the processing latency and computing load through dynamic resource scheduling, determine the optimal data processing scheme, and perform parallel data fusion to obtain the fused data result.

[0055] First, the data volume, feature complexity, and real-time requirements of the multi-source dataset are analyzed. Combined with the current system's computing resource usage (such as CPU / GPU utilization and memory usage), a dynamic resource scheduling algorithm is used to allocate processing threads and computing resources to balance processing latency and load pressure. Based on the scheduling results, the optimal data processing scheme is selected from the preset scheme library, and a multi-threaded parallel processing mechanism is started to perform spatiotemporal correlation and feature fusion on the standardized point cloud, image, and location data. Finally, a fused data result containing environmental geometric information, texture features, and precise location labels is generated.

[0056] Step S103: Perform real-time quality assessment on the fused data results. Calculate and judge the coverage integrity and accuracy level of the mapped area through multi-dimensional indicators to obtain the quality assessment results.

[0057] Specifically, indicators such as spatial coverage, point cloud density, and image overlap are extracted from the fused data results to construct a coverage integrity assessment model to determine whether there are missing or sparse areas in the mapped area; at the same time, the deviation between the fused data and the reference data (such as known landmark coordinates and high-precision map fragments) is calculated to assess the data accuracy level; the quantitative results of coverage integrity and accuracy level are integrated to form a quality assessment result that includes the quality level of each region.

[0058] Step S104: Based on the quality assessment results, construct benchmark data, optimize the path and generate control commands after simulation verification, and update the data acquisition strategy synchronously.

[0059] Based on the low-quality areas identified in the quality assessment results (such as areas with missing coverage or substandard accuracy), baseline data for path optimization is constructed by combining the UAV's current location, remaining battery power, and mission requirements. The mapping effects of different path adjustment schemes are simulated in a digital twin simulation environment to verify their feasibility and effectiveness. New flight trajectory parameters (such as waypoint coordinates and flight speed) are generated based on the simulation results and converted into control commands that the UAV can execute. At the same time, according to the characteristics of the optimized path, data acquisition strategy parameters such as sensor sampling frequency and data resolution are updated to ensure that higher quality data is obtained in key areas.

[0060] The aforementioned UAV flight mapping control method first acquires point cloud, image, and location data collected by UAV sensors. This data is then preprocessed to achieve format standardization and timestamp synchronization, forming a multi-source data set. Based on this multi-source data set, dynamic resource scheduling is used to adapt to processing latency and computational load, determining the optimal data processing scheme and performing parallel data fusion to obtain the fused data result. Subsequently, real-time quality assessment is performed on the fused data result, using multi-dimensional indicators to determine the coverage integrity and accuracy level of the mapped area, generating a quality assessment result. Finally, benchmark data is constructed based on the quality assessment result, and after simulation verification, the flight path is optimized and control commands are generated, synchronously updating the data acquisition strategy. The standardized preprocessing and timestamp synchronization of multi-source data ensure data consistency; the combination of dynamic resource scheduling and parallel data fusion improves data processing efficiency and adapts to different processing latency and computational loads; real-time quality assessment allows for timely understanding of the coverage integrity and accuracy level of the mapped area, providing accurate basis for flight path optimization; optimizing the path and updating the data acquisition strategy based on the quality assessment result forms a closed loop from data acquisition to flight control, effectively improving the accuracy and efficiency of UAV flight mapping and enhancing adaptability to complex environments.

[0061] In one embodiment, based on a multi-source data set, dynamic resource scheduling is used to adapt processing latency and computational load, determine the optimal data processing scheme, and perform parallel data fusion to obtain the fused data result. This may include the following steps:

[0062] Step S201: Based on the dynamic resource scheduling information of the multi-source data set, extract the feature parameters of each data source and decompose the system computing power through a load balancing algorithm to generate a resource supply vector.

[0063] Preferably, the dynamic resource scheduling information is a set of parameters that reflect the real-time status of the system's current computing resources (CPU / GPU / memory), task queue length, and processing priority, and is used to guide the allocation of computing power.

[0064] Feature parameters are metrics that quantify the characteristics of various data sources, such as the density of point cloud data (points / cubic meter), the resolution of image data (pixels × pixels), and the update frequency of location data (Hz).

[0065] Load balancing algorithms include round-robin, least connections, or weighted distribution based on task priority.

[0066] Step S202: The feature parameters and resource supply vector are used to construct an initial scheme space. After filtering by constraints, the stability index is calculated to form an initial processing scheme matrix with priority sorting.

[0067] Constraints include upper limits for filtering application resources (e.g., GPU utilization ≤ 80%), lower limits for latency (e.g., processing latency ≤ 100ms), and data integrity thresholds (e.g., point cloud coverage ≥ 95%).

[0068] Step S203: Perform multi-dimensional evaluation on each candidate scheme in the initial processing scheme matrix to obtain the evaluation results; the multi-dimensional evaluation includes resource utilization, data fusion timeliness and fault tolerance indicators.

[0069] Step S204: Combine real-time processing environment parameters and use a reinforcement learning model to optimize the weights of the evaluation results to generate a Pareto optimal solution set; the Pareto optimal solution set is the set of solutions that satisfy the Pareto optimal state under resource and time delay constraints.

[0070] Step S205: Select execution schemes that meet the current computing power constraints from the Pareto optimal solution set, perform parallel data fusion operations, and generate initial fusion data results.

[0071] Step S206: If the processing delay of the initial fused data results exceeds the preset threshold, dynamically adjust the resource allocation strategy based on the current load distribution.

[0072] Step S207: Based on the adjusted resource allocation strategy, dynamically adjust the resource allocation ratio according to the load prediction algorithm to obtain the final fused data result; the load prediction algorithm adopts a temporal hybrid model, combines a long short-term memory network to capture periodic load fluctuations, corrects sudden changes, updates samples through a sliding window, and outputs future load prediction values.

[0073] Specifically, based on dynamic resource scheduling information from multi-source datasets, feature parameters of each data source are extracted. The system's computing power is decomposed using a load balancing algorithm to generate a resource supply vector. The feature parameters and resource supply vector are used to construct an initial scheme space. After constraint filtering, stability indices are calculated, forming an initial processing scheme matrix with priority ranking. Each candidate scheme in this matrix is ​​evaluated from multiple dimensions based on resource utilization, data fusion timeliness, and fault tolerance, yielding evaluation results. Combining real-time processing environment parameters, a reinforcement learning model is used to optimize the weights of the evaluation results, generating a set of schemes that satisfy Pareto optimality under resource and latency constraints. Execution schemes that meet the current computing power constraints are selected from this set, and parallel data fusion operations are performed to generate initial fused data results. If the processing latency of the initial fused data results exceeds a preset threshold, the resource allocation strategy is dynamically adjusted based on the current load distribution. Then, based on the adjusted strategy, a load prediction algorithm using a temporal hybrid model (combining a long short-term memory network to capture periodic load fluctuations, correct for sudden changes, update samples through a sliding window, and output future load prediction values) is used to dynamically adjust the resource allocation ratio, resulting in the final fused data results.

[0074] This embodiment achieves precise matching between data features and system computing power through feature parameter extraction and resource supply vector generation; initial scheme space construction and priority ranking provide an orderly foundation for scheme selection; multi-dimensional evaluation and reinforcement learning optimization ensure the scheme is optimal in terms of resources, timeliness, and fault tolerance; the application of Pareto optimal solution set ensures the effectiveness of the scheme under resource and latency constraints; and the use of dynamic adjustment of resource allocation strategy and load prediction algorithm can promptly deal with situations where latency exceeds the threshold, ultimately improving the efficiency and stability of parallel data fusion and ensuring the quality of fused data results.

[0075] In one embodiment, the initial processing scheme matrix is ​​constructed using the following formula:

[0076] M ij =filter(tensor(F) i ,S j ))×λ j

[0077] Among them, M ij This represents the element in the i-th row and j-th column of the initial processing scheme matrix M, corresponding to the quantized value of the i-th feasible scheme in the j-th evaluation dimension. n represents the total number of feasible solutions retained after constraint filtering, k represents the number of dimensions for evaluating the solutions, and F i S represents the feature parameter vector of the i-th data source. j Let F represent the resource supply vector corresponding to the j-th evaluation dimension, and let tensor(F) be the vector. i ,S j) represents the feature vector F of the data source. i With resource supply vector S j Tensor product operations are performed to generate the original combination of schemes through multidimensional space mapping. `filter(·)` represents the constraint filtering function, and λ... j This represents the stability weight coefficient of the j-th evaluation dimension.

[0078] This embodiment achieves a multi-dimensional spatial mapping between data characteristics and resource allocation by performing a tensor product operation between the data source feature parameter vector and the resource supply vector, providing a quantitative basis for generating a comprehensive combination of original solutions. Combined with a constraint filtering function, invalid solutions that do not meet constraints such as resources and latency can be accurately eliminated, ensuring that the retained feasible solutions have practical execution value. By introducing a stability weight coefficient to weight and quantify each evaluation dimension, the matrix elements can intuitively reflect the comprehensive performance of different feasible solutions in each evaluation dimension, and an ordered initial processing solution matrix is ​​formed by prioritizing the sorting.

[0079] In one embodiment, the parallel data fusion operation may include:

[0080] Step S301: Based on the parallel processing rules in the execution scheme, perform spatiotemporal alignment preprocessing on the multi-source data set, and obtain the multi-source data type by secondary verification of data structure features.

[0081] Specifically, the multi-source dataset is first preprocessed with spatiotemporal alignment according to parallel processing rules (including thread allocation rules, synchronization mechanism rules, and data interface rules). This is achieved by calibrating the timestamps of each data source with the system's reference clock and transforming the spatial coordinates of point cloud, image, and location data to a unified coordinate system to eliminate spatiotemporal deviations. Subsequently, a secondary verification of the data structure characteristics is performed, which involves analyzing the data's storage format (such as the .pcd format for point clouds and the .jpg format for images) and field information (such as the latitude and longitude fields in location data). Finally, the multi-source data type to which each data belongs is determined.

[0082] Step S302: Allocate independent processing threads according to the multi-source data types, and synchronously perform special processing such as point cloud noise reduction and stitching, image feature extraction and matching, and spatiotemporal calibration of location information to generate intermediate feature data.

[0083] Step S303: The intermediate feature data output by each thread is correlated and fused using a multimodal fusion model. Through feature weight allocation and error compensation calculation, a structured initial fusion data result is output. The multimodal fusion model is generated by pre-training on historical multi-source data and adapted to various drone scenarios through transfer learning.

[0084] Multimodal fusion models can process and integrate various types of data such as point clouds, images, and locations, and achieve cross-modal information fusion by learning the correlation patterns between data.

[0085] Error compensation can correct systematic and random errors generated during the acquisition and processing of multi-source data through preset algorithms, thereby reducing the impact of errors on the fusion results.

[0086] Specifically, based on the parallel processing rules in the execution scheme, the multi-source data set is first preprocessed with spatiotemporal alignment. Secondary verification of data structure features is used to determine the data types, including point clouds, images, and location information. Then, independent processing threads are allocated according to the data types to synchronously execute specialized processing—denoising and stitching of point cloud data, feature extraction and matching of image data, and spatiotemporal calibration of location information—generating corresponding intermediate feature data. Finally, a multimodal fusion model is used to correlate and fuse the intermediate feature data output from each thread. Through feature weight allocation and error compensation calculation, a structured initial fusion data result is output. The multimodal fusion model is pre-trained from historical multi-source data and adapted for various UAV scenarios through transfer learning.

[0087] In this embodiment, spatiotemporal alignment preprocessing and secondary data type verification ensure the consistency of multi-source data in the temporal and spatial dimensions and the accuracy of type attribution; independent threads are allocated according to data type to execute special processing synchronously, which improves data processing efficiency and can give full play to the system's parallel computing capabilities; the multimodal fusion model, combined with feature weight allocation and error compensation, effectively integrates different types of intermediate feature data, improving the accuracy and completeness of the initial fusion data results; and the model is pre-trained and adapted to UAV scenarios through transfer learning, further ensuring the adaptability and reliability of the fusion effect.

[0088] In one embodiment, real-time quality assessment of the fused data results, which involves calculating and determining the coverage integrity and accuracy level of the mapped area using multi-dimensional indicators, may include the following steps:

[0089] Step S401: Extract spatial coverage data from the fused data results, and construct a coverage integrity evaluation matrix based on the comparison results of spatial coverage and preset threshold.

[0090] Preferably, spatial coverage data refers to the coverage ratio of the fused data results within the target mapping area, usually counted in grid units, such as the proportion of data points within a 10m × 10m grid.

[0091] The coverage integrity assessment matrix presents the coverage compliance status of each sub-region in matrix form, with element values ​​of 0 (not compliant) or 1 (compliant), and rows / columns corresponding to the grid coordinates of the mapped region.

[0092] Step S402: Extract point cloud features of non-compliant areas based on the coverage integrity assessment matrix, calculate feature matching degree, normal vector consistency and geometric constraint error, and generate accuracy deviation tensor.

[0093] The non-compliant area is the area with an element value of 0 in the coverage integrity assessment matrix, that is, the area where the data coverage density is lower than the preset threshold.

[0094] Step S403: Construct a Gaussian mixture model using the accuracy deviation tensor to determine the regional accuracy confidence level, and trigger the correction process through Monte Carlo simulation comparison.

[0095] Specifically, a Gaussian mixture model is trained using sample data of the precision deviation tensor. The model consists of K Gaussian components (K is set according to the complexity of the deviation distribution), and each component corresponds to a deviation pattern. The model outputs the precision confidence level (0-100%) of the non-compliant region, which is the probability that the deviation is within the allowable range. When performing Monte Carlo simulation, 1000 random error samples are generated and compared with the actual precision deviation tensor. If the proportion of simulated deviations exceeding the allowable range exceeds 5%, an adaptive correction process is triggered.

[0096] Monte Carlo simulation comparison is a method that simulates different error scenarios through multiple random samplings, compares the simulation results with the actual deviations, and determines whether to trigger the correction process.

[0097] Step S404: In the calibration process, the non-compliant areas are resampled, and the registration parameters are optimized by combining the location information to generate calibrated fusion data with enhanced spatiotemporal consistency.

[0098] Step S405: Extract multi-scale feature descriptors from the corrected fused data to construct a 3D scene graph structure, calculate topological similarity based on graph neural network, and generate the final quality assessment result.

[0099] Furthermore, multi-scale feature descriptors are feature vectors extracted from different resolutions (e.g., 1 meter, 0.5 meters), containing information such as the geometric shape and texture details of the region. The 3D scene graph structure is a 3D spatial topology represented by nodes (feature points) and edges (feature relationships). Nodes store feature descriptors, and edges represent the distance and angular relationships between features.

[0100] From the fused data after correction, feature descriptors, such as PFH (Point Feature Histogram) and SIFT (Scale Invariant Feature Transform), are extracted at both large scale (capturing the overall outline of the region) and small scale (capturing detailed features). Using feature points as nodes and spatial relationships between features as edges, a 3D scene graph structure is constructed. The scene graph is then input into a graph neural network, which aggregates features from adjacent nodes through a message passing mechanism. The topological similarity score between each sub-region in the graph and a preset standard scene graph is calculated. Based on the score, quality levels (e.g., excellent, medium, poor) are assigned, generating a final quality assessment result that includes the quality levels of each region.

[0101] Specifically, spatial coverage data is extracted from the fused data results and compared with a preset threshold. A coverage integrity assessment matrix is ​​constructed based on the comparison results. Point cloud features of non-compliant areas are extracted based on this matrix. A precision deviation tensor is generated by calculating feature matching degree, normal vector consistency, and geometric constraint error. A Gaussian mixture model is constructed using the precision deviation tensor to determine the regional precision confidence level. An adaptive correction process is then triggered by Monte Carlo simulation comparison. In the correction process, non-compliant areas are resampled, and registration parameters are optimized by combining location information to generate corrected fused data with enhanced spatiotemporal consistency. Finally, multi-scale feature descriptors are extracted from the corrected fused data to construct a 3D scene graph structure. Topological similarity is calculated based on graph neural networks to generate the final quality assessment result.

[0102] This embodiment can accurately locate missing or sparse areas in the mapping by comparing spatial coverage with a preset threshold and constructing a coverage integrity evaluation matrix. The accuracy deviation tensor generated based on the point cloud features of the non-compliant areas can comprehensively reflect the deviation of the mapping accuracy of the area. The combination of Gaussian mixture model and Monte Carlo simulation provides a scientific basis for triggering the correction process. The resampling and registration parameter optimization of the non-compliant areas in the correction process effectively improves the spatiotemporal consistency of the fused data. Finally, the quality evaluation results generated by the 3D scene graph structure and graph neural network calculation can comprehensively and accurately reflect the coverage integrity and accuracy level of the mapped area.

[0103] In one embodiment, the precision deviation tensor can be calculated using the following formula:

[0104] T = ω1·H + ω2·N + ω3·G

[0105] Where T represents the precision bias tensor, and H represents the feature matching degree matrix. p i This represents the coordinates of the feature points in the point cloud to be evaluated. The coordinates of the reference feature point are represented by n, the number of matching point pairs is represented by n, and the normal vector consistency matrix is ​​represented by N. n jThis represents the normal vector of the point cloud to be evaluated. Let I represent the reference normal vector, m represent the number of sampling points, I represent the 3×3 identity matrix, and G represent the geometric constraint error matrix. d l ω1 represents the distance from the point to the reference plane, k represents the number of constraint points, and ω1, ω2, and ω3 represent the weights of feature matching degree, normal vector consistency, and geometric constraint error, respectively.

[0106] This embodiment achieves multi-dimensional quantification of the accuracy deviation of substandard areas by weighted integration of the feature matching degree matrix, the normal vector consistency matrix, and the geometric constraint error matrix. The feature matching degree reflects the accuracy of the correspondence between point cloud features, the normal vector consistency reflects the degree of conformity of surface geometric attributes, and the geometric constraint error quantifies the deviation of spatial position. The accuracy deviation tensor formed by the combination of the three with weight coefficients can comprehensively and accurately characterize the accuracy defects of the area, effectively improving the scientific nature of quality assessment and the pertinence of correction operations.

[0107] In one embodiment, constructing benchmark data based on quality assessment results, optimizing the path and generating control commands after simulation verification, and synchronously updating the data acquisition strategy may include the following steps:

[0108] Step S501: Construct a flight control benchmark dataset based on the quality assessment results, integrate the current trajectory parameters of the UAV with the initial configuration of the data acquisition strategy, and generate an input benchmark.

[0109] The flight control benchmark dataset contains key information such as the quality level of the mapped area, the coordinates of the non-compliant areas, and the accuracy deviation value.

[0110] Trajectory parameters are quantitative indicators that describe the flight trajectory of a UAV, including waypoint coordinates, flight speed, heading angle, waypoint curvature, etc., which determine the flight path and attitude of the UAV.

[0111] The initial configuration of the data acquisition strategy includes preset sensor operating parameters before the UAV performs the mission, such as lidar sampling frequency, camera exposure time, and data compression ratio.

[0112] Step S502: Based on the three-dimensional mapping error field and feature sparsity distribution in the input benchmark analytical quality assessment results, construct a multi-objective decision model for flight path optimization.

[0113] The three-dimensional mapping error field is represented by a three-dimensional grid, which shows the distribution of accuracy deviation in the mapping area. The value of each grid node corresponds to the magnitude of the mapping error at that location.

[0114] Feature sparsity distribution reflects the distribution of the density of data features (such as point clouds and image feature points) within the mapping area, and identifies sparse areas where data collection is insufficient.

[0115] Specifically, based on the trajectory parameters and data acquisition configuration integrated in the input benchmark, specific data of the 3D mapping error field are extracted from the quality assessment results, namely the error values ​​and distribution range of each region, while clarifying the location and area of ​​the sparse regions in the feature sparsity distribution. Combining the flight performance limitations of the UAV (such as maximum turning radius and minimum flight speed) and mission requirements (such as mapping coverage and accuracy threshold), a multi-objective decision model is constructed. This model takes reducing 3D mapping errors and reducing feature sparsity regions as the main optimization objectives, while also taking into account the length of the flight path and energy consumption. By setting the weight coefficients of each objective, the multi-objective problem is transformed into a solvable single-objective optimization problem.

[0116] Step S503: Based on the multi-objective decision model, solve the global optimal solution space of trajectory parameters and generate dynamic control commands including the curvature smoothness of the cover point and the sensor triggering timing.

[0117] The multi-objective decision model comprehensively considers multiple objectives such as improving mapping accuracy, flight path smoothness, and data acquisition efficiency to solve the mathematical model of the optimal path scheme.

[0118] Based on the multi-objective decision-making model, the trajectory parameters are solved using particle swarm optimization or genetic algorithms. Under the premise of satisfying flight constraints, the global optimal solution space of the trajectory parameters is searched. The solution with the best overall performance is selected from the solution space to generate dynamic control commands. Among them, the waypoint curvature smoothness command ensures the smooth flight attitude of the UAV and avoids sharp turns by controlling the rate of change of curvature between adjacent waypoints. The sensor trigger timing command precisely controls the start and stop time of sensors (such as cameras and lidar) according to the waypoint position and flight speed to ensure that sufficient data can be collected in key areas.

[0119] Step S504: Perform full-scene simulation of mapping performance on the trajectory parameters adjusted after executing the dynamic control command. If the point cloud density and image overlap in the key area do not meet the preset standard, start the backup path planning based on graph optimization and output the globally optimal flight path correction scheme.

[0120] Using a digital twin simulation platform, a full-scenario simulation of mapping efficiency is performed on the trajectory parameters adjusted after executing dynamic control commands. The flight process and data acquisition of the UAV under this trajectory are simulated, and the point cloud density (number of points per unit volume) and image overlap (percentage of overlapping areas between adjacent images) of key areas are calculated. The calculation results are compared with preset standards (e.g., point cloud density ≥ 50 points / cubic meter, image overlap ≥ 70%). If the standards are not met, a graph-based alternative path planning is initiated. Waypoints are used as nodes in the graph, and the flight costs between waypoints (e.g., distance, energy consumption) are used as edge weights. By iteratively optimizing the graph structure, a path that can cover the key areas and meet the data acquisition standards is found, and finally, the globally optimal flight path correction scheme is output.

[0121] Step S505: Based on the globally optimal flight path correction scheme, generate a flight control command set through trajectory tracking, and synchronously update the data acquisition strategy, including sensor exposure delay and data compression ratio parameters.

[0122] Based on the globally optimal flight path correction scheme, the position, speed, and attitude that the UAV needs to reach at each moment are calculated through trajectory tracking algorithms, and then converted into a specific flight control command set, including throttle size, control surface deflection angle, etc., to control the UAV to fly along the corrected path. At the same time, according to the characteristics of the corrected path, such as the need for denser data collection in sparse data areas, the data collection strategy parameters are updated synchronously, the sensor exposure delay is adjusted to ensure image clarity, and the data compression ratio is modified to balance data volume and transmission efficiency, ensuring that the collected data can meet the mapping requirements.

[0123] Specifically, a flight control benchmark dataset is constructed based on the quality assessment results. This dataset integrates the current trajectory parameters of the UAV (such as waypoint coordinates, flight speed, and heading angle) with the initial configuration of the data acquisition strategy (such as sensor sampling frequency and data resolution) to generate an input benchmark. Based on this input benchmark, the 3D mapping error field (reflecting mapping accuracy deviations in different regions) and feature sparsity distribution (identifying areas with insufficient data acquisition) in the quality assessment results are analyzed. A multi-objective decision-making model for flight path optimization, aimed at improving mapping accuracy and reducing feature sparsity regions, is constructed. The global optimal solution space for the trajectory parameters is solved using the multi-objective decision-making model, generating dynamic control commands that include waypoint curvature smoothness (ensuring flight stability) and sensor trigger timing (ensuring data acquisition synchronization). A full-scene simulation of the mapping performance is performed on the trajectory parameters adjusted after executing the dynamic control commands. If the point cloud density and image overlap in key areas do not meet the preset standards, a graph-optimized backup path planning is initiated, outputting a globally optimal flight path correction scheme. Finally, a flight control command set is generated through trajectory tracking based on this correction scheme, synchronously updating the data acquisition strategy, including sensor exposure delay and data compression ratio parameters.

[0124] This embodiment achieves precise conversion of quality assessment results into flight control parameters by constructing a benchmark dataset and a multi-objective decision-making model, ensuring that path optimization balances mapping accuracy and flight stability. The combination of dynamic control commands and backup path planning enhances fault tolerance in complex scenarios and guarantees the quality of data acquisition in key areas. Synchronous updates to the data acquisition strategy adapt sensor parameters to the flight trajectory. Finally, the control command set generated through trajectory tracking achieves coordinated optimization of flight path and data acquisition, significantly improving the efficiency and accuracy of UAV mapping.

[0125] In one embodiment, such as Figure 2 As shown, this application also provides a flight mapping control system for an unmanned aerial vehicle (UAV), which may include:

[0126] The multi-source data acquisition module 601 is used to acquire point cloud, image and location data collected by UAV sensors. After preprocessing, the format is standardized and the timestamp is synchronized to obtain a multi-source data set.

[0127] The data fusion processing module 602 is used to adapt the processing latency and computing load based on the multi-source data set through dynamic resource scheduling, determine the optimal data processing scheme, and perform parallel data fusion to obtain the fused data result.

[0128] The data quality assessment module 603 is used to perform real-time quality assessment on the fused data results. It calculates and judges the coverage integrity and accuracy level of the mapped area through multi-dimensional indicators to obtain the quality assessment results.

[0129] The flight strategy optimization module 604 is used to construct baseline data based on the quality assessment results, optimize the path and generate control commands after simulation verification, and update the data acquisition strategy synchronously.

[0130] The aforementioned UAV flight mapping control system comprises a multi-source data acquisition module that acquires point cloud, image, and location data collected by UAV sensors. Through preprocessing, it standardizes the format and synchronizes timestamps to form a multi-source data set, providing a unified and time-consistent data source for subsequent processing. The data fusion processing module, based on this multi-source data set, dynamically schedules resources to adapt to processing latency and computational load, determines the optimal data processing scheme, and performs parallel data fusion to generate fused data results, achieving efficient integration of multi-source data. The data quality assessment module performs real-time quality assessment of the fused data results, using multi-dimensional indicators to calculate and judge the coverage integrity and accuracy level of the mapped area, outputting quality assessment results to provide a basis for path optimization. The flight strategy optimization module constructs benchmark data based on the quality assessment results, optimizes the flight path after simulation verification, generates control commands, and synchronously updates the data acquisition strategy.

[0131] The preprocessing of the multi-source data acquisition module ensures data consistency and availability, laying the foundation for subsequent fusion; the dynamic scheduling and parallel fusion of the data fusion processing module improves data processing efficiency and adapts to the computing needs of different scenarios; the multi-dimensional evaluation of the data quality assessment module ensures the reliability of the fused data and provides accurate reference for path optimization; the flight strategy optimization module optimizes the path and strategy based on the evaluation results, realizing dynamic adjustment and closed-loop control of flight mapping, and improving the accuracy, efficiency and adaptability of UAV flight mapping as a whole.

[0132] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0133] In one embodiment, a drone is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the flight mapping control method, system, and steps of the drone as described above.

[0134] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0135] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0136] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A flight mapping control method for an unmanned aerial vehicle (UAV), characterized in that, The method includes: The point cloud, image, and location data collected by the drone's sensors are acquired, and after preprocessing, the format is standardized and the timestamp is synchronized to obtain a multi-source data set; Based on the multi-source data set, dynamic resource scheduling is used to adapt to processing latency and computational load, determine the optimal data processing scheme, and perform parallel data fusion to obtain fused data results. The fused data results are evaluated in real time, and the coverage integrity and accuracy level of the mapped area are determined by multi-dimensional index calculation to obtain the quality evaluation result; Based on the quality assessment results, benchmark data is constructed, and after simulation verification, the path is optimized and control commands are generated, while the data acquisition strategy is updated synchronously.

2. The method according to claim 1, characterized in that, The process involves dynamically scheduling resources to adapt to processing latency and computational load based on the multi-source data set, determining the optimal data processing scheme, and performing parallel data fusion to obtain the fused data result, including: Based on the dynamic resource scheduling information of the multi-source data set, the characteristic parameters of each data source are extracted and the system computing power is decomposed through a load balancing algorithm to generate a resource supply vector. The feature parameters and resource supply vector are used to construct an initial scheme space. After filtering by constraints, the stability index is calculated to form an initial processing scheme matrix with priority ranking. A multidimensional evaluation is performed on each candidate scheme in the initial processing scheme matrix to obtain the evaluation results; the multidimensional evaluation includes resource utilization rate, data fusion timeliness and fault tolerance rate indicators; By combining real-time processing environmental parameters, the evaluation results are weighted and optimized using a reinforcement learning model to generate a Pareto optimal solution set; the Pareto optimal solution set is a set of solutions that satisfy the Pareto optimal state under resource and time delay constraints. Execution schemes that meet the current computing power constraints are selected from the Pareto optimal solution set, and parallel data fusion operations are performed to generate initial fusion data results; If the processing delay of the initial fusion data results exceeds a preset threshold, the resource allocation strategy will be dynamically adjusted based on the current load distribution. Based on the adjusted resource allocation strategy, the resource allocation ratio is dynamically adjusted according to the load prediction algorithm to obtain the final fused data result. The load prediction algorithm adopts a temporal hybrid model, combined with a long short-term memory network to capture periodic load fluctuations, correct sudden changes, update samples through a sliding window, and output future load prediction values.

3. The method according to claim 2, characterized in that, The initial processing scheme matrix is ​​constructed using the following formula: M ij =filter(tensor(F i ,S j ))×λ j Among them, M ij This represents the element in the i-th row and j-th column of the initial processing scheme matrix M, corresponding to the quantized value of the i-th feasible scheme in the j-th evaluation dimension. n represents the total number of feasible solutions retained after constraint filtering, k represents the number of dimensions for evaluating the solutions, and F i S represents the feature parameter vector of the i-th data source. j Let F represent the resource supply vector corresponding to the j-th evaluation dimension, and let tensor(F) be the vector. i ,S j ) represents the feature vector F of the data source. i With resource supply vector S j Tensor product operations are performed to generate the original combination of schemes through multidimensional space mapping. `filter(·)` represents the constraint filtering function, and λ... j This represents the stability weight coefficient of the j-th evaluation dimension.

4. The method according to claim 2, characterized in that, The parallel data fusion operation includes: Based on the parallel processing rules in the execution scheme, the multi-source data set is preprocessed with spatiotemporal alignment, and the multi-source data type is obtained by secondary verification of data structure features to determine the type attribution. Independent processing threads are allocated according to the multi-source data types to synchronously perform special processing such as point cloud noise reduction and stitching, image feature extraction and matching, and spatiotemporal calibration of location information to generate intermediate feature data; The intermediate feature data output by each thread is correlated and fused using a multimodal fusion model. Through feature weight allocation and error compensation calculation, a structured initial fusion data result is output. The multimodal fusion model is generated by pre-training on historical multi-source data and adapted to various drone scenarios through transfer learning.

5. The method according to claim 1, characterized in that, The real-time quality assessment of the fused data results, which calculates and judges the coverage integrity and accuracy level of the mapped area using multi-dimensional indicators, includes: Spatial coverage data is extracted from the fused data results, and a coverage integrity evaluation matrix is ​​constructed based on the comparison results of spatial coverage and preset threshold. Based on the coverage integrity assessment matrix, point cloud features of unqualified areas are extracted, and feature matching degree, normal vector consistency and geometric constraint error are calculated to generate a precision deviation tensor. The Gaussian mixture model is constructed using the aforementioned accuracy deviation tensor to determine the regional accuracy confidence level, and the correction process is triggered by comparison through Monte Carlo simulation. In the correction process, the non-compliant areas are resampled, and the registration parameters are optimized by combining the location information to generate corrected fusion data with enhanced spatiotemporal consistency. Multi-scale feature descriptors are extracted from the corrected and fused data to construct a 3D scene graph structure. Topological similarity is calculated based on graph neural networks to generate the final quality assessment result.

6. The method according to claim 5, characterized in that, The precision deviation tensor is calculated using the following formula: T = ω1·H + ω2·N + v3·G Where T represents the precision bias tensor, and H represents the feature matching degree matrix. p i This represents the coordinates of the feature points in the point cloud to be evaluated. The coordinates of the reference feature point are represented by n, the number of matching point pairs is represented by n, and the normal vector consistency matrix is ​​represented by N. n j This represents the normal vector of the point cloud to be evaluated. Let I represent the reference normal vector, m represent the number of sampling points, I represent the 3×3 identity matrix, and G represent the geometric constraint error matrix. d l ω1 represents the distance from the point to the reference plane, k represents the number of constraint points, and ω1, ω2, and ω3 represent the weights of feature matching degree, normal vector consistency, and geometric constraint error, respectively.

7. The method according to claim 1, characterized in that, Based on the quality assessment results, benchmark data is constructed, and after simulation verification, the path is optimized and control commands are generated. The data acquisition strategy is updated synchronously, including: Based on the quality assessment results, a flight control benchmark dataset is constructed, and the current trajectory parameters of the UAV and the initial configuration of the data acquisition strategy are integrated to generate an input benchmark. Based on the input benchmark, the three-dimensional mapping error field and feature sparsity distribution in the quality assessment results are analyzed to construct a multi-objective decision model for flight path optimization; Based on the global optimal solution space of the trajectory parameters obtained by the multi-objective decision model, dynamic control commands including the curvature smoothness of the cover point and the sensor triggering timing are generated. The mapping performance of the trajectory parameters adjusted after the execution of the dynamic control command is simulated in a full-scene simulation. If the point cloud density and image overlap in the key area do not meet the preset standard, the backup path planning based on graph optimization is initiated, and the globally optimal flight path correction scheme is output. Based on the global optimal flight path correction scheme, a flight control command set is generated through trajectory tracking, and the data acquisition strategy, including sensor exposure delay and data compression ratio parameters, is updated synchronously.

8. A flight mapping control system for an unmanned aerial vehicle (UAV), characterized in that, The system includes: The multi-source data acquisition module is used to acquire point cloud, image and location data collected by UAV sensors. After preprocessing, the format is standardized and the timestamp is synchronized to obtain a multi-source data set. The data fusion processing module is used to adapt processing latency and computing load through dynamic resource scheduling based on the multi-source data set, determine the optimal data processing scheme and perform parallel data fusion to obtain fused data results; The data quality assessment module is used to perform real-time quality assessment on the fused data results. It calculates and judges the coverage integrity and accuracy level of the mapped area through multi-dimensional indicators to obtain the quality assessment results. The flight strategy optimization module is used to construct benchmark data based on the quality assessment results, optimize the path and generate control commands after simulation verification, and update the data acquisition strategy synchronously.

9. A drone, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images