Unmanned aerial vehicle-based topographic mapping method and system, device, medium

Abstract

The present application relates to the technical field of topographic mapping, and discloses a topographic mapping method and system based on a UAV, equipment and a medium. The system comprises a topographic data acquisition module, a feature extraction module, a data decomposition module and a parameter distribution module. The topographic data acquisition module collects original topographic data of a target area with the aid of a multi-source sensor carried by a UAV; the feature extraction module performs density clustering processing on the original data to obtain a topographic feature point cloud data set; the data decomposition module separates the feature point cloud data set into high-frequency topographic detail data and low-frequency topographic contour data through adaptive wavelet decomposition; and the parameter distribution module analyzes user mapping instruction semantics to generate a task weight vector, and accordingly distributes mapping parameters between a laser scanning system and an optical imaging system. The system enhances the adaptability and comprehensive effect of topographic mapping through multi-source acquisition, feature extraction, data decomposition and dynamic parameter distribution, and can meet diversified requirements.

Description

Technical Field

[0001] This invention relates to the field of topographic mapping technology, specifically to topographic mapping methods, systems, equipment, and media based on unmanned aerial vehicles (UAVs). Background Technology

[0002] Topographic mapping is an important means of obtaining information on the Earth's surface morphology and geographic location, and it has wide applications in fields such as geological exploration, urban planning, and disaster monitoring. Traditional topographic mapping mostly relies on manual field surveying or manned aerial surveying. Manual surveying is inefficient and poses significant safety risks in complex terrain areas such as mountains and swamps. While manned aerial surveying can cover larger areas, it is limited by factors such as takeoff and landing sites and weather conditions, resulting in poor flexibility and high costs.

[0003] With the development of UAV technology, UAV topographic mapping is gradually becoming the mainstream method. Existing UAV mapping systems are usually equipped with a single sensor, such as LiDAR or optical camera, which results in a limited variety of terrain data and makes it difficult to comprehensively reflect terrain features. For example, point cloud data obtained solely by LiDAR is easily affected by occlusion in vegetated areas, leading to a decrease in data accuracy; image data obtained solely by optical camera is prone to measurement errors due to viewing angle deviations in areas with significant terrain undulations.

[0004] Existing systems lack accurate extraction and effective decomposition of terrain features during data processing. Large amounts of raw data are used directly for subsequent processing without filtering, which not only increases the complexity of data processing but may also affect the accuracy of surveying results due to redundant data. Furthermore, in terms of surveying parameter allocation, existing systems mostly use fixed parameter settings, failing to dynamically adjust them according to the user's specific surveying needs and the actual characteristics of the terrain data. This makes it difficult to balance surveying efficiency and accuracy, and cannot meet the diverse surveying needs in different scenarios. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, device, and medium for topographic mapping based on unmanned aerial vehicles (UAVs) to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a terrain mapping system based on unmanned aerial vehicles (UAVs), the system comprising:

[0007] The terrain data acquisition module is used to acquire raw terrain data of the target area through multi-source sensors carried by the UAV.

[0008] The feature extraction module is used to perform density clustering processing on the raw terrain data to generate a terrain feature point cloud dataset.

[0009] The data decomposition module is used to perform adaptive wavelet decomposition processing on the terrain feature point cloud dataset to separate high-frequency terrain detail data and low-frequency terrain contour data.

[0010] The parameter allocation module is used to parse the semantics of the surveying instructions input by the user, generate a surveying task weight vector, and allocate surveying parameters between the laser scanning system and the optical imaging system based on the high-frequency terrain detail data.

[0011] The hierarchical optimization module is used to construct a two-layer optimization model for terrain mapping based on the mapping parameter allocation results. The two-layer optimization model for terrain mapping includes a scanning accuracy optimization layer and a resource consumption optimization layer.

[0012] The offline database module is used to pre-store the optimal mapping parameter configuration set under different terrain complexity conditions. The optimal mapping parameter configuration set is generated through multi-objective offline optimization.

[0013] The interpolation calculation module is used to perform dual-domain interpolation calculation of the weight domain and the terrain domain in the offline database module based on the surveying task weight vector and the current terrain complexity, and output the dynamic surveying pole configuration.

[0014] Preferably, the feature extraction module performs density clustering processing including:

[0015] The original terrain data is subjected to elevation normalization processing to generate an elevation normalization matrix;

[0016] The sum of squared clustering errors of the elevation normalization matrix is ​​calculated based on the density clustering algorithm, and the optimal number of clusters is determined by the stable point of the error rate of change.

[0017] Initial cluster centers are generated based on the optimal number of clusters. The cluster centers are then iteratively updated until convergence, and the terrain feature point cloud dataset is output.

[0018] Preferably, the data decomposition module performs adaptive wavelet decomposition processing including:

[0019] The terrain feature point cloud dataset is input into the wavelet basis function library, and the wavelet decomposition level is dynamically selected according to the terrain undulation gradient.

[0020] The terrain feature point cloud dataset is separated into high-frequency terrain detail data and low-frequency terrain contour data by multi-layer wavelet packet decomposition.

[0021] The high-frequency terrain detail data is input into the parameter allocation module, and the low-frequency terrain contour data serves as the benchmark for three-dimensional terrain modeling.

[0022] Preferably, the parameter allocation module generates the surveying task weight vector by including:

[0023] Semantic keywords in user commands are parsed and mapped to precision weights, efficiency weights, energy consumption weights, path smoothness weights, and anti-interference weights.

[0024] The mapping result is normalized into a five-dimensional mapping task weight vector, where the sum of the weights of each dimension is 1;

[0025] Based on the spectral distribution characteristics of the high-frequency terrain detail data, the allocation ratio of laser scanning parameters and optical imaging parameters is associated with the mapping task weight vector.

[0026] Preferably, the hierarchical optimization module constructs a two-layer optimization model for terrain mapping, including:

[0027] The scanning accuracy optimization layer takes minimizing the point cloud reconstruction error as its objective function, and the constraints include scanning angle threshold and point cloud density threshold.

[0028] The resource consumption optimization layer takes the weighted minimization of UAV flight time and energy consumption as the objective function, and the constraints include battery capacity threshold and data storage threshold.

[0029] The high-frequency terrain detail data is simultaneously input into the scanning accuracy optimization layer and the resource consumption optimization layer.

[0030] Preferably, the construction of the offline database module includes:

[0031] Define the dimensions of terrain complexity, including elevation variance, mean slope, and land cover type;

[0032] Define the dimension of the weight anchor points to cover the convex hull space of the weight vector of the surveying task;

[0033] For each weighted anchor point and terrain complexity combination, the optimal mapping parameter configuration set is generated through multi-objective particle swarm optimization.

[0034] Preferably, the interpolation calculation module performs two-domain interpolation calculations including:

[0035] Based on the weight vector of the surveying task, the convex combination coefficients are solved in the convex hull of the weight anchor point to generate the weight domain interpolation result.

[0036] Based on the current terrain complexity, radial basis function interpolation is performed on the terrain condition point set corresponding to the weighted anchor point to generate terrain domain interpolation results;

[0037] The weight domain interpolation results and the terrain domain interpolation results are combined, and the dynamic mapping pole configuration is output to the parameter allocation module.

[0038] Preferably, the present invention also includes a UAV-based terrain mapping method, applied to the UAV-based terrain mapping system described above, the method comprising the following steps:

[0039] Step 1: Acquire raw terrain data of the target area using multi-source sensors mounted on the drone;

[0040] Step 2: Perform density clustering processing on the raw terrain data to generate a terrain feature point cloud dataset;

[0041] Step 3: Perform adaptive wavelet decomposition on the terrain feature point cloud dataset to separate high-frequency terrain detail data from low-frequency terrain contour data;

[0042] Step 4: Parse the semantics of the surveying instructions input by the user, generate a surveying task weight vector, and allocate surveying parameters between the laser scanning system and the optical imaging system based on the high-frequency terrain detail data.

[0043] Preferably, the present invention further includes a device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the UAV-based terrain mapping system as described above.

[0044] Preferably, the present invention further includes a medium storing a computer program that, when executed by a processor, implements the UAV-based terrain mapping system as described above.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] By utilizing multi-source sensors to acquire raw terrain data of the target area through the terrain data acquisition module, terrain information can be captured from multiple dimensions, enriching the diversity and comprehensiveness of the data. The collaborative work of multiple source sensors can effectively compensate for the shortcomings of a single sensor under specific terrain conditions. For example, the combination of laser scanning data and optical imaging data can reduce the interference of vegetation, buildings, etc., on terrain measurements, making the acquired raw data more consistent with the actual terrain features.

[0047] The feature extraction module performs density clustering on the raw terrain data to generate a terrain feature point cloud dataset. This process enables precise screening and refinement of the raw data. Density clustering can identify areas with significant terrain features based on the density of data points, eliminating a large amount of redundant data and reducing the burden on subsequent data processing. Simultaneously, focusing on key terrain feature point clouds makes subsequent analysis and processing more targeted, avoiding interference from irrelevant data in the mapping results and helping to improve data processing efficiency.

[0048] The data decomposition module employs adaptive wavelet decomposition to process the terrain feature point cloud dataset, separating high-frequency terrain detail data from low-frequency terrain contour data, thus achieving hierarchical processing of the terrain data. High-frequency data reflects subtle undulations and local features of the terrain, while low-frequency data presents the overall contour and macroscopic trend of the terrain. This decomposition method allows the system to focus on processing the corresponding data parts according to different surveying needs, providing a data foundation for subsequent detailed and general surveying.

[0049] The parameter allocation module parses the semantics of the user-input surveying commands, generates a surveying task weight vector, and allocates surveying parameters between the laser scanning system and the optical imaging system accordingly, making the surveying process more aligned with the user's actual needs. Through semantic parsing, the system can accurately understand the user's requirements for surveying accuracy, scope, and focus, and then dynamically adjust the working parameters of the two systems. For example, when the user needs to obtain detailed terrain details of a certain area, the system can allocate parameters to the laser scanning system that are more conducive to capturing details; when the user is more concerned with the overall outline of the terrain, the system can optimize the parameter settings of the optical imaging system, achieving a reasonable allocation of surveying resources, allowing the performance of the laser scanning system and the optical imaging system to be fully utilized, and synergistically improving the surveying results. Attached Figure Description

[0050] Figure 1 This is a timing diagram of the UAV-based terrain mapping system described in this invention;

[0051] Figure 2 A flowchart for density clustering processing;

[0052] Figure 3 A flowchart for adaptive wavelet decomposition processing;

[0053] Figure 4 Flowchart for constructing a two-level optimization model;

[0054] Figure 5 This is a flowchart for two-domain interpolation calculation. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Please see Figure 1This invention provides a terrain mapping system based on unmanned aerial vehicles (UAVs). The system includes: a terrain data acquisition module, a feature extraction module, a data decomposition module, and a parameter allocation module. Specific implementation details are as follows:

[0057] The terrain data acquisition module scans and images the target area using multi-source sensors mounted on the UAV, acquiring raw terrain data including 3D coordinates and reflection intensity. The feature extraction module receives the raw terrain data and executes a density clustering algorithm. This preprocessing preprocesses the raw data, applying density clustering to identify and extract the point set most representative of the terrain structure features, generating a terrain feature point cloud dataset. The data decomposition module receives the terrain feature point cloud dataset and applies adaptive wavelet decomposition. Based on the local undulation characteristics of the terrain data, this processing dynamically selects appropriate wavelet basis functions and decomposition levels, separating the input dataset into high-frequency terrain detail data containing subtle terrain variations and low-frequency terrain contour data reflecting the main terrain undulation structure through wavelet transform. The parameter allocation module receives user-input surveying commands, parses the command semantics using natural language processing technology, identifies the user's preferences for accuracy, efficiency, and energy consumption in the surveying task, and quantifies them into a multi-dimensional surveying task weight vector. Simultaneously, this module receives high-frequency terrain detail data and analyzes its spectral distribution characteristics. Based on the mapping task weight vector and the spectral characteristics of high-frequency terrain detail data, this module calculates and allocates specific operating parameters for the laser scanning system and the optical imaging system to balance the mapping effect under different task requirements.

[0058] Example 1: See Figure 2This demonstrates the operation of the hierarchical optimization module in a real-world terrain mapping task. Assuming the target area is a hilly region with sparse vegetation and exposed rock, after the UAV, equipped with multi-source sensors, completes initial data acquisition, the terrain data acquisition module obtains raw point clouds containing 3D coordinates and reflection intensity. The feature extraction module performs density clustering on this data, identifying feature point sets representing ridgelines, valley lines, and slope inflection points. The data decomposition module uses adaptive wavelet decomposition to separate the feature point cloud into high-frequency data (reflecting rock texture and small gullies) and low-frequency data (reflecting the overall morphology of the mountain). The parameter allocation module interprets the user instruction "Focus on capturing geological fissures, with a flight time of no less than 2 hours" and maps it to a weight vector [accuracy 0.4, efficiency 0.2, energy consumption 0.3, path smoothing 0.1, anti-interference 0.0]. At this point, the hierarchical optimization module is activated. The scanning accuracy optimization layer receives high-frequency terrain detail data, which shows that the target area has dense high-frequency components (corresponding to geological fissure features). This layer sets the objective function to minimize the point cloud reconstruction error in the fissure area. Constraints include: the laser scanning incident angle must not exceed 55 degrees (to avoid distortion from rock wall reflections), and the point cloud density in the fissure area must reach more than 200 points per square meter. The resource consumption optimization layer synchronously receives the same high-frequency data, and the objective function is set as minimizing the weighted sum of flight time and total energy consumption (coefficients α=0.6, β=0.4). Constraints include: the maximum battery output energy is 20000mAh, and the remaining capacity of the onboard memory is 128GB. The offline database module pre-stores the optimal parameter configuration for hilly terrain. For example, under the weighted anchor point [accuracy 0.5, efficiency 0.3, energy consumption 0.2, 0, 0], the optimal configuration corresponding to the terrain complexity (elevation variance 15.2, average slope 25°, surface cover type 2) is: laser scanning line density 120 lines / second, optical imaging resolution 0.05 meters. The interpolation calculation module first performs weight domain interpolation: the current weight vector lies within the convex hull formed by anchor points A1[0.5,0.3,0.2,0,0] and A2[0.3,0.2,0.4,0.1,0]. The convex combination coefficients are then calculated. =0.65, =0.35. Extract the parameter configuration sets corresponding to these two anchor points from the database.

[0059] During the terrain domain interpolation stage, the terrain complexity of the current area is calculated in real time: elevation variance 18.7 (calculated using the standard deviation of elevation values ​​in the feature point cloud), mean slope 28° (analyzed using triangular mesh normal vectors), and land cover type 2 (classified by reflection intensity). Based on the weighted domain interpolation results, Gaussian radial basis functions are used for terrain complexity interpolation. Using neighboring working point data stored in the database (e.g., elevation variance 16.1 / slope 26°, elevation variance 20.3 / slope 30°) as nodes, the Euclidean distance between the current terrain parameters and each node is calculated, and the node parameter configuration is fused according to the distance weight. The final output dynamic mapping pole configuration is as follows: laser scanning line density is increased to 135 lines / second (to ensure crack capture accuracy), optical imaging resolution is adjusted to 0.08 meters (to reduce data volume), scanning angle range is set to ±50° (to meet the 55-degree incident angle constraint), and flight speed is reduced to 3.5 meters / second (to balance energy consumption and point cloud density). After the configuration is transmitted to the parameter allocation module, the module makes fine adjustments based on the spatial distribution characteristics of the high-frequency data: in the northwest slope area with dense fissures, the laser scanning line density is temporarily increased to 150 lines / second; in the southeast gentle slope area with vegetation cover, it is appropriately reduced to 120 lines / second.

[0060] The UAV control system executes new parameter configurations for mapping operations. During flight, the terrain data acquisition module continuously acquires new point cloud data, and the feature extraction module updates the feature point set every 5 minutes. When flying to areas with abrupt terrain changes, the real-time calculated terrain complexity value changes (elevation variance increases to 25.3, average slope reaches 35°). The interpolation calculation module immediately triggers recalculation: maintaining the original convex combination coefficients in the weight domain, reselecting neighboring working condition nodes in the terrain domain (elevation variance 24.8 / slope 34°, elevation variance 26.1 / slope 37°), and outputting the updated dynamic configuration: scan line density increased to 145 lines / second, flight speed further reduced to 2.8 m / second, and the multispectral camera turned off to reduce energy consumption. The resource consumption optimization layer continuously monitors the system status. When it detects that the battery energy consumption rate is higher than expected, this layer sends a coordination request to the scan accuracy optimization layer. Through iterative negotiation using a two-layer model, while ensuring scanning accuracy in the fractured region, the point cloud density threshold for the smoother areas was adjusted from 200 points / m² to 180 points / m², bringing the total energy consumption back within a safe threshold. The final mapping task was completed within 1 hour and 45 minutes, with all collected data meeting the accuracy requirements for fracture analysis, and the remaining battery capacity maintained above 15%. The entire process achieved an adaptive balance between accuracy targets and resource constraints through dynamic configuration adjustments.

[0061] Example 2: See Figure 3This paper demonstrates the operation of the feature extraction module in a mine topographic mapping task. An unmanned aerial vehicle (UAV) conducts aerial surveys of the open-pit mine, and the topographic data acquisition module acquires raw point cloud data including blasting benches, transport ramps, and rock stripping surfaces, totaling approximately 1.2 million 3D points. The feature extraction module first performs elevation normalization: the lowest point in the identified area is 342.5 meters above sea level (located in a waterlogged area), and the highest point is 498.7 meters above sea level (located at the top of a rock pile), with an elevation range of 156.2 meters. Subtracting 342.5 from the elevation values ​​of all points and dividing by 156.2 generates an elevation normalization matrix within the interval [0,1]. This process eliminates the influence of absolute elevation differences on subsequent clustering, making a 10-meter-high transport ramp comparable to a 50-meter-high rock wall in the feature space. In the density clustering stage, the system sets the test range for the number of clusters K to 2 to 12, performing 10 K-means clustering iterations for each K value (to avoid local optima), and recording the sum of squared clustering errors (SSE). When K=5, the SSE is 1.24; when K=6, the SSE decreases to 1.18, with an error change rate of 4.8%; when K=7, the SSE is 1.15, with a change rate of 2.5%; and when K=8, the SSE is 1.13, with a change rate of 1.7%. The system determines that the error change rate stabilizes within the 2% threshold after K=7, therefore the optimal number of clusters is determined to be 7. The K-means++ algorithm is used to initialize the cluster centers: the first center point is selected (located on the southeast transport ramp), and subsequent center points are selected based on the point furthest from the existing centers, ultimately obtaining 7 initial centers distributed at the bottom of the mine, the rock wall, and the edge of the steps.

[0062] The iterative update process needs to be continuous. After the first iteration, all points are assigned to the nearest center based on Euclidean distance, and the coordinates of each cluster center are recalculated. The cluster center of the northwest rock wall moves 12 meters upwards, and the cluster center of the southern accumulation area moves 8 meters downwards. In the third iteration, the cluster center at the bottom of the mine pit moves less than 0.5 meters, reaching the convergence threshold. The final output terrain feature point cloud dataset contains 7 groups with a total of 8500 feature points: Cluster 1 represents the 45° blasting bench line (2100 points), Cluster 2 represents the bottom plane of the mine pit (1200 points), Cluster 3 represents the northwest vertical rock wall (1800 points), Cluster 4 represents the centerline of the transport ramp (900 points), Cluster 5 represents the ore accumulation cone (1500 points), Cluster 6 represents the drainage ditch trajectory (800 points), and Cluster 7 represents the equipment operating platform (1200 points). These feature points accurately capture the key terrain structures of the mine.

[0063] The data decomposition module receives the feature point cloud dataset and calculates the terrain undulation gradient: analyzing the elevation standard deviation within a 20×20 meter grid, the average gradient at the bottom of the mine pit is 3.8°, the average gradient of the transport ramp is 8.2°, and the average gradient of the northwest rock wall is 41.5°. Based on the gradient distribution, the Daubechies4 wavelet basis function is selected, employing a 2-level decomposition for gentle areas and a 4-level decomposition for steep areas. Through wavelet packet decomposition, the data at the bottom of the mine pit is separated into low-frequency contours (terrain undulations with wavelengths >15 meters) and a small amount of high-frequency details (gravel distribution with wavelengths 2-15 meters); the northwest rock wall is decomposed into rich high-frequency components with wavelengths of 0.5-4 meters, corresponding to the rock mass fracture network. The high-frequency terrain detail data is transmitted to the parameter allocation module, which detects three high-frequency anomaly areas: area A (middle section of the rock wall) has dense components with wavelengths of 0.8-1.2 meters, area B (step junction) has periodic components with wavelengths of 1.5-2 meters, and area C (ore pile) exhibits irregular high-frequency noise. Low-frequency topographic contour data generated a 3D skeleton model: the mine pit has a bowl-shaped structure with the deepest point located at the center; three transport ramps spiral upwards at a 12% slope; the northwest rock wall maintains a 75° inclination. This model serves as the basic framework for subsequent refined modeling. When the UAV flew over the new exploration area, the feature extraction module started real-time updates, revealing an unknown radial gully cluster in the new area. After elevation normalization, 11 potential clusters were detected. The system dynamically adjusted the clustering: the SSE change rate dropped sharply when the initial test K=10, determining the optimal number of clusters to be 10. The three newly added clusters captured the main gully ridge (wavelength 8 meters), the confluence of branch gullies (wavelength 3 meters), and the erosion deposits (wavelength 1.5 meters), respectively. The data decomposition module then increased the wavelet decomposition level of the area to 4 layers, separating the high-frequency data of millimeter-level erosion texture on the gully walls. The entire process, through adaptive processing, accurately identified the characteristic topology of the newly added geological structures.

[0064] Example 3: See Figure 4 This demonstrates that in a task involving deformation monitoring of urban building complexes, the parameter allocation module received the text instruction "Continuously track the development of external wall cracks, completing 20 hectares of coverage daily." The semantic parsing engine identified that "tracking cracks" was mapped to an accuracy weight, "completing daily" to an efficiency weight, and "continuously" implied a need for anti-interference. Using a pre-defined semantic scoring matrix, "cracks" was assigned 4.5 points for accuracy, "20 hectares" to efficiency, and "external wall" was associated with an anti-interference weight of 3 points (considering the reflective properties of building materials). Energy consumption and path smoothness each received a base score of 2 points. After normalization, a five-dimensional mapping task weight vector W=[0.29,0.26,0.13,0.13,0.19] was generated. This vector reflects the need for a balance between accuracy stability and operational efficiency in long-term monitoring.

[0065] High-frequency terrain detail data was obtained from facade scans of three high-rise buildings. The data shows that Building A exhibits high-frequency vibration components with wavelengths of 0.1-0.3 mm (38%), Building B shows crack propagation signals of 1-2 mm (45%), and Building C shows decorative layer peeling characteristics of 5-8 mm (17%). The parameter allocation module establishes a resource allocation model for laser scanning and optical imaging.

[0066]

[0067] In the formula: This represents the proportion of laser scanning resources. This indicates a precision weight value of 0.29. The high-frequency energy value associated with the crack is 0.45. The energy consumption weighting value is 0.13. The total high-frequency energy value is 1.0. (Calculated...) This means that 98% of the data acquisition resources are allocated to the laser scanning system. The specific implementation plan is as follows: a scanning frequency of 200 lines / second and a dot spacing of 0.01 meters are used in the crack area of ​​Building B; the frequency is reduced to 150 lines / second in the vibration area of ​​Building A; and 80 lines / second combined with optical imaging is used in the non-critical areas of Building C. In the daily task allocation, crack monitoring accounts for 60% of flight time, vibration analysis accounts for 30%, and rapid scanning of the remaining areas accounts for 10%.

[0068] The two-layer model of the hierarchical optimization module is activated synchronously. The scanning accuracy optimization layer sets the target as crack width measurement error ≤0.05mm, with constraints including: the angle between the scanning incident angle and the building normal ≤30° (to avoid facade data distortion), and the point cloud overlap rate in the crack area ≥80%. This layer accesses phase information from high-frequency data in real time and dynamically adjusts the scanning path: when a sudden change in crack direction is detected, cross-scan lines perpendicular to the crack direction are automatically added; during periods of strong sunlight (10:00-14:00), multi-echo recognition technology is used to compensate for point cloud loss caused by strong light interference. The resource consumption optimization layer constructs a flight time-storage joint model, considering urban airspace restrictions, with the upper limit of flight speed set at 8m / s. The model introduces a data value density factor. Defined as:

[0069]

[0070] when When the energy level falls below a threshold of 0.15 (e.g., in the decorative layer area of ​​Building C), the optical imaging-dominant mode is automatically triggered. The battery management subsystem provides real-time feedback on remaining power. When the power level drops below 40%, the resource consumption optimization layer sends a coordination request to the scanning accuracy layer: while maintaining the scanning accuracy of the core crack area, switch the vibration monitoring mode of Building A from continuous scanning to interval sampling, reducing total energy consumption by 18%. On the third day of the mission, a sudden rainfall event occurred, altering the reflectivity of the building surfaces. High-frequency data shows that rainwater generates high-frequency noise (wavelength 0.5-1mm) on the glass curtain wall of Building B, with the energy distribution changing to 25% in Area A, 50% in Area B (including noise), and 25% in Area C. The parameter allocation module recalculates. =0.82, correspondingly increasing the optical imaging weight to 18%. Hierarchical optimization model adjustments: the scanning accuracy layer enables the raindrop filtering algorithm, and the resource consumption layer increases the data transmission compression ratio to 4:1. Simultaneously, the anti-interference weight is automatically increased to 0.25, triggering scanning path optimization: avoiding the densely rained southwest facade and prioritizing scanning the less rain-affected northeast side.

[0071] Upon completion of the 15th day of monitoring, the system detected two new microcracks in Building B. The parameter allocation module automatically increased the local scan frequency to 240 lines / second based on the crack energy intensity, while the weight vector was dynamically updated. The resource consumption optimization layer initiated a special coordination effort: reducing the daily coverage area to 18 hectares and freeing up 15% of flight time for high-frequency monitoring of newly added cracks; simultaneously optimizing the charging strategy and utilizing the midday rest period for rapid recharging. This flexible mechanism allows the system to reorganize its monitoring strategy without interrupting its mission when new risk points are discovered.

[0072] The entire process utilizes weight vector-driven parameter allocation, coupled with real-time feedback from a two-layer optimization model, to achieve sustainable operation of millimeter-level deformation monitoring in complex urban environments. When metal decorative components are present on building facades, the scanning accuracy layer automatically activates polarization filtering to suppress specular reflection; when encountering temporary air traffic control, the resource consumption layer replans routes to reduce detour energy consumption. Ultimately, within a 28-day monitoring period, the system fully captured the expansion trajectories of three main cracks, with daily energy consumption controlled within 92% of the design value, verifying the adaptability of dynamic weight allocation and two-layer optimization in long-term surveying tasks.

[0073] Example 4: This example demonstrates the implementation process of building an offline database module in a volcanic geomorphological mapping project. First, the terrain complexity dimensions are defined: elevation variance, which is obtained by calculating the elevation standard deviation of the volcanic cone area, where the elevation ranges from 1200 meters to 2450 meters, with a standard deviation of 182.3; mean slope, which is calculated using the triangulation method to achieve an average slope of 38.7° on the cone slope; and surface cover type, which is coded as a combination of type 4 (exposed volcanic rock area) and type 2 (sparse vegetation in geothermal area). The weighted anchor points cover the convex hull of the five-dimensional space, and five key anchor points are selected: A1[0.6,0.1,0.2,0.1,0] (accuracy priority), A2[0.2,0.5,0.2,0.1,0] (efficiency priority), A3[0.3,0.2,0.4,0.1,0] (energy-sensitive), A4[0.4,0.3,0.1,0.1,0.1] (balanced), and A5[0.25,0.25,0.25,0.15,0.1] (uniform distribution). Each anchor point corresponds to six terrain condition combinations. The optimal mapping parameter configuration table for volcanic landforms is used to display the optimized parameter combinations for different conditions stored in the offline database module, as shown in Table 1.

[0074] Table 1: Optimal mapping parameter configuration for volcanic landforms.

[0075]

[0076] For the highly complex conditions of anchor point A1 (elevation variance 180-210, slope 40-45°, surface type 4), multi-objective particle swarm optimization was initiated. Fifty particles were initialized, each representing a set of parameter configurations: laser scanning frequency (80-220 lines / second), optical imaging resolution (0.01-0.05 meters), and flight altitude (50-150 meters). During iteration, particle fitness was evaluated using a dual-objective approach: objective 1 was the lava flow texture reconstruction error (simulated using historical data), and objective 2 was the energy consumption per flight (including scanning power consumption and data transmission power consumption). In the 20th generation population, three dominant particles emerged: particle X (line density 200 / resolution 0.015) had the smallest reconstruction error but the highest energy consumption; particle Y (line density 170 / resolution 0.022) had an 8% increase in error but a 22% decrease in energy consumption; and particle Z (line density 185 / resolution 0.018) fell between the two. After 100 iterations, the Pareto front contains 17 non-dominated solutions. Based on the project's emphasis on accuracy, the particle X configuration was selected and stored in the database. For the same terrain condition at anchor point A3 (energy-sensitive), the optimization process exhibited different characteristics. The optimal solution in the Pareto front was a line density of 110 lines / second with a resolution of 0.03 meters. At this point, energy consumption was reduced by 41% compared to the A1 condition, but reconstruction error increased by 15%. This configuration is suitable for periodic monitoring tasks, allowing for a moderate loss of accuracy to extend the monitoring cycle. When mixed surface types exist, optimization added constraints: a water reflectivity threshold limited the scanning angle. In the final special configuration, the laser scanning line density was reduced to 95 lines / second, and polarization filtering was enabled for optical imaging to suppress water surface reflection.

[0077] The database construction period lasted two weeks, covering 5 weighted anchor points × 6 terrain conditions × 3 typical solutions (accuracy-first / energy-first / balanced solutions), totaling 90 configurations. Each configuration contains 12 parameter items: in addition to basic scan parameters, it also includes derived parameters such as flight speed matching coefficients and data transmission compression levels. These configurations are stored using a spatial index structure, supporting rapid retrieval by subsequent interpolation modules.

[0078] In emergency monitoring missions of active volcanoes, this database immediately proved its worth. When a UAV detected a newly added fissure zone with a terrain complexity of elevation variance 195, slope 42°, and surface type 4, the interpolation module directly called the high-precision configuration (line density 200 / resolution 0.015) corresponding to anchor point A1, without requiring real-time optimization calculations. For newly emerging volcanic ash-covered areas (surface type code added 5), the system initiated incremental optimization: based on the nearest neighbor configuration (surface type 4), it generated ash-zone-specific parameters through constraint adjustments and dynamically expanded the new configuration to the database. This mechanism enables the offline database to continuously evolve and adapt to unforeseen mapping scenarios.

[0079] Example 5: See Figure 5The interpolation calculation module demonstrated its core function in the dynamic monitoring task of river delta topography. When the UAV system began mapping this complex area where water and land intersect, it first obtained the current mapping task weight vector W=[0.35,0.25,0.20,0.10,0.10]. The accuracy weight of 0.35 reflects the user's high requirements for monitoring river channel changes, and the efficiency weight of 0.25 corresponds to the indicator that 30 kilometers of riverbank must be scanned every day. At the same time, the topography analysis subsystem calculated the topographic complexity characteristics of the current area: elevation variance of 28.7 (reflecting the dramatic undulations of the tidal zone), average slope of 5.2° (low slope but with micro-topographical abrupt changes), and mixed surface cover type coding of 3 / 5 (intermingling of water and wetland vegetation). Weighted domain interpolation is first located in the offline database module. The current weight vector W lies within the convex hull formed by three predefined weight anchor points: anchor point A1 [0.4, 0.3, 0.2, 0.1, 0] represents a balance between accuracy and efficiency, anchor point A2 [0.3, 0.2, 0.3, 0.1, 0.1] is energy-sensitive, and anchor point A3 [0.5, 0.2, 0.1, 0.1, 0.1] is extreme accuracy-priority. By solving the linear equations, the convex combination coefficients of W with respect to these three anchor points are calculated to be 0.55, 0.30, and 0.15, respectively. This means that the current task parameter configuration should mainly refer to the setting of anchor point A1, and appropriately incorporate the characteristics of A2 and A3.

[0080] During the terrain domain interpolation stage, the system identifies four work points in the database whose current terrain complexity is closest to that of the current terrain: work point X (elevation variance 25 / slope 4° / type 3), work point Y (elevation variance 30 / slope 6° / type 5), work point Z (elevation variance 28 / slope 5° / type 3-5 mixed), and work point W (elevation variance 32 / slope 7° / type 5). An improved radial basis function interpolation method is used, assigning a weight to each work point proportional to its similarity to the current terrain. Work point Z receives the highest weight of 0.38 because its mixed type highly matches the current region; work point Y has a weight of 0.28, reflecting the similarity of elevation variance; work points X and W receive weights of 0.22 and 0.12, respectively.

[0081] The fusion process of the dual-domain interpolation results exhibits adaptive characteristics. In the southern region, where tidal channels are densely distributed, the weighted domain interpolation results dominate (coefficient 0.7), employing a high scan density configuration of A1 anchor points to ensure detailed capture of channel edges. In the northern region, covered by wetland vegetation, the topographic domain interpolation weight is increased to 0.6, adopting the strategy recommended by the working point Y—reducing laser power and enhancing optical imaging—to avoid errors caused by vegetation penetration. The final generated dynamic mapping pole configuration exhibits spatial differences: the laser scan line density in the channel area is 185 lines / second with a resolution of 0.02 meters; in the wetland area, it is adjusted to a combination of 120 lines / second and 0.04 meters; and in the transition zone, a compromise scheme of 155 lines / second and 0.03 meters is adopted. When the UAV enters the estuary sandbar area, real-time topographic analysis reveals new complexity characteristics: the elevation variance surges to 45, the average slope drops to 2.8°, and a new surface type code of sandy soil, 6, is added. The interpolation calculation module immediately initiates incremental adjustments: maintaining the original convex combination coefficients in the weight domain, and adding two temporary working points in the topographic domain (elevation variance 40 / slope 3° / type 6 and elevation variance 50 / slope 2° / type 6), quickly generating a sandbar-specific configuration through local interpolation. This configuration significantly reduces optical imaging exposure time (avoiding overexposure due to sand surface reflection) while increasing the laser scanning pulse frequency (capturing the microstructure of loose sand particles).

[0082] During the mission, the system encountered the outer bands of a typhoon, with winds increasing to level 6. The system automatically increased the anti-interference weight to 0.25, triggering an adjustment to the interpolation strategy: a temporary anchor point A4 [0.3, 0.2, 0.2, 0.1, 0.2] (anti-interference enhanced type) was added to the weight domain, and the convex combination coefficients were recalculated; a wind influence factor was added to the terrain domain, assigning higher weights to high-altitude work points. The final dynamic configuration output included wind-resistant flight mode parameters: reducing the flight altitude to 50 meters, increasing the scan line spacing by 20%, and enabling the Doppler compensation algorithm to eliminate point cloud displacement caused by wind speed. After completing the main river channel scan, unexpected mangrove expansion was discovered in the tributary area. The new land cover type (code 7) triggered a database extrapolation mechanism: based on the dielectric properties and height distribution of mangroves, a laser multiple echo recognition function was added to the nearest neighbor vegetation type (code 5) configuration, reducing the optical imaging color saturation requirements. These adjustments were achieved through real-time interpolation calculations without interrupting the mapping process.

[0083] The entire monitoring period lasted 21 days, during which the interpolation module performed 187 dynamic configuration adjustments, including 43 weight domain interpolation corrections, 92 topographic domain interpolation updates, and 52 dual-domain fusion strategy optimizations. The average response time for each adjustment was 1.7 seconds, ensuring the system consistently adapted to the complex and varied topographic features and environmental conditions of the delta region. The final dataset comprehensively recorded key changes such as main channel shift, sandbar morphological evolution, and wetland vegetation expansion. The mapping parameter configurations for all sub-regions maintained dynamic matching with topographic features and mission objectives. This adaptive mechanism based on dual-domain interpolation effectively solved the problems of uneven accuracy or resource waste caused by parameter fixation in traditional surveying systems under complex environments.

[0084] It should be noted that, in this document, relational terms such as "first" and "second" are used only 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 process, method, article, or apparatus.

[0085] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An unmanned aerial vehicle (UAV) based topographic mapping system, comprising: include: The terrain data acquisition module is used to acquire raw terrain data of the target area through multi-source sensors carried by the UAV. The feature extraction module is used to perform density clustering processing on the raw terrain data to generate a terrain feature point cloud dataset. The data decomposition module is used to perform adaptive wavelet decomposition processing on the terrain feature point cloud dataset to separate high-frequency terrain detail data and low-frequency terrain contour data. The parameter allocation module is used to parse the semantics of the surveying instructions input by the user, generate a surveying task weight vector, and allocate surveying parameters between the laser scanning system and the optical imaging system based on the high-frequency terrain detail data. The hierarchical optimization module is used to construct a two-layer optimization model for terrain mapping based on the mapping parameter allocation results. The two-layer optimization model for terrain mapping includes a scanning accuracy optimization layer and a resource consumption optimization layer. The offline database module is used to pre-store the optimal mapping parameter configuration set under different terrain complexity conditions. The optimal mapping parameter configuration set is generated through multi-objective offline optimization. The interpolation calculation module is used to perform dual-domain interpolation calculation of the weight domain and the terrain domain in the offline database module based on the surveying task weight vector and the current terrain complexity, and output the dynamic surveying pole configuration. The interpolation calculation module performs two-domain interpolation calculations, including: Based on the weight vector of the surveying task, the convex combination coefficients are solved in the convex hull of the weight anchor point to generate the weight domain interpolation result. Based on the current terrain complexity, radial basis function interpolation is performed on the terrain condition point set corresponding to the weighted anchor point to generate terrain domain interpolation results; The weight domain interpolation results and the terrain domain interpolation results are combined, and the dynamic mapping pole configuration is output to the parameter allocation module.

2. The drone-based topographic mapping system of claim 1, wherein, The feature extraction module performs density clustering processing including: The original terrain data is subjected to elevation normalization processing to generate an elevation normalization matrix; The sum of squared clustering errors of the elevation normalization matrix is ​​calculated based on the density clustering algorithm, and the optimal number of clusters is determined by the stable point of the error rate of change. Initial cluster centers are generated based on the optimal number of clusters. The cluster centers are then iteratively updated until convergence, and the terrain feature point cloud dataset is output.

3. The drone-based topographic mapping system of claim 1, wherein, The data decomposition module performs adaptive wavelet decomposition processing, including: The terrain feature point cloud dataset is input into the wavelet basis function library, and the wavelet decomposition level is dynamically selected according to the terrain undulation gradient. The terrain feature point cloud dataset is separated into high-frequency terrain detail data and low-frequency terrain contour data by multi-layer wavelet packet decomposition. The high-frequency terrain detail data is input into the parameter allocation module, and the low-frequency terrain contour data serves as the benchmark for three-dimensional terrain modeling.

4. The drone-based topographic mapping system of claim 1, wherein, The parameter allocation module generates a mapping task weight vector including: Semantic keywords in user commands are parsed and mapped to precision weights, efficiency weights, energy consumption weights, path smoothness weights, and anti-interference weights. The mapping result is normalized into a five-dimensional mapping task weight vector, where the sum of the weights of each dimension is 1; Based on the spectral distribution characteristics of the high-frequency terrain detail data, the allocation ratio of laser scanning parameters and optical imaging parameters is associated with the mapping task weight vector.

5. The UAV-based terrain mapping system according to claim 1, characterized in that, The hierarchical optimization module constructs a two-layer optimization model for terrain mapping, including: The scanning accuracy optimization layer takes minimizing the point cloud reconstruction error as its objective function, and the constraints include scanning angle threshold and point cloud density threshold. The resource consumption optimization layer takes the weighted minimization of UAV flight time and energy consumption as the objective function, and the constraints include battery capacity threshold and data storage threshold. The high-frequency terrain detail data is simultaneously input into the scanning accuracy optimization layer and the resource consumption optimization layer.

6. The drone-based topographic mapping system of claim 1, wherein, The construction of the offline database module includes: Define the dimensions of terrain complexity, including elevation variance, mean slope, and land cover type; Define the dimension of the weight anchor points to cover the convex hull space of the weight vector of the surveying task; For each weighted anchor point and terrain complexity combination, the optimal mapping parameter configuration set is generated through multi-objective particle swarm optimization.

7. A UAV-based terrain mapping method, applied to the UAV-based terrain mapping system as described in any one of claims 1 to 6, characterized in that, Includes the following steps: Step 1: Acquire raw terrain data of the target area using multi-source sensors mounted on the drone; Step 2: Perform density clustering processing on the raw terrain data to generate a terrain feature point cloud dataset; Step 3: Perform adaptive wavelet decomposition on the terrain feature point cloud dataset to separate high-frequency terrain detail data from low-frequency terrain contour data; Step 4: Parse the semantics of the surveying instructions input by the user, generate a surveying task weight vector, and allocate surveying parameters between the laser scanning system and the optical imaging system based on the high-frequency terrain detail data.

8. An apparatus, comprising: It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the steps of the UAV-based terrain mapping method as described in claim 7.

9. A medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the UAV-based terrain mapping method as described in claim 7.

Images