Geological mapping method based on unmanned aerial vehicle three-dimensional modeling

Abstract

The invention discloses a geological mapping method based on unmanned aerial vehicle three-dimensional modeling, and relates to the technical field of geological survey and intelligent path planning, and the method comprises the steps: S1, extracting an initial feature data set from geological information through constructing a three-dimensional space data processing frame, decomposing the original data into a plurality of sub-region data sets, obtaining a preliminary classification geological information set, and carrying out the calculation of the preliminary classification geological information set; s2, analyzing abnormal values and boundary information in the sub-region data set according to the preliminarily classified geological information set, and determining a potential key geological feature point position set, and S3, matching the point position set with geological environment parameters according to the key geological feature point position set, recalibrating the point positions if the matching degree is lower than a preset threshold value, and if the matching degree is lower than the preset threshold value, recalibrating the point positions if the matching degree is lower than the preset threshold value. An optimized geologic characteristic distribution diagram is obtained; according to the geological mapping method based on unmanned aerial vehicle three-dimensional modeling, the accuracy and efficiency of geological survey in a complex terrain environment are effectively improved, and comprehensive technical support is provided for field geological survey.

Description

Technical Field

[0001] The present invention relates to the technical field of geological survey and intelligent path planning, and in particular to a geological mapping method based on unmanned aerial vehicle (UAV) three-dimensional modeling. Background Art

[0002] As an essential foundation for Earth science research and resource exploration, geological surveys are directly linked to the strategic needs of national resource development and environmental protection. Their accuracy and efficiency have a crucial impact on scientific decision-making. During field surveys, rationally planning sampling points and routes is crucial for improving work quality.

[0003] However, current methods often struggle to balance comprehensiveness and targetedness when dealing with complex terrain and changing geological environments, particularly with regard to data integration and dynamic adjustment. Existing solutions often rely on manual judgment and lack the ability to deeply mine and optimize complex three-dimensional spatial information in real time. This leads to uneven resource allocation and limited survey efficiency when undertaking large-scale, multi-target surveys. Against this backdrop, the field of geological survey faces significant technical challenges. The primary challenge is accurately identifying key geological boundaries and anomalous areas within complex three-dimensional spatial data. These areas are often hidden within vast amounts of information and difficult to locate quickly using traditional methods. Addressing this issue further requires intelligently planning sampling point layouts and navigation routes based on the identification results, ensuring both coverage of key areas and time efficiency. These two factors are closely intertwined: the former determines the scientific nature of the planning, while the latter directly impacts the operability of field work. Failure to effectively address these two issues will inevitably lead to inefficiencies and high costs in surveys. Summary of the Invention

[0004] The purpose of this invention is to provide a geological mapping method based on drone 3D modeling, combining the 3D modeling results with geological survey needs, intelligently identifying key geological features and optimizing sampling point layout and navigation paths.

[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a geological mapping method based on UAV 3D modeling, comprising:

[0006] S1. By constructing a three-dimensional spatial data processing framework, an initial feature dataset is extracted from geological information, and the original data is decomposed into multiple sub-regional datasets to obtain a preliminary classified geological information set;

[0007] S2. Based on the preliminary classified geological information set, analyze the outliers and boundary information in the sub-regional data set to determine the potential key geological feature point set;

[0008] S3. For a set of key geological feature points, match the set of points with geological environment parameters. If the matching degree is lower than a preset threshold, recalibrate the points to obtain an optimized geological feature distribution map.

[0009] S4. Based on the optimized geological feature distribution map, perform grid analysis on the feature distribution map to determine the initial sampling point location set;

[0010] S5. For the initial set of sampling point locations, the travel cost under the terrain environment is taken into account. If the cost of the path segment is higher than a preset threshold, an alternative path is recalculated to obtain a preliminary optimized navigation path solution.

[0011] S6. Based on the preliminary optimized navigation path plan, the real-time optimization mechanism is integrated to adjust the path. If terrain obstacles are detected, the path replanning process is triggered to determine the final navigation path.

[0012] S7. Based on the final navigation path and sampling point location set, the multi-target survey task is decomposed into multiple subtask units, the task execution priority is obtained, and a field survey execution plan is obtained;

[0013] S8. Based on the field investigation implementation plan, conduct simulation analysis on the potential risk points of the implementation plan. If the risk value exceeds the preset threshold, make partial adjustments to the plan and determine the final implementation plan of the investigation.

[0014] Preferably, the S1 includes extracting an initial feature data set from geological information by constructing a three-dimensional spatial data processing framework, decomposing the original data using a multi-level grid division method based on the data distribution characteristics under complex terrain conditions, and obtaining a sub-region data set and a classification information set; decomposing the original data into multiple sub-region data sets using a multi-level grid division method, and obtaining the classification results of local geological information based on the data distribution characteristics within the sub-region set; optimizing and adjusting the classification information using a support vector machine algorithm based on the classification results of the sub-region data sets in combination with the data distribution characteristics under complex terrain conditions, and determining a geological information set; if there is a data distribution in the geological information set, In the case of uneven distribution, the sub-region set is supplemented by data interpolation method to obtain a geological information data set with balanced distribution; based on the geological information data set with balanced distribution, the classification information is processed by spatial interpolation technology for the multi-level network structure in three-dimensional space to obtain a geological information distribution map; through the geological information distribution map, combined with the initial feature data set, it is determined whether there are abnormal data points. If there are abnormal data points, they are smoothed to obtain the corrected geological information distribution result; based on the corrected geological information distribution result, the classification information in the multi-level network is integrated by data fusion technology for the data characteristics under the terrain environment to determine the final geological information comprehensive data set.

[0015] Preferably, S2 includes using data processing tools to perform hierarchical analysis on the sub-region set based on the geological information and the results of the preliminary classification, obtaining the distribution characteristics of the outlier points and boundary information, and determining the preliminary anomaly distribution range; based on the anomaly distribution range, combined with in-depth analysis methods, local feature comparison is performed on the outlier points and boundary information to obtain an abnormal area distribution map; through the distribution map, a preset threshold is used to screen the abnormal area to obtain a set of candidate points for key features; for the candidate point set, combined with the distribution law of potential features, the support vector machine algorithm is used to classify and optimize the point set to determine the key geological feature points; if the key geological feature points are unevenly distributed in the sub-region set, the point set is adjusted by a data smoothing tool to obtain a feature point set with a balanced distribution; based on the evenly distributed feature point set, spatial mapping technology is used to associate and integrate the point set based on the spatial characteristics of the geological information to obtain a comprehensive geological feature distribution map; through the comprehensive geological feature distribution map, combined with the intermediate results of the screening process, the potential features are secondary verified to determine the final key geological feature point set.

[0016] Preferably, S3 includes constructing a feature association framework for the correspondence between the geological feature point set and the environmental parameters through data integration means to obtain preliminary correspondence degree distribution data; based on the preliminary correspondence degree distribution data, a preset threshold is used for screening, and if the correspondence degree is lower than the preset threshold, the point set is corrected to obtain a corrected point distribution data set; for the corrected point distribution data set, a data comparison tool is used to analyze the matching deviation between the point set and the environmental parameters to determine a deviation point subset; based on the point subset, a spatial adjustment method is used to calibrate the position of the point subset to obtain an adjusted point distribution structure; based on the adjusted point distribution structure, a feature association framework is used to recalculate the correspondence degree between the point set and the environmental parameters to obtain updated matching distribution data; based on the updated matching distribution data, if there are still points in the matching distribution data that do not reach the preset threshold, these points are secondary optimized through a data smoothing tool to obtain a final optimized distribution map; for the final optimized distribution map, a spatial mapping technology is used to integrate the point set and the geological features to determine a comprehensive geological feature distribution view.

[0017] Preferably, the S4 includes comparing the sampling points with the distribution map of geological features using a spatial analysis tool for the initial sampling point location set, determining the coverage ratio of the sampling points in the high-density area, and if the coverage ratio does not reach a preset threshold, adjusting the sampling points to obtain an adjusted sampling point distribution data set; calculating the correlation between the sampling points and the feature distribution based on the adjusted sampling point distribution data set, analyzing whether the point distribution in the high-density area is balanced, and if the distribution deviation exceeds a preset range, performing local optimization on the points to obtain an optimized point distribution structure; processing the feature distribution in the grid division using a spatial interpolation method for the optimized point distribution structure, obtaining regional analysis data, and determining the boundary of the density area. Boundary range; according to the boundary range of the density zone, the sampling points are secondary checked through the data screening tool to determine whether all the points fall within the boundary range. If there are points that deviate from the boundary, the positions of the points that deviate from the boundary are adjusted to obtain the corrected sampling point set; for the corrected sampling point set, the data integration framework is used to correlate and map the location data with the distribution characteristics of the geological features to obtain a comprehensive sampling point distribution view; based on the comprehensive sampling point distribution view, the sampling layout and regional analysis results are superimposed and compared through the visualization processing tool to determine the final sampling point configuration plan; for the final sampling point configuration plan, the data storage module is used to archive the location data and feature distribution information to obtain a structured sampling layout data set.

[0018] Preferably, S5 includes analyzing the matching degree between the preliminary optimized navigation path plan and the terrain using a data processing tool. If the matching degree is lower than a preset threshold, the path plan is locally corrected to obtain a corrected path distribution. Based on the corrected path distribution, the correlation between the sampling points and the navigation path is calculated. If the correlation cannot cover the target area, the point distribution is redistributed to determine the adjusted point layout. Based on the adjusted point layout, the path plan and the travel cost data are integrated using an information integration framework to obtain a comprehensive path cost distribution view. Based on the comprehensive path cost distribution view, cost segments in the navigation path that exceed a pre-examination threshold are identified using a data screening tool. If the proportion of cost segments that exceed the preset threshold exceeds a preset proportion range, alternative paths are planned for the cost segments that exceed the preset threshold to obtain an optimized path structure. Based on the optimized path structure, a spatial interpolation method is used to analyze the coverage of the path plan and the sampling points to obtain regional distribution data. Based on the regional distribution data, the data verification module is used to evaluate the conformity of the path plan with the planning requirements. If the conformity does not meet the preset standard, the path structure is fine-tuned to determine the final navigation path configuration.

[0019] Preferably, the S6 includes acquiring dynamic environmental information from the field survey scene, continuously monitoring environmental changes, and obtaining the latest terrain and obstacle data distribution; according to the terrain and obstacle data distribution, using pre-established path update rules, comparing and analyzing the navigation path information; if it is detected that a terrain obstacle affects the path passage, triggering the path adjustment process, and determining a preliminary adjusted path plan; for the preliminary adjusted path plan, calculating the degree of matching between the path information and the dynamic environmental data through a spatial analysis tool; if the matching degree is lower than a preset threshold, correcting the local segment of the path to obtain a corrected path layout; according to the corrected path layout, using information The fusion framework integrates and processes the real-time collected environmental change data with path information to obtain a comprehensive environmental adaptability distribution view. Based on the comprehensive environmental adaptability distribution view, the sections of the navigation path affected by terrain obstacles are identified. If the proportion of the identified affected sections exceeds the preset range, the replanning process is initiated to determine an alternative path solution. Based on the alternative path solution, the spatial interpolation method is used to calculate the path layout and the coverage of the field survey scene to obtain path distribution data. Based on the path distribution data, the fit between the path solution and the survey scenario requirements is evaluated. If the fit does not meet the preset standard, the path layout is adjusted to determine the final navigation path configuration.

[0020] Preferably, the S7 includes analyzing the spatial distribution relationship between the sampling location and the path using a pre-established resource allocation model for the data of the navigation path and the sampling location, and obtaining a preliminary resource allocation plan; based on the preliminary allocation plan, integrating the balance principle, evaluating the distribution uniformity of the resource allocation through a spatial analysis tool; if the distribution uniformity is lower than a preset threshold, making a local adjustment to the resource allocation, and determining the adjusted configuration layout; constructing a task allocation framework for the adjusted configuration layout, decomposing the multi-objective task into multiple subtask units, and obtaining a set of decomposed task units; and using a priority sorting method to sort the subtask units according to the set of decomposed task units. The matching degree between the task unit and the resource allocation layout is calculated to obtain the task execution priority sequence; for the task execution priority sequence, the priority sequence is compared with the overall requirements of the survey plan. If the comparison result shows that the task unit execution sequence does not meet the overall requirements, the priority sequence is adjusted to determine the optimized execution sequence; based on the optimized execution sequence, the data verification tool is used to analyze the fit between the execution sequence and the complete execution plan to obtain the final field survey execution plan; for the final field survey execution plan, the execution plan is associated with the actual distribution of the navigation path and sampling locations through the data mapping method to obtain the execution path distribution.

[0021] Preferably, the S8 includes executing a plan for field investigation, preliminarily identifying potential risk points by establishing a risk assessment framework in advance, obtaining regional distribution information, and determining a preliminary risk distribution map; based on the preliminary risk distribution map, deeply analyzing the regional distribution information, if the analysis result shows that the risk value of the area exceeds the preset threshold, extracting specific risk points through a data screening tool to obtain a risk point set; for the risk point set, dynamically simulating the influence range of each risk point through a simulation analysis tool, obtaining the boundary data of the influence range from the simulation results, and determining a list of risk influence ranges; based on the risk influence range list, adopting a local adjustment mechanism, re-planning the part of the execution plan involving risk points, adjusting the relevant paths and task allocations through a spatial distribution optimization tool, and obtaining an adjusted plan fragment.

[0022] Preferably, the S8 also includes seamlessly connecting the adjusted plan fragments with other parts of the original execution plan. If a task allocation conflict is found during the connection process, the conflict is resolved through a priority sorting tool to obtain a complete integrated plan; based on the complete integrated plan, a verification tool is used to comprehensively test the feasibility of the plan. If the test results show that there are execution obstacles in the link, an alternative path is extracted through the backup plan database to determine the final investigation plan; for the final investigation plan, the plan content is correlated and matched with the actual environmental information of the field investigation to obtain environmental adaptability feedback and obtain optimized execution guidance.

[0023] It can be seen from the above technical solution that the present invention has the following beneficial effects:

[0024] This geological mapping method based on drone 3D modeling extracts an initial feature data set from massive geological information by constructing a 3D spatial data processing framework. It then uses a multi-level gridding method to decompose the original data into multiple sub-regional data sets, applies a feature extraction algorithm to identify key geological features, and constructs a feature association model to optimize the geological feature distribution map. A sampling point layout scheme is designed based on the distribution map, and a path optimization model is constructed in conjunction with navigation path planning requirements. A real-time optimization mechanism is incorporated to cope with dynamic changes in field surveys. The present invention also constructs a task allocation model based on the principle of balanced resource allocation, decomposing multi-target survey tasks into subtask units, and embeds a scientific decision support module for risk analysis, ultimately determining an executable survey plan. This system can effectively improve the accuracy and efficiency of geological surveys in complex terrain environments, providing comprehensive technical support for field geological surveys. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0027] like Figure 1 As shown, the present invention provides a technical solution: a geological mapping method based on UAV three-dimensional modeling, comprising:

[0028] S1. By constructing a three-dimensional spatial data processing framework, an initial feature dataset is extracted from geological information, and the original data is decomposed into multiple sub-regional datasets to obtain a preliminary classified geological information set;

[0029] S2. Based on the preliminary classified geological information set, analyze the outliers and boundary information in the sub-regional data set to determine the potential key geological feature point set;

[0030] S3. For a set of key geological feature points, match the set of points with geological environment parameters. If the matching degree is lower than a preset threshold, recalibrate the points to obtain an optimized geological feature distribution map.

[0031] S4. Based on the optimized geological feature distribution map, perform grid analysis on the feature distribution map to determine the initial sampling point location set;

[0032] S5. For the initial set of sampling point locations, the travel cost under the terrain environment is taken into account. If the cost of the path segment is higher than a preset threshold, an alternative path is recalculated to obtain a preliminary optimized navigation path solution.

[0033] S6. Based on the preliminary optimized navigation path plan, the real-time optimization mechanism is integrated to adjust the path. If terrain obstacles are detected, the path replanning process is triggered to determine the final navigation path.

[0034] S7. Based on the final navigation path and sampling point location set, the multi-target survey task is decomposed into multiple subtask units, the task execution priority is obtained, and a field survey execution plan is obtained;

[0035] S8. Based on the field investigation implementation plan, conduct simulation analysis on the potential risk points of the implementation plan. If the risk value exceeds the preset threshold, make partial adjustments to the plan and determine the final implementation plan of the investigation.

[0036] Based on a constructed 3D spatial data processing framework, this method uses drones to collect high-precision 3D models of geological scenes. The raw geological information is then processed and preliminarily classified by region. Through a comprehensive analysis of anomalous geological data and boundary geological features within each subregion, potential key geological locations are automatically identified and compared with geological environmental parameters in a database to determine their plausibility and reliability. If the match is insufficient, the key locations are spatially calibrated using historical data and geological evolution models to obtain a more accurate distribution map of geological features. After obtaining the optimized distribution map, a gridding strategy is used to define the survey area, and initial sampling point locations are determined based on this. Combined with the drone's path planning module, the travel cost along the sampling path is evaluated. If the cost of a path segment exceeds a pre-set warning value, the system automatically calculates an alternative route to improve overall efficiency and safety. Furthermore, a real-time optimization mechanism dynamically adjusts the path during field sampling. Once terrain obstacles are identified, the path replanning process automatically initiates, updating the navigation path. Once the navigation path is determined, the system breaks down the overall survey task into several subtasks based on the final route and sampling point distribution. It also prioritizes task execution based on factors such as resource constraints and time windows. This ultimately creates a field survey execution plan that includes task content, sampling sequence, path planning, and resource allocation. The system then uses historical case studies and a simulation engine to simulate and assess potential risks. If risks exceed a threshold, the system dynamically adjusts local strategies to ensure the feasibility and safety of the survey plan.

[0037] This implementation method achieves efficient classification and analysis of geological information through three-dimensional modeling data collected by drones, thereby improving the accuracy of geological feature identification; through the matching and calibration mechanism of points and geological parameters, it effectively eliminates misidentified points and enhances the reliability of geological distribution maps; through path cost assessment and real-time adjustment mechanism, it significantly improves the feasibility and safety of navigation paths and reduces sampling costs; multi-task decomposition and priority sorting strategies improve task execution efficiency, while risk simulation and local adjustment strategies effectively ensure the controllability and safety of program implementation.

[0038] S1 includes constructing a three-dimensional spatial data processing framework, extracting an initial feature data set from geological information, decomposing the original data using a multi-level grid partitioning method based on the data distribution characteristics under complex terrain environments, and obtaining sub-regional data sets and classification information sets; using a multi-level grid partitioning method to decompose the original data into multiple sub-regional data sets, and obtaining the classification results of local geological information based on the data distribution characteristics within the sub-regional set; based on the classification results of the sub-regional data sets and the data distribution characteristics under complex terrain environments, using the support vector machine algorithm to optimize and adjust the classification information and determine the geological information set; if there is uneven data distribution in the geological information set Based on the situation, the sub-region set is supplemented by the data interpolation method to obtain a geological information data set with uniform distribution; based on the geological information data set with uniform distribution, the classification information is processed by spatial interpolation technology for the multi-level network structure in three-dimensional space to obtain a geological information distribution map; based on the geological information distribution map, combined with the initial feature data set, it is determined whether there are abnormal data points. If there are abnormal data points, they are smoothed to obtain the corrected geological information distribution result; based on the corrected geological information distribution result, the classification information in the multi-level network is integrated by data fusion technology for the data characteristics under the terrain environment to determine the final geological information comprehensive data set.

[0039] In this embodiment, the target area is first scanned in three dimensions by a drone to obtain original geological data containing information such as elevation, slope, lithology, and landform clues. All data are uniformly processed according to a preset coordinate system and imported into a three-dimensional spatial data processing system. The first step of the system is to hierarchically process the terrain data to construct a multi-level grid structure. Specifically, the system divides the entire area into several first-level main grids according to the undulation of the terrain. The division standard of the main grid is that the coverage range of each grid is 100 meters by 100 meters in the horizontal direction, and units with an elevation change of no more than 50 meters are classified as the same level. If the elevation change within a main grid exceeds 50 meters, the main grid is automatically subdivided into 10 by 10 sub-grids, and the size of each sub-grid is 10 meters by 10 meters for more detailed geological information collection and analysis.

[0040] Data samples within each subgrid are extracted and attribute analysis is performed. Key parameters include lithology identifiers, mineral types, structural orientations, and fracture density. For these attribute data, the system uses a statistical module to calculate basic statistics such as mean, range, and standard deviation. On this basis, adjacent data points within each subgrid are paired, the attribute differences between each pair of data are calculated, and the average of all differences is taken. If this average difference falls below a preset similarity threshold (currently set at 10% of the total attribute variation range), the subgrids are classified as the same category. This preliminary classification result serves as the initial input for support vector machine training.

[0041] The support vector machine training samples are derived from historical geological survey data. The dataset contains real samples of various known lithologies, with at least 100 samples per class. The system uses a 50-fold cross-validation technique to evaluate the performance of the trained model, with a target classification accuracy of at least 95%. If the classification accuracy meets the target, the model is used to optimize the classification results for the current subgrid. This optimization process involves fine-tuning the classification boundaries to maximize the maximum margin between adjacent classes, thereby improving classification consistency.

[0042] The support vector machine training process includes six specific steps: data preparation, feature extraction, model initialization, training execution, model validation, and parameter tuning. Each step has clear input, operational logic, and judgment criteria, as follows: First, data preparation is performed. The system extracts at least 3,000 fully annotated sample data from historical geological surveys. These samples contain information such as lithology type, stratigraphic structure, and spatial location, and cover at least five major lithology categories in the region. Each sample must have a clear classification label, that is, the lithology category to which it belongs. Samples are evenly divided by category, ensuring that each category has at least 500 instances. Second, feature extraction is performed. The system extracts a fixed number of input feature variables from each sample. In this implementation, this is set to 12 feature dimensions, including: relative position values ​​of 3D spatial coordinates, stratigraphic thickness, rock density, mineral composition ratio, porosity, color coding, weathering score, structural line inclination, fracture development index, average lithology code of adjacent samples, regional slope, and distance from the nearest fault zone. All eigenvalues ​​were uniformly normalized and linearly transformed to the range of 0 to 1 using the maximum and minimum values ​​to eliminate dimensionality effects. Third, the support vector machine model was initialized. The model used a radial basis function as the kernel function to enhance classification capabilities for nonlinear boundaries. The initial penalty parameter was set to 100, and the kernel width parameter was set to 0.1, both empirically chosen. Fourth, the training process was performed. The model was fitted using 80% of the total sample data as the training set and 20% as the validation set. The training objective was to maximize the margin and minimize the classification error. Gradient descent was used to optimize the Lagrangian function during training, with an upper limit of 10,000 iterations and an early termination criterion of less than 0.0001 in the objective function. After training, model parameters were generated, including the support vector set, Lagrangian multiplier, bias term, and decision function expression. Fifth, the model was validated. Classification predictions were performed on the remaining 20% ​​of the validation set. The predicted labels were compared with the true labels, and the classification accuracy, precision, recall, and F1 score were calculated. In this embodiment, the classification accuracy is set to be not less than 95%, the precision and recall rates are not less than 93%, and the F1 score is not less than 94% as the model passing standard. Sixth, parameter tuning. If the model does not meet the above indicators, the system will adjust the penalty parameter and the kernel width parameter in turn, use the grid search method to carry out cross validation in the parameter interval, the penalty parameter is searched between 10 and 1000, the kernel function width is searched between 0.01 and 1, and the search step size is 10 and 0.05 respectively. Select the parameter combination with the best overall performance in cross validation, retrain and verify the model until all evaluation indicators are met. After completing all the above steps, the system will solidify and save the trained model as the discriminant model in the subsequent geological classification optimization step, and the classified sub-region data will be subjected to high-precision prediction and automatic classification processing.

[0043] After completing the initial classification optimization, the system further examines the spatial density of data points within the subgrids. Specifically, the number of sampling points in each subgrid is calculated. If the number of data points in a subgrid is less than 80% of the regional average data point density, it is marked as a data-sparse area. The current average density is set at 3,000 data points per square kilometer, so subgrids with less than 2,400 data points per square kilometer are marked as sparse areas. The system then performs interpolation on these areas. The interpolation algorithm uses an inverse distance weighted method, where the interpolated target point is influenced by up to 16 neighboring known points. The weight of each point is equal to the square of the inverse of the distance from the target point. For example, if the neighboring points are 10 meters, 20 meters, and 30 meters away from the target point, their weights are 0.01, 0.0025, and 0.0011, respectively. The final interpolation result is the sum of all weights multiplied by the attribute value of the corresponding known point, and then divided by the total weight.

[0044] Subsequently, the system performs spatial interpolation on all subgrids to form a preliminary geological information distribution map. This map maps and connects the representative attribute values ​​of each subgrid in space to generate a continuous distribution image. To ensure image smoothness, the system detects outliers in all subgrids. The detection standard is: if the difference between the attribute value of a subgrid and the average value of its eight adjacent grids exceeds twice the standard deviation of the attributes of all grids in the area, it is marked as an outlier. For example, if the standard deviation is 5 and the average value is 30, grids with attribute values ​​exceeding 40 or below 20 are considered outliers. The outlier processing method is smooth substitution, that is, its value is replaced with the attribute mean of the eight adjacent grids.

[0045] Finally, to form the final comprehensive geological information dataset, the system performs multi-level grid fusion. The fusion process begins with the bottom subgrid and progresses upwards. At each fusion level, the final category of the upper grid is determined by the majority category of all the subgrids below it. If there are equal categories, the category with the larger number of sample points is selected. For example, if an upper grid has four subgrids, with a category distribution of two A and two B, and the total number of points in category A is 200 and category B is 150, then category A is selected as the representative category of the upper grid.

[0046] S2 includes using data processing tools to perform hierarchical analysis on the sub-region set based on the geological information and preliminary classification results, obtaining the distribution characteristics of outlier points and boundary information, and determining the preliminary anomaly distribution range; based on the anomaly distribution range, combined with in-depth analysis methods, local feature comparison is performed on the outlier points and boundary information to obtain an abnormal area distribution map; through the distribution map, the abnormal area is screened using a preset threshold to obtain a set of candidate points for key features; for the candidate point set, combined with the distribution law of potential features, the support vector machine algorithm is used to classify and optimize the point set to determine the key geological feature points; if the key geological feature points are unevenly distributed in the sub-region set, the point set is adjusted using a data smoothing tool to obtain a set of feature points with a balanced distribution; based on the evenly distributed feature point set, spatial mapping technology is used to associate and integrate the point set based on the spatial characteristics of the geological information to obtain a comprehensive geological feature distribution map; through the comprehensive geological feature distribution map, combined with the intermediate results of the screening process, the potential features are secondary verified to determine the final set of key geological feature points.

[0047] In this implementation, the preliminary classification results are first integrated with the original geological information to form a feature dataset encompassing 12 dimensions, including spatial coordinates, lithologic identifiers, structural plane orientation, fracture development index, weathering degree, and porosity. For each subregion, the system uses a three-dimensional cubic sliding window with a fixed side length of 30 meters. This window is gradually moved across the entire region, and statistics are collected for each data point within the window. Statistical indicators include the mean and standard deviation of the lithologic code, which serve as a benchmark for identifying outliers. For each data point within the window, the system calculates the difference between its lithologic value and the window mean. If this difference is greater than twice the window standard deviation, the data point is identified as an outlier. This threshold of two times the standard deviation is the system's default anomaly determination threshold, derived from a statistical analysis of lithologic fluctuations in over 100 different tectonic zones. It is universally applicable and effectively controls the error rate. The 30-meter side length of the sliding window is determined based on the spatial resolution of the drone's 3D modeling and the minimum structural partition size for field exploration, ensuring both local feature extraction accuracy and computational efficiency.

[0048] At the same time, the system traverses the classification results of adjacent sub-areas. If the lithologic category labels of two adjacent sub-areas are different and their spatial distance is less than 20 meters, their interface is identified as a geological boundary line. The 20-meter spatial threshold is set based on the structural belt identification spacing commonly used in geological surveys to ensure the sensitivity and accuracy of boundary identification. After completing the identification of outliers and boundaries, the system maps all outliers to a three-dimensional model according to spatial coordinates and uses a density-based clustering algorithm to analyze these outliers. The clustering algorithm takes each outlier as the center, sets a cluster radius of 30 meters, and counts the number of outliers within its radius. If the number is not less than 5, the point is considered to be in an outlier cluster area. The 30-meter clustering radius is consistent with the side length of the aforementioned sliding window, which facilitates spatial consistency processing, and the clustering judgment threshold of 5 points is obtained by analyzing the minimum density of outliers in common fault structures.

[0049] For each abnormal cluster area, the system calculates the average attribute difference between the cluster center and the edge points in the three dimensions of lithology, structural surface inclination, and weathering degree. If the difference in any dimension is greater than 20%, the cluster area is retained as a candidate key area. The 20% difference ratio is determined after manual exploration and verification of multiple typical structural areas, which can significantly distinguish between structural anomalies and general geological disturbances. All abnormal points in the candidate area are extracted as a set of candidate points and input into the trained support vector machine model for binary classification prediction. This model is a radial basis function kernel model previously trained using 3,000 labeled samples. The input is a 12-dimensional feature vector for each point, and the output is a predicted label for whether it belongs to a key geological feature point.

[0050] The points whose output of the model is a key label are collected as a key point set. Next, the system performs a spatial balance analysis on the key point set. The specific steps are to divide the entire study area into 1 square kilometer grids and count the number of key points in each grid. If the ratio of the maximum value to the minimum value exceeds 3 times, the system considers that the point distribution is uneven. The density ratio threshold of 3 times is an empirical threshold, which comes from the minimum constraint standard for point sampling balance in many geological mapping projects across the country. If it is determined to be an uneven distribution, a data smoothing operation is performed: for areas where the point density is greater than the average, some points are proportionally eliminated, with the goal of making their density no more than 120% of the average value of the entire area; for areas where the density is lower than the average, the system adds smoothing points by performing linear interpolation between each point and its 8 neighboring points until the density of the area reaches more than 80% of the average value, ensuring the stability of the point density.

[0051] After point density adjustment is complete, the system constructs a connection graph between points in three-dimensional space. For each point, the system calculates the spatial distance and attribute similarity between it and other key points within a 50-meter radius. If the spatial distance between two points is less than 50 meters and their lithologic codes are identical, a connection is established. The 50-meter connection threshold is derived from the average extension length of common faults or contact zones in actual field geological structures. All connected points form a structurally connected subgraph. The system performs a depth-first traversal of each subgraph to determine whether it meets the following three criteria: First, the spatial connectivity between all points must be at least 95% complete, meaning that each point is connected to at least five other points on average; second, the attribute difference must be less than 10%, meaning that the lithologic difference between any two connected points does not exceed 10% of the lithologic codes in the entire dataset; and third, the overall spatial trend consistency must be greater than 90%, meaning that the point extension direction is less than 10 degrees from the main structural direction in the area. These three criteria are derived from the system's logical deduction of the geological model and calculations based on a large number of measured structural line directions and point attributes.

[0052] The point set contained in any connected structural subgraph that meets the above three criteria will be used as the final set of key geological feature points, which will be output for subsequent application modules such as path planning, sample selection, and geological interpretation. This process ensures that all output points are true anomaly clusters, high-confidence geological key points with spatial consistency, structural continuity, and attribute stability, and are repeatable, verifiable, and applicable to engineering projects.

[0053] A support vector machine (SVM) algorithm is used to classify and optimize the candidate point set to accurately identify and screen key geological feature points. The steps are as follows: First, feature parameters are extracted for each point in the candidate point set. These feature parameters include, but are not limited to, the point's spatial coordinates (e.g., longitude and latitude or three-dimensional position), its corresponding geological attributes (e.g., lithology, formation thickness, mineral content, color information, reflectivity), and its local gradient relative to surrounding points. A multidimensional feature vector is constructed for each point, where all dimensional values ​​are uniformly normalized, scaling all input features to between 0 and 1 to eliminate bias introduced by different dimensions in model training. Next, a training set is constructed based on a subset of manually annotated key feature points and non-key feature points, and a SVM classifier is trained using supervised learning. A radial basis function (RBF) kernel is used during training to map the input feature space to a higher-dimensional feature space to enhance separability. Kernel parameters are set, such as a penalty factor of 100 and a kernel width of 0.5. The optimal combination with the minimum error is selected from multiple candidate values ​​using a 5-fold cross-validation method. The objective function is to minimize the weighted sum of classification error and margin maximization, and the optimal hyperplane is solved using the Lagrange multiplier method. After training, the classification model is used to predict the feature vectors of all candidate points. The model outputs the classification label of each point, i.e., key point or non-key point, along with a confidence score for each classification. The system sets a confidence threshold of 0.8, which is an empirical value based on the standard of ensuring a classification accuracy greater than 90% during multiple verification processes. Only points classified as key points with a confidence score greater than or equal to 0.8 are retained as the final key geological feature points. After classification, the system performs spatial distribution statistics on all key points and analyzes their distribution within each subregion. If the number of key points in a subregion is significantly less than 30% of the average, the feature points in that region are considered unevenly distributed, and the subsequent data smoothing step is entered. This calculation process ensures that through the supervised learning ability of the support vector machine, high-precision classification optimization of complex geological feature points can be achieved in the high-dimensional feature space, thereby accurately identifying key points with geological significance and providing high-quality basic data for subsequent spatial mapping and feature verification.

[0054] S3 includes constructing a feature association framework based on the correspondence between the geological feature point set and the environmental parameters through data integration to obtain preliminary correspondence degree distribution data; based on the preliminary correspondence degree distribution data, a preset threshold is used for screening. If the correspondence degree is lower than the preset threshold, the point set is corrected to obtain a corrected point distribution data set; for the corrected point distribution data set, the matching deviation between the point set and the environmental parameters is analyzed through a data comparison tool to determine a deviation point subset; based on the point subset, a spatial adjustment method is used to calibrate the position of the point subset to obtain an adjusted point distribution structure; based on the adjusted point distribution structure, the correspondence degree between the point set and the environmental parameters is recalculated through a feature association framework to obtain updated matching distribution data; based on the updated matching distribution data, if there are still points in the matching distribution data that do not reach the preset threshold, these points are secondary optimized through a data smoothing tool to obtain a final optimized distribution map; for the final optimized distribution map, spatial mapping technology is used to integrate the point set with the geological features to determine a comprehensive geological feature distribution view.

[0055] In this implementation, a multidimensional attribute matrix is ​​first established to determine the correspondence between a set of geological feature points and environmental parameters. Rows represent each geological feature point, and columns represent its attributes and environmental parameter values. Point attributes include 12 items, including lithology code, structural surface inclination, fracture density, weathering index, porosity, color classification, hydrological type, surface sediment type, distance to the fault zone, and historical activity level. Each attribute is obtained through drone remote sensing and manual field measurements, with consistent units and standardized values. Environmental parameters include 10 items, including stratum thickness, regional slope, vegetation cover, groundwater depth, historical earthquake density, annual precipitation, surface temperature, rock resistivity, surface erosion intensity, and surrounding structural complexity, all derived from professional databases or remote sensing inversion results. By constructing a feature association framework, the 12 attributes of each geological point are used as independent variables and the 10 environmental parameters are used as dependent variables. The local regression models are constructed separately using the point-by-point multiple linear regression method. First, the data of each point are extracted from the set of geological feature points. Each point is used as an independent regression unit. Its input variables are multiple attribute values ​​of the point itself. These attribute values ​​include lithology code, structural surface inclination, fracture density, weathering grade, porosity, hydrological type code, color classification value, fault distance, fracture strength index, stratum thickness code, structural stability index and lithology hardness score, totaling 12 items; the corresponding output variables are the environmental parameter values ​​of the spatial location of the point, including the slope value corresponding to the point, groundwater depth, vegetation coverage, average annual precipitation, seismic activity frequency, resistivity, erosion intensity, sediment type number, surface temperature and tectonic disturbance level, totaling 10 items.

[0056] Before executing point-by-point regression, the system normalizes all variables to values ​​between 0 and 1 to eliminate dimensionality effects and improve regression model stability. Then, for each point, the attribute values ​​and environmental parameter values ​​of that point and its surrounding points within its subregion are extracted to form a local sample set. The size of the local sample set is determined by the spatial density of neighboring points, with a minimum sample size of 20 and a maximum sample size of 100. If insufficient sample points can be collected within a 10-meter range, the range is expanded to 20 meters until the minimum sample size is met.

[0057] Each local sample set is used to build a multivariate linear regression model for that location. In this model, 10 regression equations are constructed, each with 12 attributes as independent variables and each of the 10 environmental parameters as dependent variables. Each equation is fitted using the least squares method. The system calculates the coefficient of determination (R-squared) for each regression equation. The R-squared value measures the ability of the regression equation to explain the variance of the dependent variable. The system uses the average of the 10 R-squared values ​​as the overall "match score" for the location, measuring the overall match between the location's attributes and the environmental parameters. If, during the modeling process, the standard error of a variable is found to be greater than 50% of its regression coefficient, indicating that the variable does not contribute sufficiently to the regression results, the system will automatically remove the variable and rebuild the model to ensure model validity and stability.

[0058] After modeling is complete, the matching scores for all points are saved and used in subsequent matching screening, correction, and optimization analysis. This regression analysis is performed independently on a point-by-point basis, offering strong local adaptability and high data interpretation accuracy. It is particularly suitable for 3D geological modeling scenarios with highly heterogeneous geological environments and complex data structures. The entire process is automated, requiring no human intervention, ensuring an objective and repeatable calculation process that meets the high requirements for geological information accuracy and environmental consistency in real-world projects.

[0059] For each model, the coefficient of determination (R-squared) of the regression was calculated as the "match score" for that location. The R-squared value indicates how well the attribute explains changes in environmental parameters, ranging from 0 to 1, with higher values ​​indicating a higher degree of match. The match scores for all locations were converted to a percentage to form a preliminary distribution of correspondence.

[0060] Based on experience and analysis of extensive historical exploration data, the correlation between geological points and environmental parameters is statistically significant when the R-squared value is greater than 0.8. Therefore, a screening threshold of 80% was set. During the screening process, points with a correspondence below 80% were identified as points for correction. The correction process first locates the set of adjacent points within a 10-meter radius of the point in space and calculates the attribute mean μ and standard deviation σ of this set. The difference Δ between the current point and μ is then compared. If Δ is greater than 50% of σ, the attribute value is corrected toward μ by 50% of Δ, and all attributes are processed item by item. After correction, the multivariate regression is re-run to calculate the new matching score and update the point data. All corrected points are recombined to form the corrected point distribution dataset.

[0061] Next, the error analysis is performed on the corrected points. The system calculates the difference between the attribute value of the point and the mean value of the environmental parameter grid cell in which it is located. If the difference is greater than 20% of the standard deviation of the environmental parameter, the point is classified as a deviation point. The spatial adjustment operation is performed with the deviation point subset as input. Specifically, the main structural direction θ of the local area where the point is located is determined, and a spatial vector is constructed by the coordinates of the 8 points around it to fit the linear trend direction. If the angle between this direction and the environmental parameter gradient direction is less than 30 degrees, it is determined to be an adjustable area. The point is offset along this direction at a ratio of 50% of the original error value. The offset is measured in meters. After the adjustment is completed, the new point spatial coordinates are formed.

[0062] The matching degree is recalculated for all adjusted points to generate updated matching degree distribution data. Points with matching degrees still below 80% enter the smoothing optimization phase. This smoothing operation uses inverse distance weighted interpolation, averaging the attribute values ​​of the eight points surrounding the target point, weighted by distance. The weights are calculated as the inverse of the distance between the target point and its neighbors, and the sum of all weights is normalized. The interpolated values ​​replace the original attribute values. After interpolation is complete, the matching degree calculation is performed again to form the final set of optimized points.

[0063] Finally, all points are remapped onto the 3D geological model to construct a comprehensive view of the distribution of geological features. During this mapping process, each point is assigned a value within a corresponding voxel according to its 3D coordinate position. If a voxel contains multiple points, the attribute value of the voxel with the highest matching degree is used as the final value. This comprehensive view supports layer-by-layer cross-section display, full-space query, and visual linkage with environmental factors. It serves as part of the geological mapping output for practical applications such as geological structure analysis, mineral prediction, and project site selection.

[0064] S4 includes using spatial analysis tools to compare the sampling points with the distribution map of geological features for the initial sampling point location set, and determining the coverage ratio of the sampling points in the high-density area. If the coverage ratio does not reach the preset threshold, the sampling points are adjusted to obtain the adjusted sampling point distribution data set; based on the adjusted sampling point distribution data set, the correlation between the sampling points and the feature distribution is calculated to analyze whether the point distribution in the high-density area is balanced. If the distribution deviation exceeds the preset range, the points are locally optimized to obtain the optimized point distribution structure; for the optimized point distribution structure, the spatial interpolation method is used to process the feature distribution within the grid division, obtain regional analysis data, and determine the boundary range of the density area. ; According to the boundary range of the density zone, the sampling points are checked twice through the data screening tool to determine whether all the points fall within the boundary range. If there are points that deviate from the boundary, the positions of the points that deviate from the boundary are adjusted to obtain the corrected sampling point set; for the corrected sampling point set, the data integration framework is used to correlate and map the location data with the distribution characteristics of geological features to obtain a comprehensive sampling point distribution view; based on the comprehensive sampling point distribution view, the sampling layout and regional analysis results are superimposed and compared through the visualization processing tool to determine the final sampling point configuration plan; for the final sampling point configuration plan, the data storage module is used to archive the location data and feature distribution information to obtain a structured sampling layout data set.

[0065] In this implementation, the system first spatially overlays the input initial sampling point location set with the geological feature distribution map, analyzing whether each sampling point is located within a key geological high-density area. High-density areas are determined by whether the number of geological feature points per unit area is greater than 150% of the average point density for the entire area. For example, if there are 30,000 key points within a 200 square kilometer area with an average point density of 150 points per square kilometer, the 150% threshold is 225 points per square kilometer. The system constructs a buffer zone with a radius of 20 meters centered on each sampling point and counts whether this zone contains areas with a density exceeding the threshold. If so, the point is considered to cover a high-density area. For all sampling points, the system calculates the proportion of the total number of sampling points located in the high-density area. If this proportion is less than 80%, the sampling point position is adjusted. The 80% coverage threshold is an empirical value derived from a sample balance analysis of nearly 50 actual exploration areas, which balances sampling representativeness and point cost control. For points that need to be adjusted, the system calculates the shortest spatial distance to the nearest high-density boundary and moves the point by 70% of the distance in that direction. The new position is recorded as the updated sampling coordinate. The 70% offset coefficient is determined based on the spatial attenuation model of the point density transition zone, which ensures that the point is close to the key area without being overly concentrated.

[0066] Based on the adjusted set of sampling points, the system divides the high-density area into equal-area grids, usually set to 10 meters by 10 meters per grid. The number of sampling points in each grid is counted, and the maximum, minimum and average values ​​are recorded. The distribution deviation ratio is calculated, that is, the difference between the maximum and minimum values ​​divided by the average value. If the ratio exceeds 30%, it is judged that the point distribution is uneven. The 30% threshold is determined by the error change trend of the accuracy of geological structure identification at different sampling densities in the simulation experiment. Exceeding this threshold will significantly reduce the connectivity of the profile. The system adjusts the point positions in the areas where the number of points in the high-density grid exceeds the standard, and moves the points no more than 10 meters toward the center of the adjacent low-density area each time. The movement gives priority to the direction with the weakest connectivity, while ensuring that it does not cross the boundary outside the low-density area.

[0067] The optimized sampling point data is fed into the spatial interpolation module, which uses inverse distance weighted interpolation to process the attribute values ​​within each grid. The interpolation algorithm sets the minimum number of influential points to 6 and the maximum number of influential points to 12, with weights calculated as the inverse square of the distance. The interpolation radius is set to 20 meters. For each target grid, an estimated value is obtained by weightedly averaging the attributes of neighboring known points, ultimately forming a continuous interpolation layer of geological attributes. The system analyzes this layer and extracts contour lines of geological feature density. The system sets a contour extraction threshold of 130% of the average density. For example, if the average density is 150 points per square kilometer, areas with a density exceeding 195 points are defined as feature boundaries. Boundary extraction uses a standard contour construction algorithm, connecting points of the same density to form a continuous closed shape. The specific calculation process is as follows: First, within the geological attribute layer generated by inverse distance weighted interpolation, the system traverses all grid cells according to a regular grid structure and extracts the interpolated value for each grid cell. This value represents the density of the geological feature points at that location. The system then calculates the average density of all interpolated values ​​for the entire layer, which is recorded as the base density value. Assuming the average density is 150 points per square kilometer, the contour extraction threshold is set at 130% of this average, or 195 points per square kilometer. Next, during the traversal process, the system identifies all grid boundaries that intersect the contour density threshold (i.e., 195 points per square kilometer). The method is as follows: for any two adjacent grid cells, if one interpolated value is above 195 and the other is below 195, the boundary is considered a contour crossing boundary. The system interpolates an intermediate point at this boundary using linear interpolation. Specifically, if two adjacent grid points are P1 and P2, and their corresponding density values ​​are D1 and D2, then the contour point P is calculated between P1 and P2 using the linear weighted ratio of the contour density points. That is, when 195 is between D1 and D2, the relative position of point P on line segment P1P2 is calculated so that the density value is 195. After calculating all interpolated points that cross the boundary, the system connects all such interpolated points into line segments based on the grid topology. Each line segment represents the direction of the contour line between the grids. The system further uses a topology tracking algorithm to construct the contour line path. The algorithm searches for each possible crossing state grid by grid and connects multiple boundary line segments into a complete path. The system checks whether each path forms a closed area based on its connectivity. If a closed loop is formed, it is marked as a complete contour line graphic. Subsequently, the system performs object-oriented structural recognition on all the formed contour line closed graphics, matches each closed graphic with the interpolation point density value it contains, and filters out those abnormal lines with an area less than 10 square grids or an internal density average value less than 195, retaining those closed areas whose internal density is greater than the contour line threshold as valid geological feature density boundaries. Finally, the system outputs the closed graphics composed of these contour lines as a vector layer, which can be used for subsequent boundary zoning, path avoidance or key area planning analysis.The calculation logic of this standard contour construction algorithm can ensure the continuity, accuracy and closure of the boundary extraction results. The system then performs a spatial overlap analysis on all sampling points and the boundary layer. If a sampling point is more than 5 meters away from the nearest boundary, it is marked as a deviation point. The 5-meter distance threshold is set based on the drone modeling error and the actual exploration operation error limit, which can ensure the feasibility of the data. For the deviated point, the system translates it along the direction to the boundary, moves it to the area 3 meters inside the boundary line, and updates its coordinates. The set of all corrected sampling points is input into the data integration framework and bound to the geological attribute data to form the final spatial attribute structured record. Finally, the system compares and displays the above-mentioned comprehensive data with the geological area analysis layer through the visualization module, and marks high-density coverage points, correction points, points inside and outside the boundary, etc. with different colors on the map to provide intuitive layout results. The system scores based on three indicators: final point coverage, balance, and boundary fit. The scoring results are stored in the database together with the final sampling point coordinates, attribute fields, grid number and other information as a structured dataset. It can be exported to CSV, GDB or GeoJSON format for subsequent sampling execution system or analysis software to call.

[0068] S5 includes analyzing the match between the preliminary optimized navigation path plan and the terrain using data processing tools. If the match falls below a preset threshold, the path plan is locally modified to obtain a modified path distribution. Based on the modified path distribution, the spatial analysis module calculates the correlation between the sampling points and the navigation path. If the correlation does not cover the target area, the point distribution is redistributed to determine the adjusted point layout. For the adjusted point layout, the path plan is integrated with the travel cost data using an information integration framework to obtain a comprehensive path cost distribution view. Based on the comprehensive path cost distribution view, a data screening tool is used to identify cost segments in the navigation path that exceed a pre-examination threshold. If the proportion of cost segments exceeding the preset threshold exceeds a preset range, alternative paths are planned for these cost segments to obtain an optimized path structure. For the optimized path structure, a spatial interpolation method is used to analyze the coverage of the path plan and the sampling points to obtain regional distribution data. Based on the regional distribution data, the path plan is evaluated for compliance with planning requirements. If the compliance does not meet the preset standard, the path structure is fine-tuned to determine the final navigation path configuration.

[0069] In this implementation, the input preliminary optimized navigation path plan is first analyzed for terrain compatibility. The system reads the path plan's 3D coordinate sequence, segments it into continuous 10-meter path units, and extracts the 3D direction vector for each segment. Simultaneously, the normal vector of the corresponding surface unit below the path segment is extracted from the digital elevation model. The system uses a vector angle calculation formula to calculate the angle θ between the path direction of each segment and its corresponding terrain normal. The angle values ​​for all path segments are averaged. If the average angle is greater than 15 degrees, the path is deemed to fail the terrain adaptation requirement. 15 degrees is the maximum adaptable slope angle measured in geological field operations based on the actual maneuverability of personnel and equipment; exceeding this value results in a sharp increase in traffic risk. The system marks all path segments with angles exceeding 20 degrees as high-risk segments, inserts control nodes before and after these segments, and reconstructs the path curve using cubic Bezier interpolation. Curvature constraints are also set to ensure that the turning radius of each segment is no less than 20 meters, ensuring that the path remains continuous and smooth in high-slope transition areas without sharp angles or turns.

[0070] After completing the local correction of the path, the system loads the current set of sampling points, and constructs a spatial buffer with a radius of 50 meters with the navigation path as the center, and checks each point to see if it falls within the buffer. Calculate the number of covered points among all sampling points. If the coverage rate is less than 80%, it is marked as insufficient coverage. The 80% coverage threshold is obtained by statistically analyzing the relationship between sampling efficiency and operation efficiency in 50 typical mapping area projects. A ratio lower than this will significantly increase the probability of operation overlap and under-collection. For uncovered points, the system calculates the Euclidean distance from the point to the nearest point on the path, and performs the shortest path projection within a range of no more than 30 meters to generate new point coordinates to ensure that the new point is in an accessible area on the path. All adjusted points are updated to form the adjusted point layout.

[0071] Subsequently, the system combines the path coordinates and surface feature data to model the path travel cost. The travel cost of each path segment is calculated by weighted superposition of three parts: the first part is the terrain slope factor, the slope value range is 0 to 45 degrees, and the system is divided into five levels with weights of 0.1, 0.2, 0.3, 0.4 and 0.5 respectively; the second part is the vegetation density factor, which obtains the NDVI value from the remote sensing image and is divided into four levels with weights of 0.1, 0.2, 0.3 and 0.4 respectively; the third part is the travel time factor, which is weighted 0.3 during peak hours and 0.1 during normal hours. The cost value of each path segment is the weighted sum of the above three factors. The system generates a travel cost heat map for all path segments and counts the proportion of path segments with cost values ​​exceeding 0.7 in the total path length. If this proportion exceeds 20%, it means that there are too many high-cost segments. 0.7 is the upper limit warning value for traffic safety, and 20% is the upper limit of the acceptable risk segment ratio. Both thresholds are set based on the unmanned equipment traffic stability assessment and historical operation data standards.

[0072] For all high-cost segments, the system creates a search grid for alternative paths within 30 meters of the segment. The A-star algorithm is used to search for the minimum-cost path, and Dijkstra's algorithm is used to correct path connectivity. Alternative paths must have consistent start and end points, a length no longer than 1.2 times that of the original path, and an average segment cost less than 0.5. Once the alternative path is created, it replaces the original high-cost segment, generating an optimized path structure. This path optimization process first identifies all high-cost segments and completes the generation and replacement of alternative paths in the following steps: First, the system performs a spatial analysis of each high-cost segment, calculating its centerline. Using the centerline as a reference, it extends the search area by 15 meters on either side, creating a search area with a total width of 30 meters. This area is divided into a regular grid with a 5-meter edge length, forming a search grid matrix. Each grid cell is assigned a travel cost value derived from a combination of multiple data sources, including terrain slope, surface material type, water distribution, and the presence of obstacles. Each factor is assigned a standard weight: slope is weighted at 0.3, surface material is weighted at 0.3, water obstacles are weighted at 0.2, and other obstacles are weighted at 0.2. This weighted calculation results in a calculated travel cost for each grid. Next, the system sets the starting and ending points of the original high-cost path segment as the fixed starting and ending points of the alternative path and invokes the A-star path search algorithm within the constructed grid. The A-star algorithm uses travel cost as a cost function and sets the Euclidean distance from the start point to the end point as a heuristic function, prioritizing grid cells with shorter expected path lengths and lower costs. During the search, the system calculates the sum of the current path's cumulative cost and the estimated value of the expected target path in real time, selecting the grid with the smallest value as the extension point. The search results in a continuous grid path that minimizes the cost. The path generated by the A-star algorithm is then subjected to connectivity correction by the Dijkstra algorithm to ensure there are no breakpoints between path segments. The Dijkstra algorithm starts from the starting point of a path and gradually reevaluates the minimum cumulative cost from the starting point to each node based on the existing path until the destination is confirmed to be optimally connected. This resolves the local optimality trap that may be caused by the heuristic function and ensures that the entire path segment is unobstructed. After the path is generated, the system performs conditional verification on the alternative path. The verification criteria include: the total path length must not exceed 1.2 times the original high-cost path segment; the average cost per grid cell in the alternative path must be less than 0.5; and the starting and ending points must completely overlap with the original path. If all these conditions are met, the system deletes the original high-cost path segment and inserts the alternative path into the original path structure. The replaced path is relabeled and participates in the next round of overall navigation path optimization evaluation. Through this calculation process, the system accurately constructs alternative paths within high-cost areas, effectively reducing overall path costs, improving path accessibility, and improving task execution efficiency.

[0073] Next, the system performs another regional coverage analysis of the paths and sampling points. The system divides the entire target area into cells, each sized 25 meters by 25 meters, and counts the number of sampling points in each cell and whether they are within the path buffer. Using the inverse distance weighted interpolation method, the coverage density of the points within the path buffer is estimated. If there are continuous areas with a total area exceeding 10% but no sampling points falling within them, they are considered blind spots. 10% is the minimum control limit for spatial coverage assessment; exceeding this limit will significantly affect the representativeness of the data. The system will add auxiliary path nodes in the direction of the blind spot and update the path structure to ensure that all areas are covered.

[0074] Finally, the system verifies the conformity between the final path structure and the planning requirements through the data verification module. Evaluation indicators include: whether the average path cost is less than 0.5; whether the proportion of high-cost segments is less than 20%; whether the blind area is less than 10%; and whether the proportion of sampling points covered by the path buffer is greater than 85%. All four indicators must be met. If not, the system activates the path structure fine-tuning mechanism to improve the indicators to the target range by fine-tuning the coordinates of the control points, adjusting the buffer direction, optimizing the turning path, and other operations. After the verification is passed, the system encapsulates all the control point coordinates, segment cost values, buffer coverage results, corresponding point association numbers and other information of the path into a structured dataset, and stores it in standardized JSON format and GDB geographic database format for direct calling of subsequent tasks such as path execution, navigation scheduling, and sample arrangement.

[0075] S6 includes acquiring dynamic environmental information from field survey scenes, continuously monitoring environmental changes, and obtaining the latest terrain and obstacle data distribution; based on the terrain and obstacle data distribution, using pre-established path update rules, comparing and analyzing the navigation path information; if it is detected that terrain obstacles affect the path, triggering the path adjustment process, and determining the preliminary adjusted path plan; for the preliminary adjusted path plan, using spatial analysis tools to calculate the degree of matching between the path information and the dynamic environmental data; if the matching degree is lower than the preset threshold, correcting the local segment of the path to obtain the corrected path layout; based on the corrected path layout, using the information fusion framework The system integrates the real-time collected environmental change data with the path information to obtain a comprehensive environmental adaptability distribution view. Based on the comprehensive environmental adaptability distribution view, it identifies the sections of the navigation path affected by terrain obstacles. If the proportion of the identified affected sections exceeds the preset range, it starts the replanning process and determines the alternative path plan. Based on the alternative path plan, it uses the spatial interpolation method to calculate the path layout and the coverage of the field survey scene to obtain path distribution data. Based on the path distribution data, it evaluates the fit between the path plan and the survey scene requirements. If the fit does not meet the preset standard, it adjusts the path layout to determine the final navigation path configuration.

[0076] In this implementation, the system first utilizes high-frequency environmental sensing units deployed within the field survey area and lidar equipment mounted on drones to continuously collect data every five seconds, including terrain elevation changes, three-dimensional obstacle distribution, vegetation coverage, surface albedo, and real-time wind speed and air pressure changes. All data is synchronously mapped to a unified three-dimensional coordinate system based on a geographic reference system and structured and stored at a grid resolution of 5 meters by 5 meters. The system then calls a path monitoring module to align the current navigation path with the batch of dynamic environmental data. The alignment method calculates the Euclidean distance between the center coordinates of each path segment and the current obstacle dataset. If the shortest distance from any path segment to an obstacle is less than 5 meters, or if the slope of the elevation grid containing the current segment increases by more than 20% compared to the initial value, the path segment is considered impassable due to obstacles or terrain changes. The 5-meter distance threshold represents the minimum avoidance space for unmanned ground equipment, and the 20% slope change represents the upper limit of acceptable change, established based on manual traffic stability experiments.

[0077] For all path segments determined to be impassable, the system initiates a path adjustment process. Specifically, the system uses the start and end points of the obstructed segment as a reference point, extending the segment forward and backward by 10 meters to create a 20-meter adjustment segment. Three new control nodes are inserted within this segment, with a fixed spacing of 3 meters. The node offsets are determined by the direction of maximum obstacle density. The system uses a cubic Bezier interpolation algorithm to replace the original path segment with a new smooth curve segment, ensuring that each control point is at least 5 meters from the nearest obstacle. The interpolated result also meets a turning radius of curvature of at least 10 meters, ensuring safe passage and operational stability. When adjusting a path segment determined to be impassable, the system performs the following curve reconstruction process based on cubic Bezier interpolation to ensure that the new path curve avoids obstacles and meets turning safety constraints: First, the system determines the start and end points of the impassable path segment and extends it forward and backward by 10 meters in each direction of the path, forming a 20-meter path adjustment segment. This adjustment segment is used to insert the new path curve. The system analyzes the distribution of obstacles within the adjustment section, calculates the gradient vectors of the obstacle density in each direction, and identifies the direction in which obstacles are concentrated. The lateral offset direction of the control node is then set based on the direction of maximum obstacle density, so that the control point is away from the area with the highest obstacle density, and the offset direction is perpendicular to the obstacle gradient direction. The system then evenly inserts three new control nodes within the path adjustment section. The longitudinal spacing between these three nodes is 3 meters, and the offset distance of the control node in the direction perpendicular to the path axis is set based on the local obstacle density, with a maximum offset value of no more than 5 meters. Before each control point is inserted, the system calculates its Euclidean distance to the nearest obstacle in real time. The control point is confirmed only when the distance is greater than 5 meters; if it is less than 5 meters, it is further offset in the opposite direction until the 5-meter condition is met or it cannot be offset. With a total of five control points, including the starting point, three control points, and the end point, the system calls the cubic Bezier interpolation algorithm for path fitting. A cubic Bézier curve segment is defined by four control points. To generate a continuous, smooth curve segment, the system fits it in two stages: the first Bézier curve segment is defined by the starting point and the first, second, and third control points. The second curve segment consists of the second and third control points and the end point. Tangent consistency and continuity conditions are applied at the second and third control points to form an overall curve with continuous first-order derivatives. After the curve is constructed, the system uses differential geometry analysis to calculate the local radius of curvature of the path curve point by point. Specifically, the first and second derivatives are calculated at the interpolation points and their ratio is calculated. All curvature radii must be greater than 10 meters. If any point has a radius smaller than this, the system automatically readjusts the position or spacing of the control points until the minimum curvature radius of the entire path segment is greater than or equal to 10 meters. Finally, the system replaces the original impassable segment with this smoothed path segment reconstructed using the Bézier curve and performs smooth transition detection and adjustment at its connection with upstream and downstream segments to ensure geometric continuity and traversability of the entire path.This method effectively combines the obstacle avoidance requirements with the minimum turning radius limit of the navigation vehicle, ensuring that the adjusted path maintains drivability while achieving smooth and safe operation.

[0078] After the system completes the initial path correction, it enters the dynamic adaptability analysis phase. The analysis process involves spatially overlaying the corrected path layout with the latest environmental data. Using a rolling calculation method with a 10-meter sliding window, it evaluates parameters such as the number and density of obstacles, the fluctuation amplitude of local slopes, and the standard deviation of wind speed changes on a segment-by-segment basis. An environmental adaptability index is generated for each segment, ranging from 0 to 1, representing impassability to full adaptability. Sections with an obstacle density exceeding 3 per 100 square meters, a slope variation exceeding 25% of the initial value, and a wind speed variation exceeding 4 meters per second will be assigned an adaptability coefficient below 0.5. If the average adaptability index for all path segments falls below 0.7, or if the proportion of sections below 0.5 exceeds 10% of the total path length, the system will trigger the path replanning process.

[0079] During the replanning process, the system uses the endpoints of the impassable segment as the starting and ending points of the new route. It constructs a grid of candidate paths within a 30-meter radius, with each grid node spaced 5 meters apart. The system searches for the lowest cost path. The cost function increases the cost by 0.1 for every 5-degree increase in slope, 0.2 for every 1-meter decrease in obstacle distance, and 0.1 for every 1-meter-per-second increase in wind speed. The system selects the lowest-cost path as the new alternative and automatically seamlessly merges it with the original path.

[0080] After completing the splicing of alternative paths, the system regenerates a buffer zone based on the new path layout and performs a spatial analysis of the mission coverage of the entire path. This method establishes a 20-meter radius buffer zone centered on the path, divided into 10-meter grids. Each grid point is then determined to be covered by the path, and the total coverage ratio is calculated. If the coverage ratio is less than 90%, or if any single uncovered area exceeds 5% of the total mission area, the path solution fails to meet the mission requirements and the system performs fine-tuning operations. Specifically, the path node positions are nudged by no more than 2 meters at a time, with the direction determined by the principle of obstacle minimization, and local interpolation and smoothing are performed.

[0081] Finally, after meeting the three indicators of path safety, environmental adaptability and task coverage, the system integrates the three-dimensional coordinates of the path nodes, the adaptability index of each segment, the obstacle interference level, the path segment travel cost, the alternative segment mark and the fine-tuning record into a structured navigation data package, and generates a JSON format for field mission terminal equipment to call, and saves it in a standard geographic database format for subsequent operation scheduling, task adjustment and path tracking.

[0082] S7 includes the data of navigation paths and sampling locations, using a pre-established resource allocation model to analyze the spatial distribution relationship between sampling locations and paths, and obtain a preliminary resource allocation plan; based on the preliminary allocation plan, the equilibrium principle is integrated and the distribution uniformity of resource allocation is evaluated through spatial analysis tools. If the distribution uniformity is lower than the preset threshold, the resource allocation is locally adjusted to determine the adjusted configuration layout; based on the adjusted configuration layout, a task allocation framework is constructed to decompose the multi-objective task into multiple subtask units and obtain a set of decomposed task units; based on the set of decomposed task units, a priority sorting method is used to sort the subtask units. The matching degree between the unit and the resource configuration layout is calculated to obtain the task execution priority sequence; for the task execution priority sequence, the priority sequence is compared with the overall requirements of the survey plan. If the comparison result shows that the execution sequence of the task unit does not meet the overall requirements, the priority sequence is adjusted to determine the optimized execution sequence; based on the optimized execution sequence, the data verification tool is used to analyze the fit between the execution sequence and the complete execution plan to obtain the final field survey execution plan; for the final field survey execution plan, the execution plan is associated with the actual distribution of the navigation path and sampling locations through the data mapping method to obtain the execution path distribution.

[0083] In this implementation, the navigation path and sampling point location data obtained from the drone's 3D modeling results are first parsed. Each sampling point is recorded in its 3D coordinates and paired with the path segment based on its spatial distance. The pairing method employed is to calculate the shortest perpendicular distance from each sampling point to the centerline of each path segment. If this distance is less than or equal to 20 meters, the sampling point is assigned to the operational coverage area of ​​that path segment. The 20-meter threshold is set based on the drone's line-of-sight control standards and data from multiple field flight coverage experiments. 20 meters represents the critical distance for stable control and effective information collection. After pairing, a preliminary resource allocation model is constructed. This involves counting the number of sampling points distributed along the path segment and then proportionally allocating task resources (including operators, sample collection equipment, battery packs, etc.). The preliminary allocation strategy is to allocate resources per path segment equal to the number of sampling points multiplied by the single-point operation resource base. If the single-point resource base is set to 2 (one person and one set of equipment) and the path segment contains four sampling points, the preliminary resource allocation is 8 units.

[0084] After establishing the initial plan, the system assesses the balance of resource allocation by calculating the standard deviation of the resource count for each route segment and dividing it by the mean resource count to determine the coefficient of variation. A coefficient of variation greater than 0.25 indicates significant variability in resource allocation across segments. This 0.25 threshold is derived from the statistical average upper bound of resource utilization fluctuations from 50 field missions. The system then divides the route segments into two groups, one with excess and one with insufficient resources. Using a maximum-minimum difference adjustment method, resources are allocated from the excess segments to the insufficient segments, moving them in increments of 2 at a time until the coefficient of variation remains within 0.25. This process ensures that all route segments receive a balanced allocation of resources, avoiding either redundant or insufficient resources.

[0085] Next, a set of task units is created for each resource-allocated path segment and sampling point combination. Each task unit contains the path segment number, a list of sampling points it covers, the number of allocated resources, the estimated task execution time (the number of sampling points multiplied by the average operation time per point, typically set to 15 minutes), and the geological anomaly intensity level of the area in which the path segment is located. Each task unit is then prioritized using a weighted sum of four scoring criteria: a geological intensity score (with 0.3, 0.6, and 1 representing low, medium, and high levels, respectively); a resource adequacy score (the ratio of actual allocation to ideal value, with values ​​exceeding 1 assigned a value of 1); a path accessibility score (the inverse of the cost ratio, calculated by combining slope and obstacle statistics, ranging from 0 to 1); and a task urgency score (a score of 1 for a task's scheduled completion time less than one day from the current time and 0.5 for a task's scheduled completion time greater than three days). These scores are standardized and weighted, with weights of 0.4, 0.2, 0.2, and 0.2, respectively, to calculate the priority score for each task unit.

[0086] After the system prioritizes the task units, it compares the ranking results with the original task execution logic to check whether the task execution sequence dependency constraints are violated. For example, certain sampling points must be performed after the completion of a specific path segment. If the ranking does not conform to this logic, the task sequence is automatically topologically sorted and adjusted according to the dependency graph to generate a new task execution order that satisfies the dependency relationship. The system then uses the execution sequence verification module to calculate the logical fit of the new sequence. The inspection indicators include logical consistency rate (the ratio of the sorting to the dependency logic), resource utilization efficiency (resource unit utilization rate), and path coherence (whether the sorted execution path is connected). The three scores are weighted to obtain the total fit score. If it is less than 0.9, the system returns to adjust. The above 0.9 threshold is the critical value for stable scheduling schemes in multi-round task scheduling simulation tests.

[0087] The finalized task execution sequence is integrated with the navigation path and the spatial locations of the sampling points. Specifically, each task unit is labeled with its execution sequence number, and its path segments and sampling point spatial locations are uniformly mapped to a standard coordinate system map layer. The execution path data includes fields such as the task sequence number, execution path segment number, a list of covered sampling point coordinates, the number of required resources, the estimated execution time, and the priority score. It is output as a structured data table and spatial vector map file for use by the UAV scheduling system, task scheduling system, and field operation system. This ensures that the task execution plan is not only logically sound but also spatially feasible, with reasonable resource distribution, and is traceable and verifiable.

[0088] S8 includes the implementation plan for field surveys, which preliminarily identifies potential risk points by establishing a risk assessment framework in advance, obtains regional distribution information, and determines a preliminary risk distribution map; based on the preliminary risk distribution map, conducts in-depth analysis of the regional distribution information; if the analysis results show that the risk value of the area exceeds the preset threshold, the specific risk points are extracted through data screening tools to obtain a risk point set; for the risk point set, the impact range of each risk point is dynamically simulated through simulation analysis tools, the boundary data of the impact range is obtained from the simulation results, and a list of risk impact ranges is determined; based on the risk impact range list, a local adjustment mechanism is used to re-plan the part of the implementation plan involving risk points, The spatial distribution optimization tool is used to adjust the relevant paths and task allocation to obtain the adjusted plan fragments; for the adjusted plan fragments, the adjusted plan fragments are seamlessly connected with the other parts of the original execution plan. If any task allocation conflicts are found during the connection process, the conflicts are resolved through the priority sorting tool to obtain the integrated complete plan; based on the integrated complete plan, the verification tool is used to comprehensively test the feasibility of the plan. If the test results show that there are execution obstacles in the link, alternative paths are extracted from the backup plan database to determine the final survey plan; for the final survey plan, the plan content is correlated and matched with the actual environmental information of the field survey to obtain environmental adaptability feedback and obtain optimized execution guidance.

[0089] In this embodiment, the system first analyzes the path segments, sampling points and operation time windows involved in the field survey execution plan, and divides the plan area into standard grids with a side length of 50 meters as risk analysis units. The system then calls the historical geological disaster database, extracts the number of disasters that occurred in each grid, and divides the number by the maximum number of occurrences in the study area to obtain the geological risk factor value. The slope risk factor is converted into a risk level by the average slope angle of each grid, which is set to 0.3 for less than 15 degrees, 0.6 for 15 to 30 degrees, and 1 for more than 30 degrees. The obstacle density factor is calculated by dividing the number of known obstacles in each grid by the area of ​​the grid and normalized to a value between 0 and 1. The meteorological fluctuation factor is the ratio of the number of days with sudden weather events in the area in the past five years to the total number of observation days. The passability factor is converted into an inverse form by the surface structure stability coefficient when the path passes through the grid. The more difficult it is to pass, the higher the score. The system assigns weights of 0.3, 0.2, 0.2, 0.2, and 0.1 to these five factors, respectively. These weights are based on expert experience and regression analysis of previous mission assessments. The weighted summation yields a comprehensive risk value for each grid, and high-risk grids are marked with a risk threshold of 0.7, determined by fitting accident data from 80 past field investigations.

[0090] The system extracts all high-risk grids and marks the overlapping sampling points and path segments as a risk point set. The spatial simulation module then uses a gridded spatial propagation model to perform risk propagation analysis, centered around each risk point. The propagation process expands outward in steps of 10 meters, updating the risk gradient value at each step until it falls below 0.05. The simulation boundary represents the actual impact range of the risk point, ultimately forming a list of impact boundaries for all risk points. Based on this boundary, the system extracts all path segments and sampling points in the mission plan that fall within this range, forming a local high-risk mission segment.

[0091] Next, the system performs local optimization and adjustments to these high-risk task segments. First, the task segments are re-divided, with the paths segmented into continuous coverage areas and sampling points grouped based on whether they are affected. Path optimization uses a shortest path replacement strategy, calling on the path entries in the risk-avoidance path library that meet the shortest path requirements and do not cross high-risk grids to replace the original paths. If a sampling point is within the risk impact boundary, the point is moved outward to the center of the nearest safe grid. The moving distance is calculated using the spatial distance calculation module, which calculates the lowest total travel cost among all possible movement paths. After the adjusted task segment is constructed, the system reallocates work resources and time windows. Work resources are calculated by allocating one person and one device to each sampling point, and two people and one device to each path segment. The time window is calculated by multiplying the number of sampling points by the sampling time of 15 minutes per point, plus the path segment travel time. The travel time is calculated by dividing the path length by the standard travel speed of 4 kilometers per hour.

[0092] The system then stitches all adjusted task segments with the unaffected task segments into a complete plan. This stitching process requires continuous path segments and no conflicting time windows. If a task time or resource overlap conflict is detected, the system re-prioritizes all task segments based on a weighted sum of geological importance (assigned a value of 1 to 3), risk avoidance difficulty (assigned a value of 1 to 3), and task urgency (ordered in reverse order of task deadline from the current time). If a segment conflicts with multiple tasks, the lower-priority task will be delayed or resource allocation adjusted.

[0093] Once a complete, integrated plan is formed, the system invokes the feasibility analysis module to check path connectivity, resource availability and timely arrival, and sampling sequence compliance with geological process logic. If any failure is detected, the system invokes the backup path database and task alternatives library to replace it until all tests pass. Finally, the adjusted path segments, sampling points, operation times, and resource numbers are structured and summarized to generate the final survey plan.

[0094] Finally, the system compares the plan's compatibility with the latest environmental data, using topographic maps, communication signal maps, and weather forecast layers to match each task unit in the plan. If a mismatch occurs, such as in a signal blind spot, heavy rainfall, or an impassable path, an alternative segment is used to adjust the plan until all task points are in a safe and stable environment. This ultimately creates a complete implementation guide, including a route map, sampling task list, resource allocation documents, risk avoidance instructions, and an environmental adaptation report, ensuring the plan's operability, safety, and scheduling rationality.

[0095] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A geological mapping method based on UAV 3D modeling, characterized in that: include: S1. By constructing a three-dimensional spatial data processing framework, an initial feature dataset is extracted from geological information, and the original data is decomposed into multiple sub-regional datasets to obtain a preliminary classified geological information set; S2. Based on the preliminary classified geological information set, analyze the outliers and boundary information in the sub-regional data set to determine the potential key geological feature point set; S3. For a set of key geological feature points, match the set of points with geological environment parameters. If the matching degree is lower than a preset threshold, recalibrate the points to obtain an optimized geological feature distribution map. S4. Based on the optimized geological feature distribution map, perform grid analysis on the feature distribution map to determine the initial sampling point location set; S5. For the initial set of sampling point locations, the travel cost under the terrain environment is taken into account. If the cost of the path segment is higher than a preset threshold, an alternative path is recalculated to obtain a preliminary optimized navigation path solution. S6. Based on the preliminary optimized navigation path plan, the real-time optimization mechanism is integrated to adjust the path. If terrain obstacles are detected, the path replanning process is triggered to determine the final navigation path. S7. Based on the final navigation path and sampling point location set, the multi-target survey task is decomposed into multiple subtask units, the task execution priority is obtained, and a field survey execution plan is obtained; S8. Based on the field investigation implementation plan, conduct simulation analysis on the potential risk points of the implementation plan. If the risk value exceeds the preset threshold, make partial adjustments to the plan and determine the final implementation plan of the investigation.

2. The geological mapping method based on UAV 3D modeling according to claim 1, characterized in that: Said S1 comprises extracting an initial feature data set from geological information by constructing a three-dimensional spatial data processing framework, decomposing the original data by a multi-level grid partitioning method in view of the data distribution characteristics under a complex terrain environment, and obtaining a sub-region data set and a classification information set; decomposing the original data into a plurality of sub-region data sets by a multi-level grid partitioning method, and obtaining the classification results of local geological information in view of the data distribution characteristics within the sub-region set; optimizing and adjusting the classification information by using a support vector machine algorithm based on the classification results of the sub-region data sets in combination with the data distribution characteristics under a complex terrain environment, and determining a geological information set; if there is uneven data distribution in the geological information set In the case of the above, the sub-region set is supplemented by the data interpolation method to obtain a geological information data set with uniform distribution; based on the uniformly distributed geological information data set, the spatial interpolation technology is used to process the classification information for the multi-level network structure in three-dimensional space to obtain a geological information distribution map; through the geological information distribution map, combined with the initial feature data set, it is determined whether there are abnormal data points. If there are abnormal data points, they are smoothed to obtain the corrected geological information distribution result; based on the corrected geological information distribution result, the data fusion technology is used to integrate the classification information in the multi-level network according to the data characteristics under the terrain environment to determine the final geological information comprehensive data set.

3. The geological mapping method based on UAV 3D modeling according to claim 1, characterized in that: The S2 includes using data processing tools to perform hierarchical analysis on the sub-region set based on the geological information and the results of the preliminary classification, obtaining the distribution characteristics of the outlier points and boundary information, and determining the preliminary anomaly distribution range; based on the anomaly distribution range, combined with in-depth analysis methods, local feature comparison is performed on the outlier points and boundary information to obtain an abnormal area distribution map; through the distribution map, a preset threshold is used to screen the abnormal area to obtain a set of candidate points for key features; for the candidate point set, combined with the distribution law of potential features, the support vector machine algorithm is used to classify and optimize the point set to determine the key geological feature points; if the key geological feature points are unevenly distributed in the sub-region set, the point set is adjusted by a data smoothing tool to obtain a feature point set with a balanced distribution; based on the evenly distributed feature point set, spatial mapping technology is used to associate and integrate the point set based on the spatial characteristics of the geological information to obtain a comprehensive geological feature distribution map; through the comprehensive geological feature distribution map, combined with the intermediate results of the screening process, the potential features are secondary verified to determine the final key geological feature point set.

4. The geological mapping method based on UAV 3D modeling according to claim 1, characterized in that: The S3 includes, by means of data integration, constructing a feature association framework for the correspondence between the geological feature point set and the environmental parameters, and obtaining preliminary correspondence degree distribution data; based on the preliminary correspondence degree distribution data, screening is performed using a preset threshold, and if the correspondence degree is lower than the preset threshold, the point set is corrected to obtain a corrected point distribution data set; for the corrected point distribution data set, using a data comparison tool, analyzing the matching deviation between the point set and the environmental parameters, and determining a deviation point subset; based on the point subset, using a spatial adjustment method, positionally calibrating the point subset to obtain an adjusted point distribution structure; based on the adjusted point distribution structure, recalculating the correspondence degree between the point set and the environmental parameters through a feature association framework to obtain updated matching distribution data; based on the updated matching distribution data, if there are still points in the matching distribution data that do not reach the preset threshold, performing secondary optimization on these points through a data smoothing tool to obtain a final optimized distribution map; based on the final optimized distribution map, using spatial mapping technology to integrate the point set with the geological features to determine a comprehensive geological feature distribution view.

5. The geological mapping method based on UAV 3D modeling according to claim 1, characterized in that: The S4 includes comparing the sampling points with the distribution map of geological features using a spatial analysis tool for the initial sampling point location set, determining the coverage ratio of the sampling points in the high-density area, and adjusting the sampling points if the coverage ratio does not reach a preset threshold to obtain an adjusted sampling point distribution data set; calculating the correlation between the sampling points and the feature distribution based on the adjusted sampling point distribution data set to analyze whether the point distribution in the high-density area is balanced; if the distribution deviation exceeds a preset range, performing local optimization on the points to obtain an optimized point distribution structure; processing the feature distribution in the grid division using a spatial interpolation method for the optimized point distribution structure, obtaining regional analysis data, and determining the boundary range of the density area. Based on the density zone boundary, the sampling points are rechecked using a data screening tool to determine whether all points fall within the boundary. If there are points that deviate from the boundary, the positions of the points that deviate from the boundary are adjusted to obtain a corrected sampling point set. For the corrected sampling point set, a data integration framework is used to correlate and map the location data with the distribution characteristics of geological features to obtain a comprehensive sampling point distribution view. Based on the comprehensive sampling point distribution view, a visualization tool is used to overlay and compare the sampling layout with the regional analysis results to determine the final sampling point configuration plan. For the final sampling point configuration plan, a data storage module is used to archive the location data and feature distribution information to obtain a structured sampling layout dataset.

6. The geological mapping method based on UAV 3D modeling according to claim 1, characterized in that: S5 includes analyzing the matching degree between the path plan and the terrain using a data processing tool for the preliminary optimization of the navigation path plan. If the matching degree is lower than a preset threshold, the path plan is partially corrected to obtain a corrected path distribution. Based on the corrected path distribution, the correlation between the sampling points and the navigation path is calculated. If the correlation cannot cover the target area, the point distribution is redistributed to determine the adjusted point layout. For the adjusted point layout, an information integration framework is used to fuse the path plan with the travel cost data to obtain a comprehensive path cost distribution view. Based on the comprehensive path cost distribution view, a data screening tool is used to identify cost segments in the navigation path that are higher than the pre-examination threshold. If the proportion of cost segments higher than the preset threshold exceeds the preset proportion range, alternative paths are planned for the cost segments higher than the preset threshold to obtain an optimized path structure. For the optimized path structure, a spatial interpolation method is used to analyze the coverage of the path plan and the sampling points to obtain regional distribution data. Based on the regional distribution data, the data verification module is used to evaluate the compliance of the path plan with the planning requirements. If the compliance does not meet the preset standard, the path structure is fine-tuned to determine the final navigation path configuration.

7. The geological mapping method based on UAV 3D modeling according to claim 1, characterized in that: S6 includes acquiring dynamic environmental information from the field survey scene, continuously monitoring environmental changes, and obtaining the latest terrain and obstacle data distribution; based on the terrain and obstacle data distribution, using pre-established path update rules, comparing and analyzing the navigation path information; if a terrain obstacle is detected that affects the path, triggering a path adjustment process, and determining a preliminary adjusted path plan; For the initially adjusted path plan, the spatial analysis tool is used to calculate the degree of match between the path information and the dynamic environment data. If the match is lower than the preset threshold, the local segment of the path is corrected to obtain the corrected path layout. Based on the revised path layout, an information fusion framework is used to integrate the real-time collected environmental change data with path information to obtain a comprehensive environmental adaptability distribution view; Based on the comprehensive environmental adaptability distribution view, sections of the navigation path affected by terrain obstacles are identified. If the proportion of identified affected sections exceeds the preset range, the replanning process is initiated to determine an alternative path plan. Based on the alternative path plan, the spatial interpolation method is used to calculate the path layout and the coverage of the field survey scene to obtain path distribution data. Based on the path distribution data, the fit between the path plan and the survey scenario requirements is evaluated. If the fit does not meet the preset standard, the path layout is adjusted to determine the final navigation path configuration.

8. The geological mapping method based on UAV 3D modeling according to claim 1, characterized in that: The S7 includes using a pre-established resource allocation model for the data of the navigation path and the sampling location, analyzing the spatial distribution relationship between the sampling location and the path, and obtaining a preliminary resource allocation plan; based on the preliminary allocation plan, integrating the balance principle, using a spatial analysis tool to evaluate the distribution uniformity of the resource allocation, if the distribution uniformity is lower than a preset threshold, locally adjusting the resource allocation, and determining the adjusted configuration layout; based on the adjusted configuration layout, constructing a task allocation framework, decomposing the multi-objective task into multiple subtask units, and obtaining a set of decomposed task units; based on the set of decomposed task units, using a priority sorting method, calculating the matching degree between the subtask units and the resource allocation layout, and obtaining a task execution priority sequence; based on the task execution priority sequence, comparing the priority sequence with the overall requirements of the survey plan, if the comparison result shows that the task unit execution sequence does not meet the overall requirements, adjusting the priority sequence to determine the optimized execution sequence; based on the optimized execution sequence, using a data verification tool, analyzing the fit between the execution sequence and the complete execution plan, and obtaining the final field survey execution plan; For the final field survey execution plan, the execution plan is associated with the actual distribution of navigation paths and sampling locations through data mapping method to obtain the execution path distribution.

9. The geological mapping method based on UAV 3D modeling according to claim 1, characterized in that: Said S8 includes executing a plan for field investigation, preliminarily identifying potential risk points by establishing a risk assessment framework in advance, obtaining regional distribution information, and determining a preliminary risk distribution map; Based on the preliminary risk distribution map, the regional distribution information is deeply analyzed. If the analysis results show that the risk value of the area exceeds the preset threshold, the specific risk points are extracted through the data screening tool to obtain the risk point set; For the risk point set, the impact range of each risk point is dynamically simulated through simulation analysis tools, the boundary data of the impact range is obtained from the simulation results, and the list of risk impact range is determined; based on the list of risk impact range, a local adjustment mechanism is adopted to re-plan the part of the execution plan involving risk points, and the relevant paths and task allocation are adjusted through spatial distribution optimization tools to obtain the adjusted plan fragments.

10. The geological mapping method based on UAV 3D modeling according to claim 9, characterized in that: The step S8 further includes seamlessly integrating the adjusted plan fragment with other parts of the original execution plan. If any task allocation conflicts are found during the integration process, the conflicts are resolved using a priority sorting tool to obtain a complete integrated plan. Based on the integrated plan, verification tools are used to conduct a comprehensive test of the plan's feasibility. If the test results show that there are implementation obstacles in any link, alternative paths are extracted from the backup plan database to determine the final investigation plan; For the final survey plan, the plan content is correlated and matched with the actual environmental information of the field survey to obtain environmental adaptability feedback and obtain optimized execution guidance.

Images