Lidar-based method for updating topography of a metal mine

Abstract

The present application relates to the field of surveying and geographic information technology, and discloses a kind of metal mining area topography updating method based on laser radar.The method comprises: fusing satellite SAR remote sensing image, unmanned aerial vehicle orthographic image and mining area production operation plan, and constructs topographic change probability grid model;Based on the model, identify high-probability change area and generate targeted laser radar scanning task;High-precision scanning is carried out by unmanned aerial vehicle platform, point cloud data is obtained, and denoising, registration and ground point extraction are carried out, and local update three-dimensional digital terrain reference model is carried out.The system comprises five modules of multi-source data acquisition, probability modeling, dynamic task planning, laser scanning execution and model updating.The present application reduces resource consumption through on-demand accurate scanning and closed-loop automatic updating, topographic monitoring efficiency, accuracy and timeliness.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of surveying and geographic information technology, and particularly relates to a metal mine area terrain updating method based on a laser radar. BACKGROUND

[0002] With the continuous expansion of the development scale of metal mineral resources, the terrain of the mining area frequently changes due to operations such as blasting, excavation, transportation and natural subsidence. In order to ensure safety in production, optimize mining planning and environmental monitoring, it is urgent to update and survey the terrain of the mining area with high frequency and high precision. Laser radar has become a core technical means for mining area terrain surveying due to its advantages of active detection, all-weather operation and high spatial resolution.

[0003] However, the traditional terrain updating method based on laser radar generally adopts a full-coverage scanning strategy with fixed flight lines or preset paths. Whether the terrain changes or not, the same density of data collection is performed on the entire mining area. This passive full-scene scanning mode not only wastes a large amount of computing resources and energy, but also prolongs the data acquisition and processing cycle, making it difficult to meet the stringent requirements of real-time and efficiency of modern smart mines.

[0004] The adaptive terrain updating method based on laser radar focuses on the closed-loop integration of terrain change perception and scanning behavior control. This direction aims to identify the terrain change area in real time and dynamically adjust the flight or driving parameters of the laser radar platform accordingly, realizing intelligent allocation of scanning resources in space and time dimensions. The basic principle is to use historical terrain data as a reference to quickly compare with newly acquired point clouds to generate a probability distribution map reflecting potential changes, and to drive the decision engine of the scanning system, thereby significantly reducing the overall operation load while ensuring the surveying accuracy of key areas.

[0005] In the prior art, terrain updating usually relies on post-offline processing, that is, after completing the full-area scanning, the change detection and model reconstruction are performed through post-processing software, which cannot realize feedback adjustment during data acquisition. Even if some systems introduce a preliminary change detection mechanism, they often lack deep coupling with the motion control of the scanning platform, making it impossible to adjust the scanning density or path in real time according to the change probability. In addition, the existing methods do not effectively combine edge computing capabilities for on-site data verification and quality evaluation, making it difficult to complete the whole process of detection-decision-execution-verification in a single task.

[0006] The above defects make the existing laser radar terrain updating scheme have bottlenecks in efficiency, response speed and resource utilization, especially in large-scale and high-dynamic metal mining area application scenarios, making it difficult to support intelligent and autonomous surveying requirements, and therefore an adaptive terrain updating method that can realize the integration of perception, decision and execution is urgently needed. SUMMARY

[0007] The technical problem to be solved by the present application is to provide a laser radar-based metal mine area terrain updating method to overcome the problems of low efficiency, resource waste and delayed response to sudden terrain changes caused by full-coverage repeated scanning or manually specified area scanning when monitoring the terrain of a vast mine area in the prior art.

[0008] To solve the above technical problems, the present application provides a laser radar-based metal mine area terrain updating method, which fuses low-resolution remote sensing monitoring data and mine production operation plans to construct a terrain change probability model for the entire mine area, accurately guides high-resolution laser radar scanning resources to target scanning operations in areas that have actually changed, and realizes closed-loop updating of the three-dimensional digital terrain reference model after scanning is completed.

[0009] The present application provides a laser radar-based metal mine area terrain updating method, which comprises:

[0010] Obtaining multi-source heterogeneous monitoring data covering the entire metal mine area, the multi-source heterogeneous monitoring data including periodically obtained satellite synthetic aperture radar remote sensing image data, unmanned aerial vehicle photogrammetry orthophoto data, and real-time obtained mine digital production operation plan data;

[0011] Based on the three-dimensional digital terrain reference model of the historical period, the multi-source heterogeneous monitoring data is subjected to spatio-temporal registration and fusion processing, and a terrain change probability grid model covering the metal mine area is established, each grid cell in the terrain change probability grid model having a probability value indicating the change of the terrain of the cell;

[0012] According to the terrain change probability grid model, identify and extract the grid cell cluster with a probability value higher than the preset change determination threshold, and generate one or more target area polygons to be scanned;

[0013] For each target area polygon to be scanned, generate a laser radar scanning task instruction containing its geographic coordinates, range, and scanning accuracy requirements, and issue the task instruction to the laser radar scanning execution platform;

[0014] The laser radar scanning execution platform autonomously navigates to the location of the target area polygon to be scanned according to the received task instruction, and performs high-resolution three-dimensional laser scanning to obtain the current three-dimensional point cloud data of the area;

[0015] The obtained current three-dimensional point cloud data is processed and seamlessly fused into the three-dimensional digital terrain reference model of the historical period to generate an updated three-dimensional digital terrain reference model of the current period.

[0016] As an embodiment of the present application, the acquisition of multi-source heterogeneous monitoring data specifically includes:

[0017] Through the interface with the satellite data service provider, time-series synthetic aperture radar interferometry data sets covering the mining area range with a specific revisit period are acquired;

[0018] The unmanned aerial vehicle aerial survey platform is scheduled to fly periodically along the preset route to collect high-resolution optical images, and digital orthophoto maps are generated through photogrammetry technology;

[0019] The application program interface is accessed to the mining area production management information system to obtain structured production operation plans containing the time, spatial position, influence range, etc. of blasting, mining, and soil removal operations in the future monitoring period.

[0020] As an embodiment of the present application, the establishment of the terrain change probability grid model specifically includes:

[0021] First, the satellite synthetic aperture radar remote sensing image data is subjected to coherence change detection processing, and the interference coherence coefficient value between two consecutive period images is calculated to generate an initial change intensity map representing the fine deformation of the ground surface;

[0022] Second, the current unmanned aerial vehicle aerial survey orthophoto image data is pixel-level registered with the digital orthophoto map of the previous period, and is input into a pre-trained twin convolutional neural network for difference analysis. The network is trained to distinguish between real terrain changes and temporary ground cover changes such as vehicles and equipment, and outputs a ground feature change classification mask identified as mining, filling, or stable;

[0023] Third, the mining area digital production operation plan data is analyzed, the operation area specified in the plan is mapped to the geographic coordinate system, and a prior change probability value is assigned according to the operation type to form a prior probability map of the operation plan;

[0024] Finally, a Bayesian probability fusion framework is established, the prior probability map of the operation plan is taken as prior information, the initial change intensity map and the ground feature change classification mask are taken as observation evidence, pixel-level probability update calculation is performed, and finally the terrain change probability grid model is generated.

[0025] As an embodiment of the present application, the generation of the polygon of the target area to be scanned according to the terrain change probability grid model specifically includes:

[0026] A global change determination threshold is applied to the terrain change probability grid model for binary processing, and the grid with a probability value higher than the threshold is marked as a change grid, and the grid with a probability value less than the threshold is marked as a stable grid.

[0027] performing a density-based spatial clustering of noise algorithm on all changed grids to aggregate spatially adjacent changed grids into one or more changed region clusters;

[0028] calculating the minimum bounding polygon of each changed region cluster and taking the polygon as the target region polygon to be scanned.

[0029] As an embodiment of the present application, the generation and delivery of the laser radar scanning task instruction specifically comprises:

[0030] According to the area size, terrain complexity and average probability value in the terrain change probability grid model of each target region polygon to be scanned, a scanning point cloud density index is dynamically set for it;

[0031] querying the real-time state database of the laser radar scanning execution platform to obtain the current position, remaining power and available load information of each platform;

[0032] constructing a multi-objective optimization function, the optimization objective of which is to minimize the total completion time of all scanning tasks, and the constraint condition is the endurance and load limit of the platform;

[0033] solving the multi-objective optimization function by using a genetic algorithm or an ant colony optimization algorithm to obtain the optimal task allocation scheme and flight path planning of each platform, generate a machine executable task description file, and deliver it to the corresponding laser radar scanning execution platform through a wireless data link.

[0034] As an embodiment of the present application, the laser radar scanning execution platform is a multi-rotor unmanned aerial vehicle equipped with a high-precision laser radar scanner, an inertial navigation system and a real-time dynamic differential positioning module.

[0035] As an embodiment of the present application, the processing of the obtained current three-dimensional point cloud data and the fusion into the three-dimensional digital terrain reference model specifically comprises:

[0036] performing denoising filtering on the laser radar scanning original point cloud data to eliminate noise points caused by atmospheric suspended matter or aircraft vibration;

[0037] using the airborne inertial navigation system data and the differential correction data of the ground base station to solve the absolute geographic coordinates of the point cloud, realizing accurate geographic registration of the point cloud;

[0038] classifying the point cloud into ground points and non-ground points, and extracting pure ground point cloud;

[0039] In the three-dimensional digital terrain reference model of the historical period, the region data covered by the target region polygon to be scanned is marked and deleted;

[0040] The irregular triangle network construction algorithm or the Kriging interpolation algorithm is used to inlay the extracted pure ground point cloud into the deleted area of the three-dimensional digital terrain reference model, local accurate updating of the model is realized, and a three-dimensional digital terrain reference model of a current period is formed.

[0041] According to another aspect of the present application, a metal mine area terrain updating system based on a laser radar is also provided, which comprises:

[0042] A multi-source heterogeneous data acquisition module is configured to acquire periodic satellite synthetic aperture radar remote sensing image data, unmanned aerial vehicle photogrammetry orthographic image data and real-time mine area digitized production operation plan data covering the entire metal mine area;

[0043] A terrain change probability modeling module is connected with the multi-source heterogeneous data acquisition module and is configured to perform spatio-temporal registration and fusion processing on the acquired multi-source heterogeneous monitoring data based on a three-dimensional digital terrain reference model of a historical period, and establish a terrain change probability grid model covering the metal mine area;

[0044] A dynamic scanning task planning module is connected with the terrain change probability modeling module and is configured to identify and generate one or more to-be-scanned target area polygons according to the terrain change probability grid model, and plan laser radar scanning task instructions containing geographic coordinates, scanning parameters and an execution platform for the to-be-scanned target area polygons;

[0045] A high-precision laser radar scanning execution module is wirelessly connected with the dynamic scanning task planning module and is configured to receive and execute the scanning task instructions, perform three-dimensional laser scanning on the specified to-be-scanned target area polygons, and acquire current three-dimensional point cloud data of the area;

[0046] A three-dimensional terrain model updating module is connected with the high-precision laser radar scanning execution module and is configured to process the acquired current three-dimensional point cloud data, seamlessly fuse the current three-dimensional point cloud data into the three-dimensional digital terrain reference model of the historical period, generate an updated three-dimensional digital terrain reference model of a current period, and provide the updated model to the terrain change probability modeling module as a reference for a next period.

[0047] As an embodiment of the present application, the terrain change probability modeling module internally comprises:

[0048] A synthetic aperture radar data processing unit is configured to perform coherence change detection processing on satellite remote sensing image data;

[0049] An optical image analysis unit internally has a twin convolutional neural network and is configured to perform difference analysis on unmanned aerial vehicle orthographic image data and output a feature change classification mask;

[0050] a production plan analysis unit for converting the structured production operation plan into a priori change probability map;

[0051] a Bayesian probability fusion unit for fusing the outputs of the three units to generate a final terrain change probability raster model.

[0052] As an embodiment of the present application, the dynamic scanning task planning module internally includes:

[0053] a change area extraction unit for threshold processing and spatial clustering of the terrain change probability raster model to generate a polygon of the target area to be scanned;

[0054] a task parameter and resource matching unit internally including a multi-objective optimization algorithm solver for generating an optimal task allocation scheme and path planning according to the attributes of the area to be scanned and the real-time state of the scanning platform.

[0055] Compared with the prior art, the present application has the beneficial effects that:

[0056] 1. By constructing a terrain change probability model, the scanning resources are changed from indiscriminate full-area coverage to efficient on-demand precise investment, reducing the time, energy consumption and data storage cost of invalid scanning operations, and greatly improving the overall operation efficiency of terrain updating;

[0057] 2. The two heterogeneous data sources of remote sensing images and mine production plans are fused, and the probability is fused through the Bayesian framework, which not only can detect the terrain changes that have occurred, but also can predict the planned operation area, realize the comprehensive and accurate identification of expected changes and unexpected changes, and improve the accuracy and reliability of change detection;

[0058] 3. An automated closed-loop operation process from change detection, task planning, data acquisition to model updating is established, which maximizes the reduction of manual intervention, ensures the timeliness and continuity of terrain data updating, and provides high-time decision support for mine safety production, dynamic adjustment of mining and stripping plans, and accurate measurement of earthwork volume;

[0059] 4. Through dynamic scanning task planning and resource optimization allocation, efficient collaborative scheduling of multiple laser radar scanning execution platforms is realized, ensuring that all change areas can be monitored and covered in the shortest time under limited scanning resources, further improving the response speed and processing capacity of the system. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 is the overall technical scheme architecture diagram of the metal mine terrain updating method based on laser radar proposed by the present application;

[0061] Figure 2 is the core principle framework diagram of the topographic change probability grid model construction in the application;

[0062] Figure 3 is the logical flow framework diagram of multi-source heterogeneous monitoring data fusion and spatio-temporal registration in the application;

[0063] Figure 4 is the logical flow framework diagram of dynamic scanning task planning and multi-platform collaborative scheduling in the application;

[0064] Figure 5 is the multi-level interaction relationship and data flow diagram of high-precision laser radar scanning execution and three-dimensional terrain model closed-loop update in the application; DETAILED DESCRIPTION

[0065] Please refer to Figures 1 to 5 The application provides a metal mine area topographic updating method based on a laser radar, aiming to solve the technical problems of quickly and accurately identifying and positioning the topographic change area in a vast metal mine area, and realizing dynamic optimization allocation of scanning resources.

[0066] The method fuses low-resolution remote sensing monitoring data and mine production operation plans to construct a topographic change probability model of the whole mine area, so as to accurately guide high-resolution laser radar scanning resources to target scanning operation in the area actually changed, and realize closed-loop update of the three-dimensional digital terrain reference model after scanning.

[0067] The method comprises the following steps:

[0068] S1, acquiring multi-source heterogeneous monitoring data covering the whole metal mine area;

[0069] S2, based on the three-dimensional digital terrain reference model of the historical period, performing spatio-temporal registration and fusion processing on the multi-source heterogeneous monitoring data, and establishing a topographic change probability grid model covering the metal mine area;

[0070] S3, according to the topographic change probability grid model, identifying and extracting a grid cell cluster with a probability value higher than a preset change determination threshold, and generating one or more scanning target area polygons;

[0071] S4, for each scanning target area polygon, generating a laser radar scanning task instruction containing its geographic coordinates, range, scanning accuracy requirement, and issuing the task instruction to a laser radar scanning execution platform;

[0072] S5, the lidar scanning execution platform autonomously navigates to the location of the polygon of the target area to be scanned according to the received task instructions, and performs high-resolution three-dimensional laser scanning to obtain the current three-dimensional point cloud data of the area;

[0073] S6. Process the acquired current 3D point cloud data and seamlessly integrate it into the historical period's 3D digital terrain benchmark model to generate an updated 3D digital terrain benchmark model for the current period.

[0074] In step S1, multi-source heterogeneous monitoring data covering the entire metal mining area is acquired. The multi-source heterogeneous monitoring data includes periodically acquired satellite synthetic aperture radar remote sensing image data, UAV aerial survey orthophoto data, and real-time acquired mining area digital production operation plan data.

[0075] Specifically, through a data interface established with a satellite data service provider, a time-series synthetic aperture radar interferometry dataset covering the mining area is acquired periodically. This dataset has a fixed revisit period, typically 7 to 14 days, and a spatial resolution of 5 to 10 meters, and is used to capture millimeter-level deformation information of the Earth's surface.

[0076] Simultaneously, one or more UAV aerial survey platforms equipped with high-resolution optical cameras are dispatched to conduct periodic patrols of the mining area according to preset routes. The flight altitude is set at 300m to 500m, the forward overlap rate is greater than 80%, the lateral overlap rate is greater than 60%, and optical images with a ground sampling distance of better than 0.1m are collected. The current period's digital orthophoto map is generated through photogrammetry software.

[0077] In addition, by accessing the mining area production management information system through the application programming interface, structured production operation plan data can be obtained in real time. This data includes fields such as the start and end time, central geographic coordinates, radius of influence, operation type, and expected earthwork volume of all blasting, mining, and spoil disposal operations within the next monitoring cycle.

[0078] In step S2, based on the historical periodic three-dimensional digital terrain benchmark model, the multi-source heterogeneous monitoring data is spatiotemporally registered and fused to establish a terrain change probability raster model covering the metal mining area. This terrain change probability raster model has a spatial resolution of 10m, and each raster cell stores a floating-point value between 0 and 1, representing the probability of terrain change occurring in that cell during the current monitoring period. The specific process of establishing this model includes four sub-steps.

[0079] First, the satellite synthetic aperture radar (SAR) remote sensing image data is processed to detect changes in coherence. SAR image pairs acquired from two consecutive monitoring periods are selected, and their interferometric coherence coefficients are calculated using the following formula:

[0080] ;

[0081] wherein, and represent the complex radar echo signals of the first and second periods respectively, denotes the mean operation within the local window. The coherence coefficient The lower the value of the coherence coefficient, the greater the change in the surface scattering characteristics of the region, and thus the inference of the presence of topographic disturbances.

[0082] The calculated coherence coefficient map is normalized to map the initial change intensity map, with a pixel value range of 0-1.

[0083] Secondly, the current UAV photogrammetry orthographic image data and the digital orthographic image map of the previous period are pixel-level registered. The affine transformation model based on feature points is used for registration, the scale-invariant feature transformation algorithm is selected to extract stable corner points in the two periods of images, the random sample consensus algorithm is used to remove the mis-matching point pairs, and finally the optimal affine transformation matrix is solved to realize sub-pixel level alignment.

[0084] The aligned two-period images are input into a pre-trained twin convolutional neural network for difference analysis. The network consists of two encoder branches sharing weights and a contrast measure head, the encoder uses a deep residual network structure, and the output feature dimension is 512.

[0085] The network uses a large number of labeled samples for supervised learning in the training stage, and the labels are divided into three categories: mining area, dumping area, and stable area. The network output is the probability distribution of each pixel belonging to a class, and the class with the maximum probability is taken as the final classification result to generate a mask of ground feature change classification. The mask effectively distinguishes permanent topographic changes caused by mining activities from non-topographic changes caused by transport vehicles, temporary equipment, etc.

[0086] The mining area digital production operation plan data is parsed again. The spatial influence range of each operation record in the plan is projected into a unified geographic coordinate system to form a series of circular or polygonal regions. According to the operation type, a prior change probability value is assigned: 0.95 for blasting operations, 0.9 for open-pit mining, 0.85 for dumping operations, and 0.6 for other auxiliary operations. For areas not covered by any operation plan, a basic background probability value of 0.1 is assigned. After superimposing all regions, an operation plan prior probability map is formed.

[0087] Finally, a Bayesian probability fusion framework is established, taking the operation plan prior probability map as prior information, and the initial change intensity map and the ground feature change classification mask as observation evidence, to perform pixel-level probability update calculation. Let be the prior probability, be the likelihood function, The posterior probability is calculated.

[0088] For each grid cell, if the topographic feature change classification mask determines mining or filling, the likelihood function takes the value of the pixel value corresponding to the initial change intensity map; if it determines stability, the likelihood function takes the value of its complement. The posterior probability calculation formula is as follows:

[0089]

[0090] wherein, is a normalization constant. Through pixel-by-pixel calculation, the final terrain change probability grid model is generated.

[0091] In step S3, according to the terrain change probability grid model, a cluster of grid cells with a probability value higher than a preset change determination threshold is identified and extracted to generate one or more target area polygons to be scanned. The preset change determination threshold is set to 0.4, which is determined by historical data backtracking verification, ensuring that the missed detection rate is less than 5% while controlling the false alarm rate within an acceptable range.

[0092] The threshold is applied to the terrain change probability grid model for binary processing, and grid cells with a probability value higher than 0.4 are marked as change grids, and the rest are marked as stable grids. Then, a density-based spatial clustering noise application algorithm is performed on all change grids.

[0093] The core parameters of the algorithm include neighborhood radius and minimum point number . The neighborhood radius is set to 30m, corresponding to three grid cells; the minimum point number is set to 5. The algorithm iterates through all change grids, and change grids with a spatial distance less than the neighborhood radius and connected are aggregated into a change region cluster.

[0094] For each change region cluster, its convex hull or minimum area bounding rectangle is calculated, and further simplified into a polygon with less than 20 vertices as a target area polygon to be scanned. The vertex coordinates of the polygon are represented in the World Geodetic System, with an accuracy of centimeters.

[0095] In step S4, for each target area polygon to be scanned, a laser radar scanning task instruction containing its geographic coordinates, range, and scanning accuracy requirement is generated, and the task instruction is sent to the laser radar scanning execution platform. The process of generating the task instruction includes dynamically setting the scanning parameters and multi-platform task allocation.

[0096] First, according to the area size, terrain complexity, and average probability value of each target area polygon to be scanned in the terrain change probability grid model, a scanning point cloud density index is dynamically set for it.

[0097] ​Area less than 10000m 2 In the region, the point cloud density is set to 200 points / m². 2 Area of ​​10,000 m² 2 Up to 50000m 2 The area between them is set to 100 points / m 2 Area greater than 50,000 m² 2 The area is set to 50 points / m 2 .

[0098] The complexity of the terrain is assessed by calculating the standard deviation of the elevation within the polygon. When the standard deviation is greater than 5m, the point cloud density is increased by 20% compared to the original level. When the average probability value is higher than 0.7, the point cloud density is increased by 10%.

[0099] Next, the real-time status database of the LiDAR scanning execution platform is queried. This database records the current location, remaining battery percentage, maximum endurance, current payload status, and communication link quality of all available platforms. Assume there are three LiDAR scanning execution platforms in the system, numbered 1, 2, and 3. Platform 1 is located at the charging station on the east side of the mine area, with 90% remaining battery and a maximum endurance of 45 minutes; Platform 2 is currently executing a task left over from the previous cycle and is expected to return in 10 minutes; Platform 3 is located on the west helipad, with 75% remaining battery and an endurance of 35 minutes.

[0100] A multi-objective optimization function is then constructed. The optimization objective is to minimize the total completion time of all scanning tasks, with the following constraints: the single flight time of each platform is less than 90% of its maximum endurance; the total area of ​​the scanned region covered by each platform in a single task is less than its payload capacity limit; and all scanned regions must be allocated exactly once.

[0101] The optimization variables are the task-platform allocation matrix and the flight path sequence of each platform. An improved genetic algorithm is used to solve this optimization problem. The population size is set to 100, the crossover probability is 0.8, the mutation probability is 0.1, and the number of iterations is 200 generations. The fitness function comprehensively considers the total flight time, platform energy consumption balance, and task urgency weights. After solving, the optimal task allocation scheme is obtained: Platform 1 is responsible for two high-probability small-area areas, Platform 2 is responsible for a large-area area after returning to base, and Platform 3 is responsible for a medium-area complex terrain area.

[0102] For each platform, a detailed flight path plan is generated, including the takeoff point, waypoint sequence, scan altitude, scan speed, and return-to-home strategy. Finally, this information is encapsulated into a machine-executable task description file and transmitted to the corresponding LiDAR scanning execution platform via a 400MHz wireless data link.

[0103] In step S5, the laser radar scanning execution platform autonomously navigates to the location of the target region polygon to be scanned according to the received task instruction, and performs high-resolution three-dimensional laser scanning to obtain the current three-dimensional point cloud data of the region. The laser radar scanning execution platform is a multi-rotor unmanned aerial vehicle equipped with a high-precision laser radar scanner, an inertial navigation system, and a real-time differential positioning module.

[0104] The laser radar scanner is a 16-line mechanical rotary type, with a ranging accuracy of ±2 cm, a maximum ranging of 200 m, and a scanning frequency of 10 Hz. The inertial navigation system includes a three-axis gyroscope and a three-axis accelerometer, with a zero offset stability better than 0.1° / h. The real-time differential positioning module receives correction signals from three ground reference stations deployed in the mine area to achieve centimeter-level positioning accuracy.

[0105] After receiving the task instruction, the platform first performs self-checking to confirm that the battery voltage, communication link, and sensor status are normal.

[0106] Subsequently, the platform autonomously takes off according to the planned path and flies along the preset waypoints. Before entering the target region polygon boundary, the laser radar scanner is started 50 m away from the boundary, and the inertial navigation system and differential positioning module data are recorded synchronously.

[0107] During scanning, the platform maintains a constant height of 50 m, and the flight speed is set to 5 m / s to ensure that the point cloud density meets the task requirements. The scanning coverage is strictly limited within the polygon boundary, and data outside the boundary is automatically discarded.

[0108] After scanning is completed, the platform returns to the designated landing point along the planned return path and uploads the raw point cloud data package through the wireless link. The data package includes the timestamp, raw laser echo intensity, angle encoder reading, inertial navigation raw data, and differential positioning solution results.

[0109] In step S6, the obtained current three-dimensional point cloud data is processed and seamlessly fused into the historical period three-dimensional digital terrain reference model to generate an updated three-dimensional digital terrain reference model for the current period. The processing process includes four sub-steps.

[0110] First, the laser radar scanning raw point cloud data is denoised and filtered. A statistical outlier removal algorithm is used to calculate the average distance of each point from its nearest neighbor. If the distance is greater than the global mean plus twice the standard deviation, the point is determined to be a noise point and is removed. This step effectively removes abnormal points caused by atmospheric suspended dust, flying birds, or aircraft vibration.

[0111] Secondly, the point cloud is solved in absolute geographic coordinates by using airborne inertial navigation system data and differential correction data of ground base station. Through tight coupling Kalman filtering algorithm, the precise three-dimensional coordinates of each laser point in the world coordinate system are solved by fusing laser radar ranging, inertial attitude and differential positioning position information. The geographic registration error of the solved point cloud is controlled within 5 cm in the horizontal direction and within 10 cm in the vertical direction.

[0112] Thirdly, the ground points and non-ground points of the point cloud are classified. The gradual morphological filtering algorithm is adopted, the window size is set from 0.5 m to 10 m gradually, and the ground points are iteratively separated. For steep areas with a slope greater than 45°, a plane fitting method based on normal vector is used for correction. Finally, the pure ground point cloud is extracted, and non-ground objects such as vegetation, vehicles and equipment are removed.

[0113] Finally, in the three-dimensional digital terrain reference model of the historical period, the region data covered by the polygon of the target region to be scanned is marked and deleted. The reference model is stored in the form of irregular triangle net, and the coordinates and normal vector of each triangle are recorded. The deletion operation quickly locates the affected triangle set through spatial indexing and removes it from the model to form a hollow area.

[0114] Subsequently, the irregular triangle net construction algorithm is used to reconstruct the local triangle net by taking the extracted pure ground point cloud and the boundary points of the hollow area as input. The newly constructed triangle net and the original model are seamlessly spliced at the boundary, and the normal vector is continuous without cracks or overlaps.

[0115] The updated model is the three-dimensional digital terrain reference model of the current period, which has higher spatial resolution and accuracy than remote sensing data, and can be directly used for earthwork calculation, slope stability analysis and mining plan adjustment. The model is stored in the database and serves as the historical reference for the next monitoring period, which is called in step S2, thus forming a complete closed-loop update process.

[0116] The above method steps constitute the core technical scheme of the present application. On this basis, the present application further provides a metal mine area terrain updating system based on laser radar, which is used to support the automatic execution of the above method. The system includes a multi-source heterogeneous data acquisition module, a terrain change probability modeling module, a dynamic scanning task planning module, a high-precision laser radar scanning execution module and a three-dimensional terrain model updating module.

[0117] The multi-source heterogeneous data acquisition module is deployed in the mine area data center server and connected to the satellite data service provider interface, the unmanned aerial vehicle ground control station and the production management information system through a dedicated line. The module has a built-in timing task scheduler that triggers data pulling operations at preset intervals and performs preliminary format verification and metadata extraction on the original data to ensure data integrity and timeliness.

[0118] The terrain change probability modeling module is connected with the multi-source heterogeneous data acquisition module, receives the preliminary processed remote sensing image, orthographic image and production plan data. The module internally integrates a synthetic aperture radar data processing unit, an optical image analysis unit, a production plan analysis unit and a Bayesian probability fusion unit.

[0119] The synthetic aperture radar data processing unit calls an interferometric measurement professional library to perform coherence calculation and phase unwrapping; the optical image analysis unit loads a pre-trained twin convolutional neural network model and runs in a graphics processor accelerated environment; the production plan analysis unit converts the job plan into a geospatial object through a structured analysis engine; and the Bayesian probability fusion unit performs pixel-level probability updating under a unified grid calculation framework to output a terrain change probability grid model.

[0120] The dynamic scanning task planning module is connected with the terrain change probability modeling module and receives the probability grid model as input. The module internally includes a change area extraction unit and a task parameter and resource matching unit. The change area extraction unit calls a spatial clustering algorithm library to perform threshold segmentation and clustering analysis; and the task parameter and resource matching unit maintains a real-time platform state table and embeds a multi-objective optimization algorithm solver to support switching between genetic algorithm and ant colony optimization algorithm. The module outputs a standardized task instruction file and issues it to each laser radar scanning execution platform through a secure encryption channel.

[0121] The high-precision laser radar scanning execution module is composed of multiple unmanned aerial vehicle platforms carrying laser radars, and each platform is equipped with an embedded flight controller. The controller has a built-in task analysis engine that can analyze the issued task instructions and drive the flight control system to perform autonomous navigation and scanning operations. The platform maintains bidirectional communication with the ground station through a 400MHz data link to transmit state information and point cloud data in real time.

[0122] The three-dimensional terrain model updating module is deployed on a high-performance computing cluster and receives raw point cloud data from each platform. The module integrates a point cloud processing pipeline to sequentially perform denoising, registration, classification and model fusion operations. The processing results are written into a three-dimensional geographic information system database and provided to other business systems in the mining area through an application programming interface.

[0123] In summary, the present application realizes intelligent prediction, accurate positioning and efficient updating of terrain changes in a metal mining area by constructing a multi-level perception system from macro remote sensing to micro laser radar, combined with prior knowledge of production plans. The entire process does not require human intervention, improving the timeliness, accuracy and resource utilization efficiency of terrain monitoring, and providing a solid data foundation for intelligent mine construction.

[0124] It is to be understood that the terminology used herein such as first and second, and the like, is only used to distinguish one entity or action from another entity or action, and does not necessarily require or imply any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0125] While embodiments of the present application have been shown and described with reference to particular embodiments thereof, it will be understood by those skilled in the art that various changes in form and details can be made therein without departing from the spirit and scope of the application. The scope of the application is defined by the appended claims and their equivalents.

Claims

1. A method for updating the terrain of a metal mining area based on lidar, characterized in that, include: Acquire multi-source heterogeneous monitoring data covering the entire metal mining area. The multi-source heterogeneous monitoring data includes periodically acquired satellite synthetic aperture radar remote sensing image data, UAV aerial survey orthophoto data, and real-time acquired mining area digital production operation plan data. Based on a historical periodic three-dimensional digital terrain benchmark model, spatiotemporal registration and fusion processing are performed on the multi-source heterogeneous monitoring data to establish a terrain change probability raster model covering the metal mining area, specifically including: The satellite synthetic aperture radar remote sensing image data is subjected to coherence change detection processing. By calculating the interferometric coherence coefficient value between two consecutive periods of images, an initial change intensity map characterizing the subtle deformation of the Earth's surface is generated. The current UAV aerial survey orthophoto data is registered pixel-level with the digital orthophoto map of the previous period, and then input into a pre-trained twin convolutional neural network for difference analysis. The twin convolutional neural network is trained to distinguish between real terrain changes and temporary surface cover changes caused by vehicles and equipment, and outputs a classification mask for changes in ground features identified as mining, landfill, or stable features. The digital production operation plan data of the mining area is analyzed, the operation area specified in the plan is mapped to the geographic coordinate system, and a prior change probability value is assigned according to the operation type to form an operation plan prior probability map; A Bayesian probability fusion framework is established, which uses the prior probability map of the operation plan as prior information, and the initial change intensity map and the land feature change classification mask as observation evidence to perform pixel-level probability update calculations and finally generate the terrain change probability raster model. Each grid cell in the terrain change probability grid model has a probability value indicating that the terrain of that cell has changed. Based on the terrain change probability grid model, identify and extract grid cell clusters with probability values ​​higher than a preset change judgment threshold, and generate one or more polygons of the target area to be scanned. For each polygon of the target area to be scanned, a lidar scanning task instruction containing its geographic coordinates, range, and scanning accuracy requirements is generated, and the task instruction is sent to the lidar scanning execution platform. The lidar scanning execution platform autonomously navigates to the location of the polygon in the target area to be scanned according to the received task instructions, and performs high-resolution three-dimensional laser scanning to obtain the current three-dimensional point cloud data of the area; The acquired current 3D point cloud data is processed and seamlessly integrated into the historical 3D digital terrain benchmark model to generate an updated 3D digital terrain benchmark model for the current period.

2. The method for updating the terrain of a metal mining area based on lidar according to claim 1, characterized in that, Obtaining multi-source heterogeneous monitoring data covering the entire metal mining area specifically includes: By using an interface with a satellite data service provider, we periodically acquire time-series synthetic aperture radar interferometry datasets covering the mining area with specific revisit periods. The drone aerial survey platform is dispatched to conduct periodic patrols of the mining area according to a preset route, collect high-resolution optical images, and generate digital orthophoto maps through photogrammetry technology. By accessing the mining area production management information system through the application programming interface, a structured production operation plan containing information on the time, spatial location, and impact range of blasting, mining, and spoil disposal operations within the next monitoring cycle can be obtained.

3. The method for updating the terrain of a metal mining area based on lidar according to claim 1, characterized in that, The coherence change detection processing of the satellite synthetic aperture radar remote sensing image data includes: Synthetic Aperture Radar (SAR) image pairs acquired from two consecutive monitoring cycles were selected, and their interferometric coherence coefficients were calculated. The formula is: ； and These represent the complex radar echo signals of the first and second cycles, respectively. This indicates the calculation of the mean within a local window; The calculated coherence coefficient map is normalized and mapped to an initial variation intensity map, with pixel values ​​ranging from 0 to 1.

4. The method for updating the terrain of a metal mining area based on lidar according to claim 1, characterized in that, The current UAV aerial survey orthophoto data is pixel-level registered with the previous period's digital orthophoto image, and then input into a pre-trained Siamese convolutional neural network for difference analysis, including: A feature-point-based affine transformation model is used to perform sub-pixel-level alignment of the two images. The aligned two-phase images are input into a Siamese convolutional neural network consisting of two encoder branches with shared weights and a contrast metric head. The encoder adopts a deep residual network structure. The network outputs the probability distribution of the category to which each pixel belongs, and takes the category with the highest probability as the final classification result to generate a land cover change classification mask.

5. The method for updating the terrain of a metal mining area based on lidar according to claim 1, characterized in that, The digital production operation plan data of the mining area is analyzed, the operation areas specified in the plan are mapped to a geographic coordinate system, and a prior probability value of change is assigned according to the operation type to form an operation plan prior probability map, including: Project the spatial influence range of each work record onto a unified geographic coordinate system to form a circular or polygonal area; After all regions are superimposed, a prior probability map of the work plan is formed.

6. The method for updating the terrain of a metal mining area based on lidar according to claim 1, characterized in that, A Bayesian probabilistic fusion framework is established for pixel-level probability update calculations, including: set up For prior probability, Let be the likelihood function. This is the posterior probability; If the land cover change classification mask determines that it is mining or landfill, the likelihood function value is the pixel value corresponding to the initial change intensity map; if it is determined to be stable, the likelihood function value is its complement. Through the formula: ； The normalization constant is used; the posterior probability is calculated pixel by pixel to generate the terrain change probability raster model.

7. The method for updating the terrain of a metal mining area based on lidar according to claim 6, characterized in that, Generate a polygon of the target area to be scanned based on the terrain change probability raster model, including: A global change determination threshold is applied to the terrain change probability grid model for binarization. Grids with probability values ​​higher than the threshold are marked as changing grids, and those with probability values ​​lower than the threshold are marked as stable grids. A density-based spatial clustering noise application algorithm is applied to all changing rasters to aggregate spatially adjacent changing rasters into one or more changing region clusters. Calculate the minimum bounding polygon for each cluster of changing regions, and use this polygon as the polygon of the target region to be scanned.

8. The method for updating the terrain of a metal mining area based on lidar according to claim 1, characterized in that, Generate and issue lidar scanning task instructions, including: Based on the area size of each target region polygon to be scanned, the complexity of the terrain, and its average probability value in the terrain change probability raster model, a scanning point cloud density index is dynamically set for it. Query the real-time status database of the lidar scanning execution platform to obtain the current location, remaining power, and available payload information of each platform; Construct a multi-objective optimization function whose optimization objective is to minimize the total completion time of all scanning tasks, with constraints being platform endurance and load limitations; The multi-objective optimization function is solved using a genetic algorithm or an ant colony optimization algorithm to obtain the optimal task allocation scheme and flight path planning for each platform, generate a machine-executable task description file, and send it to the corresponding lidar scanning execution platform via a wireless data link.

9. The method for updating the terrain of a metal mining area based on lidar according to claim 1, characterized in that, The acquired 3D point cloud data is processed and fused into the 3D digital terrain benchmark model, including: The raw point cloud data scanned by lidar is denoised and filtered to remove noise points caused by atmospheric suspended particles or aircraft vibrations. The point cloud is calculated by using the differential correction data from the airborne inertial navigation system and the ground base station to achieve accurate georegistration. Classify the point cloud into ground points and non-ground points, and extract the pure ground point cloud; In the three-dimensional digital terrain benchmark model of the historical period, the area data covered by the polygon of the target area to be scanned is marked and deleted; By employing an irregular triangular mesh construction algorithm or a kriging interpolation algorithm, the extracted pure ground point cloud is inlaid into the deleted area of ​​the 3D digital terrain benchmark model, thereby achieving local accurate updates of the model and forming the 3D digital terrain benchmark model for the current period.

Images