Mining movable unmanned cleaning machine control method and system based on visual detection

Abstract

The invention discloses a mining movable unmanned cleaning machine control method and system based on visual detection, and the method comprises the steps: carrying out the real-time collection of temperature, humidity and obstacle density distribution data according to a multi-point sensor network disposed in a mine, obtaining a mine environment original data set, carrying out the data cleaning, and carrying out the real-time collection of the temperature, humidity and obstacle density distribution data; obtaining environment cleaning data; according to the environment cleaning data, mine environment situation analysis is carried out, the real-time environment state of each area is obtained, obstacle density distribution and risk signal intensity detection are carried out, and a high-risk area boundary division result is obtained; and according to a high-risk area boundary division result, calculating a safe stay position, obtaining a safe stay coordinate, generating an equipment moving path, obtaining a moving path scheme, evaluating risk intensity, obtaining risk signal intensity, optimizing the moving path, and obtaining a final moving path. By means of the method, safe and accurate control over the underground unmanned cleaning machine can be achieved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) technology, and in particular to a control method and system for a mobile unmanned cleaning machine for mining based on vision detection. Background Technology

[0002] With the deepening of intelligent mine construction, the demand for dynamic risk perception, real-time obstacle avoidance control, and full-coverage cleaning in underground roadway cleaning operations is becoming increasingly prominent. Its application scenarios cover complex environments such as high-gas mines and narrow, deformed roadways, requiring the control system not only to accurately identify visual features of water seepage / gas accumulation but also to integrate environmental changes and equipment status data in real time to optimize path decisions. For example, in high-risk conditions such as sudden changes in gas concentration gradients or water seepage from roof cracks, it is necessary to capture abnormal texture features in real time using multispectral vision and dynamically reconstruct safe paths to avoid collisions or cleaning blind spots.

[0003] In existing technologies, static environment modeling and rule-driven control methods are commonly used. For example, a point cloud map is constructed using LiDAR SLAM, an emergency stop mechanism is triggered based on a preset safety threshold, and a fixed trajectory algorithm is used to execute the cleaning task. However, control strategies based on a single sensor or pre-programmed logic cannot respond in real time to dynamic changes in the roadway (such as temporary material stockpiles obstructing visual access) and risk propagation characteristics (such as a 3-second lag in traditional sensors when the gas diffusion rate is >0.8 m / s). The lack of visual detection depth leads to the failure of the risk-path coordination mechanism, resulting in insufficient cleaning coverage and excessive emergency response, making it difficult to meet the stringent requirements of mine safety regulations.

[0004] In summary, existing technologies lack the ability to perform dynamic risk modeling and closed-loop optimization driven by visual inspection, and cannot integrate environmental semantics and equipment status in real time, resulting in delayed risk response, inaccurate path planning, and difficulty in achieving safe control of unmanned cleaning machines in underground mines. Summary of the Invention

[0005] This invention provides a control method and system for a mobile unmanned cleaning machine for mining based on vision detection, so as to achieve safe and precise control of the unmanned cleaning machine underground.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a control method for a vision-based mobile unmanned cleaning machine for mining, comprising: Based on the multi-point sensor network deployed in the mine, real-time data collection of temperature, humidity and obstacle density distribution is performed to obtain the raw dataset of the mine environment; Based on the original dataset of the mine environment, a data cleaning operation is performed to obtain environmental cleaned data; Based on the environmental cleaning data, a mine environmental situation analysis is conducted to obtain the real-time environmental status of each area; Based on the real-time environmental conditions of each region, obstacle density distribution and risk signal intensity are detected to obtain the boundary delineation results of high-risk areas; Based on the boundary delineation results of the high-risk areas, the safe stopping position is calculated to obtain the safe stopping coordinates; Based on the safe stopping coordinates and the real-time environmental status of each area, the equipment movement path is generated to obtain the movement path scheme; Based on the aforementioned movement path scheme, a risk intensity assessment is performed to obtain the risk signal strength; Based on the strength of the risk signal, the movement path is optimized to obtain the final movement path.

[0007] As an optional implementation, the step involves real-time acquisition of temperature, humidity, and obstacle density distribution data using a multi-point sensor network deployed within the mine to obtain a raw dataset of the mine environment, including: Based on the real-time monitoring data of the multi-point sensor network, synchronous acquisition of temperature, humidity and obstacle density distribution data in multiple areas is performed to obtain the original environmental data set. Based on the original environmental data set, regional classification and storage operations are performed to obtain the original dataset of the mine environment.

[0008] As an optional implementation, the step of performing data cleaning operations based on the original mine environment dataset to obtain environmental cleaned data includes: Based on the original dataset of the mine environment, timestamp alignment and spatial coordinate standardization operations are performed to obtain the spatiotemporal dataset of the environment. Based on the aforementioned environmental spatiotemporal dataset, outlier detection and removal operations are performed to obtain an environmental filtering dataset. Based on the environmental filtering dataset, multi-sensor data fusion and redundant information compression operations are performed to obtain environmental cleaning data.

[0009] As an optional implementation, the step of performing mine environmental situation analysis based on the environmental cleaning data to obtain the real-time environmental status of each area includes: Based on the environmental cleaning data, a spatial distribution mapping operation is performed to obtain a three-dimensional environmental parameter distribution; Based on the distribution of the three-dimensional environmental parameters, a region segmentation operation is performed to obtain the mine area division results; Based on the results of the mine area division and the distribution of three-dimensional environmental parameters, an environmental risk level assessment was conducted to obtain a preliminary risk status assessment for each area. Based on the preliminary risk assessment of each region, dynamic status updates and visualization operations are performed to obtain the real-time environmental status of each region.

[0010] As an optional implementation, the step of detecting obstacle density distribution and risk signal intensity based on the real-time environmental conditions of each region to obtain the boundary delineation result of high-risk areas includes: Based on the real-time environmental status of each region, a multi-dimensional risk feature correlation analysis is performed to obtain a risk correlation feature map; Based on the risk association feature map, a density-intensity overlay visualization operation based on the heat map is performed to obtain a heat distribution map of high-risk areas; Based on the heat map of the high-risk area, an edge detection operation is performed to obtain the preliminary boundary coordinates of the high-risk area; Based on the preliminary boundary coordinates of the high-risk area, boundary continuity verification and topology optimization are performed to obtain the boundary delineation results of the high-risk area.

[0011] As an optional implementation, the step of calculating the safe stay location and obtaining the safe stay coordinates based on the high-risk area boundary delineation results includes: Based on the high-risk area boundary delineation results, a safe distance buffer zone is generated to obtain the safe protection area; Based on the aforementioned security protection area, a flat area screening operation is performed to obtain a basic candidate location set; Based on the basic candidate location set, a location security margin assessment is performed to obtain an effective safe location set; Based on the set of effective safe locations, path reachability analysis and optimal coordinate selection are performed to obtain safe stopping coordinates.

[0012] As an optional implementation, the step of generating a device movement path based on the safe stopping coordinates and the real-time environmental status of each area to obtain a movement path scheme includes: Based on the safe stopping coordinates and the real-time environmental status of each area, a global topology path planning operation is performed to obtain an initial obstacle avoidance path sequence; Based on the initial obstacle avoidance path sequence, a path correction operation is performed to obtain a safe path update sequence; Based on the security path update sequence, multi-objective optimization and coverage integrity verification operations are performed to obtain the energy-optimal path scheme. Based on the energy-optimal path scheme and the preset cleaning task constraints, segmented speed planning is performed to obtain the movement path scheme.

[0013] As an optional implementation, the step of assessing the risk intensity based on the movement path scheme to obtain the risk signal strength includes: Based on the aforementioned movement path scheme, a node-level risk parameter fusion operation is performed to obtain a multi-dimensional risk feature vector set. Based on the multidimensional risk feature vector set, dynamic weight allocation and intensity prediction are performed to obtain the predicted value of path segment risk intensity. Based on the predicted risk intensity value of the path segment, an intensity correction operation is performed to obtain the risk signal intensity.

[0014] As an optional implementation, the step of optimizing the movement path based on the risk signal strength to obtain the final movement path includes: Based on the strength of the risk signal, the risk level is determined, and a path optimization instruction set is obtained; Based on the path optimization instruction set, the risk avoidance area is updated to obtain the optimization constraints. Based on the aforementioned optimization constraints, multi-objective real-time replanning is performed to obtain candidate optimization path schemes; Based on the candidate optimized path schemes, emergency avoidance strategies are injected and path feasibility is verified to obtain the final movement path.

[0015] Secondly, the present invention provides a control system for a mobile unmanned cleaning machine for mining based on vision detection, comprising: The data acquisition module is used to collect real-time data on temperature, humidity and obstacle density distribution based on a multi-point sensor network deployed in the mine, and obtain the raw dataset of the mine environment. The data cleaning module is used to perform data cleaning operations based on the original dataset of the mine environment to obtain environmental cleaned data. The status analysis module is used to perform mine environmental situation analysis based on the environmental cleaning data to obtain the real-time environmental status of each area. The boundary delineation module is used to detect obstacle density distribution and risk signal intensity based on the real-time environmental status of each region, and obtain the boundary delineation results of high-risk areas. The location calculation module is used to calculate the safe stay location based on the boundary delineation results of the high-risk area and obtain the safe stay coordinates; The path generation module is used to generate a movement path for the device based on the safe stopping coordinates and the real-time environmental status of each area, thereby obtaining a movement path scheme. The risk assessment module is used to assess the risk intensity based on the movement path scheme and obtain the risk signal strength. The optimization control module is used to optimize the movement path based on the strength of the risk signal to obtain the final movement path.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention solves the problem of response lag caused by environmental data distortion in traditional methods by real-time fusion of temperature, humidity and obstacle density data from multiple sensors and data cleaning technology to eliminate outlier interference, thereby effectively improving the speed of mine risk identification.

[0017] (2) This invention dynamically generates heat maps of high-risk areas and coordinates of safe stays, and injects the risk boundary division results into the path planning algorithm in real time, so as to realize that the cleaning path is automatically updated as the environment changes, effectively solving the problem of insufficient coverage and safety blind spots caused by traditional static paths.

[0018] (3) The present invention adopts a risk intensity-driven path optimization mechanism. When the detected risk signal is greater than the warning threshold, it autonomously triggers the injection of risk avoidance strategy and multi-objective replanning to solve the problem of frequent shutdown of equipment on fixed routes and effectively improve the continuous operation capability of equipment. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a control method for a mobile unmanned cleaning machine for mining based on vision detection, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the control system structure of a mobile unmanned cleaning machine for mining based on vision detection, provided in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Reference Figure 1 The first embodiment of the present invention provides a control method for a mobile unmanned cleaning machine for mining based on vision detection, comprising the following steps: S11, based on the multi-point sensor network deployed in the mine, real-time data collection of temperature, humidity and obstacle density distribution is performed to obtain the raw dataset of the mine environment; S12, Perform data cleaning operation based on the original dataset of the mine environment to obtain environmental cleaning data; S13. Based on the environmental cleaning data, perform mine environmental situation analysis to obtain the real-time environmental status of each area. S14, based on the real-time environmental status of each region, perform obstacle density distribution and risk signal intensity detection to obtain the boundary delineation result of high-risk areas; S15, Based on the boundary delineation results of the high-risk area, calculate the safe stopping position and obtain the safe stopping coordinates; S16. Based on the safe stopping coordinates and the real-time environmental status of each area, generate the equipment movement path to obtain the movement path scheme. S17. Based on the aforementioned movement path scheme, a risk intensity assessment is performed to obtain the risk signal strength. S18. Based on the strength of the risk signal, optimize the movement path to obtain the final movement path.

[0022] In step S11, the real-time acquisition of temperature, humidity, and obstacle density distribution data is performed based on a multi-point sensor network deployed within the mine to obtain the raw dataset of the mine environment, including: Based on the real-time monitoring data of the multi-point sensor network, synchronous acquisition of temperature, humidity and obstacle density distribution data in multiple areas is performed to obtain the original environmental data set. Based on the original environmental data set, regional classification and storage operations are performed to obtain the original dataset of the mine environment.

[0023] It should be noted that this step, through the collaborative work and intelligent synchronization mechanism of a multi-point sensor network, achieves millisecond-level acquisition and storage of multi-dimensional mine environment data, solving the problems of incomplete coverage and excessive latency caused by traditional manual inspections or single-point sensors. Multi-sensor data synchronization acquisition refers to the process of acquiring environmental parameters in parallel across multiple areas based on timestamp alignment and network collaboration protocols. Specifically, raw monitoring data is first acquired in real time from sensor nodes (including temperature sensors, humidity sensors, and lidar obstacle detectors) deployed on the roof, sidewalls, and ground of the mine roadway. Secondly, the data streams of all nodes are synchronized in time using wireless communication protocols (such as ZigBee or LoRa) to ensure that temperature, humidity, and obstacle density data are aligned at the same moment. Then, data format unification processing is performed, converting temperature values ​​to degrees Celsius, humidity values ​​to percentages, and obstacle density values ​​to the number of obstacles per unit volume. Finally, all synchronized data is integrated to generate a raw environmental data set containing timestamps, spatial coordinates, and the three types of parameter values.

[0024] The regional classification and storage operation refers to the process of efficiently organizing environmental data through spatial grid mapping and database indexing technology. In practice, the mine roadways are first divided into predefined grid areas (e.g., each 10m x 10m unit) and assigned a unique area number. Next, based on the spatial coordinate information in the original environmental dataset, temperature, humidity, and obstacle density data are classified into their corresponding grid areas. Then, a relational database (such as MySQL or SQLite) is used to construct classification and storage tables. The temperature data table stores the area number, timestamp, and temperature value; the humidity data table stores the area number, timestamp, and humidity value; and the obstacle density data table stores the area number, timestamp, and density value. Finally, data compression and index optimization are performed to generate a directly queryable original mine environmental dataset.

[0025] In step S12, the data cleaning operation based on the original mine environment dataset to obtain environmental cleaned data includes: Based on the original dataset of the mine environment, timestamp alignment and spatial coordinate standardization operations are performed to obtain the spatiotemporal dataset of the environment. Based on the aforementioned environmental spatiotemporal dataset, outlier detection and removal operations are performed to obtain an environmental filtering dataset. Based on the environmental filtering dataset, multi-sensor data fusion and redundant information compression operations are performed to obtain environmental cleaning data.

[0026] The timestamp alignment and spatial coordinate standardization operation refers to the process of achieving unified positioning of multi-source data based on spatiotemporal reference transformation technology. In practice, firstly, the timestamp information of all sensors is extracted from the original mine environment dataset and calibrated to the atomic clock time reference via the NTP protocol. Secondly, a global coordinate system for the roadway is established (the origin is the mine entrance, the X-axis extends along the main roadway, the Y-axis is horizontal and perpendicular to the X-axis pointing to the left roadway wall, and the Z-axis is vertically upward, forming a right-handed coordinate system). Then, through rigid transformation (including translation compensation: calculating spatial offset based on installation location; rotation matrix calculation: correcting sensor orientation deviation (e.g., rotating the Y-axis of sidewall sensors by -15°)), the local coordinates collected by each sensor are transformed to this unified reference. Next, spatiotemporal mapping tables are established for temperature, humidity, and obstacle density data to ensure that data from the same location at the same time correspond uniquely. Finally, the calibrated data is integrated to generate an environmental spatiotemporal dataset containing standardized timestamps, a unified coordinate system location, and three types of parameters.

[0027] It should be noted that outlier detection and removal refers to the process of identifying invalid data through dynamic fluctuation analysis and box plot statistical models. In practice, the environmental spatiotemporal dataset is first divided by region, and the historical mean and standard deviation of temperature, humidity, and obstacle density for each region are calculated. Secondly, dynamic judgment rules are set: temperatures exceeding ±3 times the standard deviation, humidity changes >10% / second, and obstacle density experiencing a sudden surge of 300% are all marked as outliers. Then, a sliding window mechanism (10-second window size) is used to scan the data stream frame by frame, adding invalid labels to outlier data. Finally, physical removal is performed, and the outlier locations are recorded, generating an environmental filtered dataset that retains valid data.

[0028] It is worth noting that multi-sensor data fusion and redundant information compression refers to the process of data dimensionality reduction based on principal component analysis and feature weighting algorithms. In practice, firstly, the temperature, humidity, and obstacle density in the environmental filtering dataset are constructed as a three-dimensional feature vector; secondly, the PCA algorithm is used to calculate feature weights (temperature weight 0.4, humidity weight 0.3, density weight 0.3), generating the fusion parameter value F = 0.4T + 0.3H + 0.3D, where F, T, H, and D represent the fusion parameter value, temperature value, humidity value, and density value, respectively; then, data points with changes of <5% within 10 consecutive seconds are merged, retaining the mean as the representative value; finally, duplicate information is compressed to generate environmental cleaning data containing only key features.

[0029] In step S13, the step of performing mine environmental situation analysis based on the environmental cleaning data to obtain the real-time environmental status of each area includes: Based on the environmental cleaning data, a spatial distribution mapping operation is performed to obtain a three-dimensional environmental parameter distribution; Based on the distribution of the three-dimensional environmental parameters, a region segmentation operation is performed to obtain the mine area division results; Based on the results of the mine area division and the distribution of three-dimensional environmental parameters, an environmental risk level assessment was conducted to obtain a preliminary risk status assessment for each area. Based on the preliminary risk assessment of each region, dynamic status updates and visualization operations are performed to obtain the real-time environmental status of each region.

[0030] It should be noted that this step addresses the shortcomings of traditional two-dimensional planar analysis in perceiving three-dimensional risks in mines by employing 3D environmental modeling and dynamic risk quantification techniques. Spatial distribution mapping refers to the process of constructing a 3D mine environment model based on point cloud reconstruction and spatial interpolation algorithms. Specifically, this involves first extracting spatial coordinates and parameters such as temperature, humidity, and obstacle density from the environmental cleaning data; secondly, using Kriging interpolation to transform discrete data points into a continuous 3D raster model (resolution 0.5m × 0.5m × 0.5m); then generating a thermal gradient layer for temperature data, an isosurface layer for humidity data, and a point density layer for obstacle density; finally, superimposing these three layers of data to construct a 3D environmental parameter distribution model. Region segmentation refers to the process of intelligently dividing risk units using density clustering and boundary optimization techniques. In practice, the comprehensive risk coefficient R for each grid point is first calculated based on the three-dimensional environmental parameter distribution model: R = 0.5 × (T / T_max) + 0.3 × (H / H_max) + 0.2 × (D / D_max) (where T, H, and D represent temperature, humidity, and density values, respectively; T_max, H_max, and D_max represent the maximum allowable temperature threshold, humidity safety limit, and density limit, respectively). Next, the DBSCAN clustering algorithm is used to aggregate continuous grids with similar R values ​​into independent regions (minimum clustering unit ≥ 8 cubic meters). Then, boundary smoothing is performed to eliminate jagged edges. Finally, a unique number is assigned to each region and vertex coordinates are recorded to generate a dataset of mine area division results.

[0031] The environmental risk level assessment operation refers to the process of quantifying regional risks based on fuzzy logic and dynamic weight allocation. In practice, firstly, three core indicators—average temperature, peak humidity, and maximum obstacle density—are extracted for each defined region. Secondly, a risk rule base is established: temperature > 40°C triggers high-temperature risk, humidity > 95% triggers flooding risk, and obstacle density > 50 obstacles / cubic meter triggers collision risk. Then, the risk intensity is calculated based on the degree to which the indicators exceed limits (e.g., high-temperature risk intensity I_h = 1.5^(T-40)). The calculated high-temperature risk intensity, flooding risk intensity, and collision risk intensity are used as inputs to a fuzzy inference system (such as MATLAB's Fuzzy Logic Toolbox) for fusion. Finally, the three risk intensity values ​​are fused to generate a regional risk level (levels 1-5), resulting in a preliminary risk status assessment report for each region. Dynamic status update and visualization refers to the process of visualizing the environmental situation through a real-time rendering engine and incremental learning algorithms. In practice, the risk level assessment results are first mapped to a three-dimensional mine model, and each area is marked with a gradient from red (level 5) to green (level 1). Secondly, a new data stream is received every 2 seconds, and the risk value of the area is updated using a sliding window mechanism (the area is re-rendered immediately when the change is greater than 10%). Then, the risk diffusion trend is dynamically displayed on the visualization interface (such as arrows indicating that the red area is spreading to adjacent roadways). Finally, a real-time environmental status matrix of each area with timestamps is output.

[0032] In step S14, the step of detecting obstacle density distribution and risk signal intensity based on the real-time environmental status of each region to obtain the boundary delineation result of high-risk areas includes: Based on the real-time environmental status of each region, a multi-dimensional risk feature correlation analysis is performed to obtain a risk correlation feature map; Based on the risk association feature map, a density-intensity overlay visualization operation based on the heat map is performed to obtain a heat distribution map of high-risk areas; Based on the heat map of the high-risk area, an edge detection operation is performed to obtain the preliminary boundary coordinates of the high-risk area; Based on the preliminary boundary coordinates of the high-risk area, boundary continuity verification and topology optimization are performed to obtain the boundary delineation results of the high-risk area.

[0033] It should be noted that this step addresses the boundary ambiguity problem caused by the reliance on manual experience in traditional high-risk area delineation through multi-dimensional risk correlation and dynamic topology optimization techniques. The multi-dimensional risk feature correlation analysis refers to the process of mining risk coupling relationships based on covariance matrices and graph neural networks. Specifically, it first extracts three core features from the real-time environmental conditions of each area: temperature anomaly, humidity gradient, and obstacle density mutation rate. Next, it calculates the covariance matrix between features (temperature-obstacle density correlation coefficient) to construct a feature correlation graph (nodes represent feature types, and edge weights represent correlation strength). Then, it uses a graph convolutional network (GCN) to learn high-risk patterns (when the obstacle density mutation rate > 20% / second and the temperature anomaly > 3σ, the risk signal strength is automatically increased by 2 levels). Finally, it generates a risk correlation feature graph with weighted coefficients.

[0034] The density-intensity overlay visualization operation refers to the process of achieving a three-dimensional presentation of the risk situation through a heatmap fusion rendering engine. Specifically, the obstacle density distribution map is first converted into a blue gradient heatmap (the higher the density, the darker the blue); then, the risk signal intensity map is converted into a red gradient heatmap (the higher the intensity, the darker the red); next, an Alpha channel overlay algorithm is used to generate a dual-channel heatmap. The Alpha channel fusion formula is Output = α×Red + (1-α)×Blue (where α is the risk intensity normalization value, Red is the risk signal intensity value, and Blue is the obstacle density value). The red-blue intersection area (purple) marks the high-risk area, and the pure red area marks the emergency risk area. Finally, a high-risk area heatmap with a resolution of 0.1 meters per pixel (including three-dimensional spatial coordinates) is output. Edge detection and preliminary boundary generation refers to the process of extracting the contours of high-risk areas based on the Canny operator and morphological processing. In practice, the heat map of high-risk areas is first subjected to Gaussian filtering for noise reduction (kernel size 5×5); then, an adaptive double-threshold Canny algorithm is used to detect edges: a high threshold (0.7) is used for high temperature difference areas (red-blue boundary), and a low threshold (0.3) is used for homogeneous areas. The adaptive threshold calculation formula is Threshold = 0.4 + 0.03×|∇G| (∇G is the image gradient magnitude); then, morphological closing operation is performed to fill the contour gaps (3×3 circular structure kernel); finally, the set of boundary points outside the connected domain is extracted to generate the preliminary boundary coordinate sequence of high-risk areas (such as coordinate chain [(x1,y1,z1), (x2,y2,z2)...]).

[0035] It is worth noting that the boundary continuity verification and topology optimization operation refers to the process of smoothing the boundary through Delaunay triangulation and spline curve fitting. In practice, firstly, breakpoints in the initial boundary coordinates are detected (a distance greater than 1 meter between adjacent points is considered a break); secondly, Delaunay triangulation is used to connect the breakpoints and generate transition patches. Firstly, the breakpoints and auxiliary points are combined into an input point set, sorted by spatial location, and a topological index is established. Then, initial triangles are generated based on the empty circle property, iteratively checking whether new points are located within the circumcircle of existing triangles—if a point is detected within the circumcircle, the triangle is split, and the mesh is continuously optimized, forcing all triangles to have a minimum interior angle greater than 25 degrees to avoid generating distorted triangular elements. Finally, a triangulation result conforming to the Delaunay criterion is output, generating transition patches. Then, B-spline curve fitting is used to smooth the jagged boundary (curvature change threshold < 0.15 rad / m); finally, a closed and smooth boundary delineation result for high-risk regions is output.

[0036] In step S15, the step of calculating the safe stay location and obtaining the safe stay coordinates based on the high-risk area boundary delineation results includes: Based on the high-risk area boundary delineation results, a safe distance buffer zone is generated to obtain the safe protection area; Based on the aforementioned security protection area, a flat area screening operation is performed to obtain a basic candidate location set; Based on the basic candidate location set, a location security margin assessment is performed to obtain an effective safe location set; Based on the set of effective safe locations, path reachability analysis and optimal coordinate selection are performed to obtain safe stopping coordinates.

[0037] It should be noted that the safe distance buffer zone generation operation refers to the process of constructing a dynamic protection zone based on the risk intensity gradient and roadway structure constraints. In practice, firstly, the topological relationships and risk level data in the high-risk area boundary delineation results are analyzed; secondly, a buffer distance benchmark value is set according to the risk level (Level 1 risk: 5 meters, with the distance increasing by 1.8 times for each additional level); then, the buffer zone is trimmed in conjunction with roadway width constraints (e.g., the buffer distance in narrow roadway areas is compressed by 30%); finally, a ring-shaped safe protection zone is generated that encloses the high-risk area, with its inner boundary at least ≥ the dynamically calculated value from the danger zone and its outer boundary at least ≥ 0.8 meters of passage space from the roadway wall.

[0038] The flat area screening operation refers to the process of initially selecting habitable locations through ground tilt angle detection and obstacle density analysis. Specifically, this involves first sampling ground tilt angle data within a 1m x 1m grid within the safety protection area; secondly, screening grid points with tilt angles < 5° (slope tolerance corresponds to the personnel stability threshold); then, integrating the real-time obstacle density map and eliminating grids with a density > 2 obstacles / square meter; finally, generating a basic candidate location set containing spatial coordinates and terrain parameters (retaining an average of 12-15 candidate points per 100 meters of tunnel). The location safety margin assessment operation refers to the process of quantifying location safety based on a risk field strength attenuation model and multi-hazard coupling analysis. Specifically, this involves first calculating the Euclidean distance D between each candidate point and the nearest high-risk area; then, assessing the risk field strength attenuation value. ( The risk level is classified as high-risk area. The Euclidean distance between the candidate point and the nearest high-risk area is used; then, secondary disaster impact factors are superimposed: roof stability coefficient (ground radar data), ventilation dead zone marker (area with wind speed < 0.3 m / s). The roof stability coefficient is normalized into a risk contribution value, and the ventilation dead zone marker is directly used as a high-risk penalty item. The total impact of secondary disasters is combined, and the calculation formula is: Total disaster impact = weighting coefficient of roof stability × risk contribution value + high-risk penalty item; finally, a comprehensive safety margin score (0-100 points) is generated, and the calculation formula is: = 100 × × (1 - (k is the attenuation coefficient,) (As a secondary disaster factor), points with a score ≥80 are retained to form a set of effective safe locations.

[0039] It is worth noting that path reachability analysis and optimal coordinate selection refer to the process of making the best risk avoidance location decision through the A algorithm and multi-objective optimization. In specific implementation, firstly, the current location of personnel is obtained and a roadway topology network map is constructed; secondly, the improved A algorithm is used to calculate the shortest path to each safe point (cost function = path length × 0.7 + traversal risk value × 0.3), where the traversal risk value is calculated as: 0.6 × the sum of risks of the edges traversed by the path + 0.4 × the maximum risk value of all grids traversed by the path; then, reachable points with path costs less than a threshold are selected; finally, the coordinates with the highest safety margin and the lowest path cost are selected as the final safe stopping coordinates (if there are ties, the location near the ventilation opening is preferred).

[0040] In step S16, generating a device movement path based on the safe stopping coordinates and the real-time environmental status of each area to obtain a movement path scheme includes: Based on the safe stopping coordinates and the real-time environmental status of each area, a global topology path planning operation is performed to obtain an initial obstacle avoidance path sequence; Based on the initial obstacle avoidance path sequence, a path correction operation is performed to obtain a safe path update sequence; Based on the security path update sequence, multi-objective optimization and coverage integrity verification operations are performed to obtain the energy-optimal path scheme. Based on the energy-optimal path scheme and the preset cleaning task constraints, segmented speed planning is performed to obtain the movement path scheme.

[0041] It should be noted that the global topology path planning operation refers to the process of constructing an initial navigation path based on Dijkstra's algorithm and a three-dimensional risk field. In specific implementation, the mine space is first discretized into a three-dimensional grid map with a resolution of 0.5 meters; then, a grid passage cost C = 0.6 × risk value + 0.3 × slope + 0.1 × obstacle density is assigned according to the real-time environmental status of each area; then, with the current position of the equipment as the starting point and the safe stopping coordinates as the ending point, the Dijkstra algorithm with risk constraints is executed (skipping dead grids with a cost > 8); finally, the initial obstacle avoidance path sequence composed of continuous grid coordinates is output (such as the path point set [P1(x1,y1,z1), P2(x2,y2,z2)...]).

[0042] The real-time path correction operation refers to the process of dynamically updating the path through rolling time-domain control and risk diffusion prediction. In practice, firstly, environmental status update data is received every 0.5 seconds; secondly, risk mutation points (grids with a risk value change rate > 15% / second) are detected in the path sequence; then, local path replanning is performed using an artificial potential field method: a repulsion potential field is generated for high-risk areas. ,in The repulsive potential field represents the strength of the repulsive force exerted by an obstacle on a mobile device (such as a drone). This represents the actual spatial distance from the device's current location to the nearest obstacle. To determine the maximum effective range of the repulsive force generated by the obstacle, an attractive potential field is generated in the safe zone. ( (The distance from the current point to the target); finally, the global path and local potential field are fused to generate a safe path update sequence. Multi-objective optimization and coverage integrity verification operation refers to the process of making energy-optimal path decisions based on the NSGA-II algorithm and coverage analysis. In specific implementation, firstly, a dual objective function is established: total path length F1, risk exposure integral F2 = Σ(risk value × passage time), where the risk value refers to the real-time dynamic risk score of each spatial unit (grid / segment) traversed by the path, reflecting the degree of environmental hazard at that location; passage time represents the time required for the device to pass through a specific grid or path segment. Secondly, a Pareto front solution set (50 sets of non-dominated path schemes) is generated; then, the task coverage integrity is verified: the path coverage rate C = actual coverage area / area to be cleaned is calculated for the area to be cleaned; finally, the path that satisfies C≥98% and has the lowest energy consumption is selected as the energy-optimal path scheme (energy consumption model E = 0.8 × distance + 0.2 × ∑(slope angle × segment length)).

[0043] It is worth noting that segmented speed planning refers to the process of generating a speed curve envelope by combining equipment dynamics constraints and task requirements. In practice, the task constraints are first analyzed and cleared: maximum speed Vmax = 1.5 m / s, emergency stop acceleration Amax = 0.3g; secondly, segmented speeds are set according to the path curvature ρ: Vmax is used for straight sections (ρ < 0.05 / m), and 0.4Vmax is used for curved sections (ρ > 0.2 / m); then a safety buffer is added: a slow speed segment of 0.2Vmax is inserted 10 meters before and after areas with a risk level > 3; finally, a movement path scheme with velocity vectors is generated (format: {coordinate, velocity, acceleration} triple sequence).

[0044] In step S17, the risk intensity assessment based on the movement path scheme to obtain the risk signal strength includes: Based on the aforementioned movement path scheme, a node-level risk parameter fusion operation is performed to obtain a multi-dimensional risk feature vector set. Based on the multidimensional risk feature vector set, dynamic weight allocation and intensity prediction are performed to obtain the predicted value of path segment risk intensity. Based on the predicted risk intensity value of the path segment, an intensity correction operation is performed to obtain the risk signal intensity.

[0045] It should be noted that the node-level risk parameter fusion operation refers to the process of aggregating multi-source risk features based on a spatiotemporal encoder. In specific implementation, firstly, seven-dimensional parameters are extracted from each path node along the movement path scheme, including: real-time gas concentration (%); roof stress value (MPa); obstacle dynamic density (numbers / m²); temperature gradient change rate (℃ / s); historical accident frequency (normalized value); equipment vibration amplitude (g); and communication signal attenuation rate (dB / m). Secondly, a spatiotemporal encoder (CNN+GRU) is used to generate a 128-dimensional feature vector: first, spatial correlation is extracted through a convolutional layer (3×3 kernels), then temporal dependence is captured through a GRU layer; finally, a multi-dimensional risk feature vector set is output (format: {node ID: [f1,f2,...,f128]}).

[0046] The dynamic weight allocation and intensity prediction operation refers to the process of quantitatively predicting risk intensity through an attention mechanism and an LSTM network. In practice, a dual-channel LSTM prediction model is first constructed, including: Channel 1: Path Topology LSTM (input node connectivity); Channel 2: Environmental Evolution LSTM (input feature vector temporal changes). Next, a dynamic attention weight module is designed: input features include topological hidden states (h_top): vectors representing the structural features of the road network (such as path connectivity, node distance, etc.) and environmental hidden states (h_env): vectors representing real-time environmental risks (such as gas concentration, temperature, and other sensor data). The specific weight calculation process is as follows: first, the topological hidden states and environmental hidden states are concatenated into a joint feature vector; then, a linear transformation is performed using the trainable parameter matrix Wa; finally, the Softmax function is applied to the transformation result to generate attention weights (including topological feature weights and environmental feature weights). The two types of features are then weighted and fused: fused feature = α × topological hidden state + β × environmental hidden state. Risk intensity prediction: First, the fused features are linearly transformed using the trainable parameter matrix Wr; then, the transformation result is compressed to the (0,1) interval using the Sigmoid function; finally, the result is multiplied by 10 to convert it into a 0-10 level risk score: Risk intensity = Sigmoid(Wr × fused features) × 10, outputting the predicted risk intensity value for each path segment.

[0047] It is worth noting that the real-time intensity correction operation refers to the process of dynamically calibrating the predicted value based on sensor feedback and Kalman filtering. In practice, firstly, a laser methane sensor and a microseismometer are deployed to collect actual risk data in real time; secondly, the residual between the predicted and measured values ​​is calculated: Residual = |R_pred - R_real|, where R_pred is the predicted value, R_real is the measured value, and Residual is the residual. Then, an adaptive Kalman filter is used to update the prediction model: when the residual < 1.0: weak update (process noise Q reduced by 50%); when the residual ≥ 1.0: strong update (observation noise R reduced by 70%). Finally, the corrected risk signal intensity is output.

[0048] In step S18, optimizing the movement path based on the risk signal strength to obtain the final movement path includes: Based on the strength of the risk signal, the risk level is determined, and a path optimization instruction set is obtained; Based on the path optimization instruction set, the risk avoidance area is updated to obtain the optimization constraints. Based on the aforementioned optimization constraints, multi-objective real-time replanning is performed to obtain candidate optimization path schemes; Based on the candidate optimized path schemes, emergency avoidance strategies are injected and path feasibility is verified to obtain the final movement path.

[0049] It should be noted that the risk level determination operation refers to the process of generating optimized instructions based on dynamic threshold rules. In practice, firstly, dynamic warning thresholds are calculated based on equipment status and environmental parameters: the base threshold is set to level 6.0; when the battery level is below 20%, the threshold increases by 20%; when the communication signal strength is above -85 dB, the threshold decreases by 30%. Secondly, a three-level determination rule is constructed: if the risk signal strength is between 6.0 and 7.9, a "partial detour" instruction is generated; if it is between 8.0 and 8.9, an "emergency turn" instruction is generated; and if it reaches level 9.0 or above, an "emergency stop awaiting rescue" instruction is generated. Then, the instruction strength is fine-tuned based on the equipment's mechanical load rate (the instruction strength coefficient increases by 0.2 when the load is >80%). Finally, a path optimization instruction set containing the instruction type, effective range, and strength coefficient is output (e.g., {Instruction type: emergency turn, effective range: nodes P23 to P47, strength coefficient: 0.8}).

[0050] The risk avoidance area update operation refers to the process of reconstructing map constraints. In practice, the initial restricted area range is first determined based on the command type: the "partial detour" command generates a circular restricted area with a radius of 3 meters; the "emergency turn" command generates a restricted area with a radius of 5 meters and expands it to three adjacent lanes; the "emergency stop awaiting rescue" command sets the entire path as a restricted area. Secondly, the restricted area is dynamically adjusted using a risk diffusion model: when the risk diffusion speed exceeds 0.5 meters per second, the restricted area radius expands exponentially over time (increasing by 10% for every second, switching to linear growth when the speed decreases for two consecutive seconds). Then, the path cost function is updated, with the new cost value multiplied by a coefficient equal to "1 plus the risk level divided by 10". Finally, the optimized constraints, including the restricted area coordinate set and the corrected cost matrix, are output. The multi-objective real-time replanning operation refers to the process of generating candidate path schemes. In practice, a three-objective optimization model is first established: the first objective is to shorten the path length (weight reduced by 10% in emergency situations); the second objective is to reduce risk exposure (weight increased by 30% in high-risk situations); and the third objective is to control energy consumption increases. Secondly, a decompositional multi-objective evolutionary algorithm is employed: 50 sets of solutions are initialized, and the solution set is updated through neighborhood crossover over 20 iterations. Then, based on information entropy theory, the weights of each objective are calculated (60% for safety, 30% for energy efficiency, and 10% for smoothness), and the scheme with the highest comprehensive score is selected. Finally, three candidate paths and their objective function values ​​are output.

[0051] It's worth noting that the emergency evacuation strategy injection and path feasibility verification operation refers to the process of ensuring the safe execution of the path. In specific implementation, strategies are first injected according to risk level: for risks of 6-7, a deceleration command of 40% of maximum speed is added 20 meters before the risk point; for risks of 8-9, an S-shaped trajectory with a curvature of 0.15 per meter is inserted; for risks above level 9, a 3-meter-long emergency stopping zone with a friction coefficient exceeding 0.7 is designated. Secondly, four-dimensional verification is performed: kinematic verification ensures the maximum centripetal acceleration is less than 0.25 times the acceleration due to gravity; dynamic verification requires the emergency stopping distance to be less than 80% of the visual distance; energy consumption verification limits the estimated energy consumption to no more than 90% of the remaining battery power; and communication verification ensures the signal strength throughout the path is higher than -90 dB. Finally, the final movement path, including path coordinates, speed strategy, and emergency markings, is output.

[0052] Reference Figure 2 The second embodiment of the present invention provides a control system for a vision-based mobile unmanned cleaning machine for mining, comprising: The data acquisition module is used to collect real-time data on temperature, humidity and obstacle density distribution based on a multi-point sensor network deployed in the mine, and obtain the raw dataset of the mine environment. The data cleaning module is used to perform data cleaning operations based on the original dataset of the mine environment to obtain environmental cleaned data. The status analysis module is used to perform mine environmental situation analysis based on the environmental cleaning data to obtain the real-time environmental status of each area. The boundary delineation module is used to detect obstacle density distribution and risk signal intensity based on the real-time environmental status of each region, and obtain the boundary delineation results of high-risk areas. The location calculation module is used to calculate the safe stay location based on the boundary delineation results of the high-risk area and obtain the safe stay coordinates; The path generation module is used to generate a movement path for the device based on the safe stopping coordinates and the real-time environmental status of each area, thereby obtaining a movement path scheme. The risk assessment module is used to assess the risk intensity based on the movement path scheme and obtain the risk signal strength. The optimization control module is used to optimize the movement path based on the strength of the risk signal to obtain the final movement path.

[0053] It should be noted that the vision-based mobile unmanned cleaning machine control system for mines provided in this embodiment of the invention is used to execute all the process steps of the vision-based mobile unmanned cleaning machine control method for mines in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0054] This invention also provides a terminal device. The terminal device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a vision-based detection control program for a mobile unmanned cleaning machine in a mine. When the processor executes the computer program, it implements the steps described in the various vision-based detection control method embodiments for mobile unmanned cleaning machines in mines, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above system embodiments.

[0055] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0056] The terminal device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The terminal device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of terminal devices and do not constitute a limitation on the terminal device. It may include more or fewer components than described above, or a combination of certain components, or different components. For example, the terminal device may also include input / output devices, network access devices, buses, etc.

[0057] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0058] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0059] Wherein, if the modules / units integrated in the terminal device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0060] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0061] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A control method for a mobile unmanned cleaning machine for mining based on vision detection, characterized in that, include: Based on the multi-point sensor network deployed in the mine, real-time data collection of temperature, humidity and obstacle density distribution is performed to obtain the raw dataset of the mine environment; Based on the original dataset of the mine environment, a data cleaning operation is performed to obtain environmental cleaned data; Based on the environmental cleaning data, a mine environmental situation analysis is conducted to obtain the real-time environmental status of each area; Based on the real-time environmental conditions of each region, obstacle density distribution and risk signal intensity are detected to obtain the boundary delineation results of high-risk areas; Based on the boundary delineation results of the high-risk areas, the safe stopping position is calculated to obtain the safe stopping coordinates; Based on the safe stopping coordinates and the real-time environmental status of each area, the equipment movement path is generated to obtain the movement path scheme; Based on the aforementioned movement path scheme, a risk intensity assessment is performed to obtain the risk signal strength; Based on the strength of the risk signal, the movement path is optimized to obtain the final movement path.

2. The control method for a vision-based mobile unmanned cleaning machine for mining, as described in claim 1, is characterized in that... The process involves real-time acquisition of temperature, humidity, and obstacle density distribution data using a multi-point sensor network deployed within the mine, resulting in a raw dataset of the mine environment, including: Based on the real-time monitoring data of the multi-point sensor network, synchronous acquisition of temperature, humidity and obstacle density distribution data in multiple areas is performed to obtain the original environmental data set. Based on the original environmental data set, regional classification and storage operations are performed to obtain the original dataset of the mine environment.

3. The control method for a vision-based mobile unmanned cleaning machine for mining, as described in claim 1, is characterized in that... The step of performing data cleaning operations based on the original mine environment dataset to obtain environmental cleaned data includes: Based on the original dataset of the mine environment, timestamp alignment and spatial coordinate standardization operations are performed to obtain the spatiotemporal dataset of the environment. Based on the aforementioned environmental spatiotemporal dataset, outlier detection and removal operations are performed to obtain an environmental filtering dataset. Based on the environmental filtering dataset, multi-sensor data fusion and redundant information compression operations are performed to obtain environmental cleaning data.

4. The control method for a vision-based mobile unmanned cleaning machine for mining, as described in claim 1, is characterized in that... The step of performing mine environmental situation analysis based on the environmental cleaning data to obtain the real-time environmental status of each area includes: Based on the environmental cleaning data, a spatial distribution mapping operation is performed to obtain a three-dimensional environmental parameter distribution; Based on the distribution of the three-dimensional environmental parameters, a region segmentation operation is performed to obtain the mine area division results; Based on the results of the mine area division and the distribution of three-dimensional environmental parameters, an environmental risk level assessment was conducted to obtain a preliminary risk status assessment for each area. Based on the preliminary risk assessment of each region, dynamic status updates and visualization operations are performed to obtain the real-time environmental status of each region.

5. The control method for a vision-based mobile unmanned cleaning machine for mining, as described in claim 1, is characterized in that... The process of detecting obstacle density distribution and risk signal intensity based on the real-time environmental conditions of each region to obtain the boundary delineation results of high-risk areas includes: Based on the real-time environmental status of each region, a multi-dimensional risk feature correlation analysis is performed to obtain a risk correlation feature map; Based on the risk association feature map, a density-intensity overlay visualization operation based on the heat map is performed to obtain a heat distribution map of high-risk areas; Based on the heat map of the high-risk area, an edge detection operation is performed to obtain the preliminary boundary coordinates of the high-risk area; Based on the preliminary boundary coordinates of the high-risk area, boundary continuity verification and topology optimization are performed to obtain the boundary delineation results of the high-risk area.

6. The control method for a vision-based mobile unmanned cleaning machine for mining, as described in claim 1, is characterized in that... The step of calculating the safe stay location and obtaining the safe stay coordinates based on the high-risk area boundary delineation results includes: Based on the high-risk area boundary delineation results, a safe distance buffer zone is generated to obtain the safe protection area; Based on the aforementioned security protection area, a flat area screening operation is performed to obtain a basic candidate location set; Based on the basic candidate location set, a location security margin assessment is performed to obtain an effective safe location set; Based on the set of effective safe locations, path reachability analysis and optimal coordinate selection are performed to obtain safe stopping coordinates.

7. The control method for a vision-based mobile unmanned cleaning machine for mining, as described in claim 1, is characterized in that... The step of generating a device movement path based on the safe stopping coordinates and the real-time environmental status of each area to obtain a movement path scheme includes: Based on the safe stopping coordinates and the real-time environmental status of each area, a global topology path planning operation is performed to obtain an initial obstacle avoidance path sequence; Based on the initial obstacle avoidance path sequence, a path correction operation is performed to obtain a safe path update sequence; Based on the security path update sequence, multi-objective optimization and coverage integrity verification operations are performed to obtain the energy-optimal path scheme. Based on the energy-optimal path scheme and the preset cleaning task constraints, segmented speed planning is performed to obtain the movement path scheme.

8. The control method for a vision-based mobile unmanned cleaning machine for mining, as described in claim 1, is characterized in that... The step of assessing the risk intensity based on the movement path scheme to obtain the risk signal strength includes: Based on the aforementioned movement path scheme, a node-level risk parameter fusion operation is performed to obtain a multi-dimensional risk feature vector set. Based on the multidimensional risk feature vector set, dynamic weight allocation and intensity prediction are performed to obtain the predicted value of path segment risk intensity. Based on the predicted risk intensity value of the path segment, an intensity correction operation is performed to obtain the risk signal intensity.

9. The control method for a vision-based mobile unmanned cleaning machine for mining, as described in claim 1, is characterized in that... The step of optimizing the movement path based on the risk signal strength to obtain the final movement path includes: Based on the strength of the risk signal, the risk level is determined, and a path optimization instruction set is obtained; Based on the path optimization instruction set, the risk avoidance area is updated to obtain the optimization constraints. Based on the aforementioned optimization constraints, multi-objective real-time replanning is performed to obtain candidate optimization path schemes; Based on the candidate optimized path schemes, emergency avoidance strategies are injected and path feasibility is verified to obtain the final movement path.

10. A control system for a vision-based mobile unmanned cleaning machine for mining, characterized in that, include: The data acquisition module is used to collect real-time data on temperature, humidity and obstacle density distribution based on a multi-point sensor network deployed in the mine, and obtain the raw dataset of the mine environment. The data cleaning module is used to perform data cleaning operations based on the original dataset of the mine environment to obtain environmental cleaned data. The status analysis module is used to perform mine environmental situation analysis based on the environmental cleaning data to obtain the real-time environmental status of each area. The boundary delineation module is used to detect obstacle density distribution and risk signal intensity based on the real-time environmental status of each region, and obtain the boundary delineation results of high-risk areas. The location calculation module is used to calculate the safe stay location based on the boundary delineation results of the high-risk area and obtain the safe stay coordinates; The path generation module is used to generate a movement path for the device based on the safe stopping coordinates and the real-time environmental status of each area, thereby obtaining a movement path scheme. The risk assessment module is used to assess the risk intensity based on the movement path scheme and obtain the risk signal strength. The optimization control module is used to optimize the movement path based on the strength of the risk signal to obtain the final movement path.

Images