Mine geological environment dynamic monitoring method based on unmanned aerial vehicle remote sensing

Abstract

The invention belongs to the technical field of remote sensing technology and ecological environment monitoring, and particularly relates to a mine geological environment dynamic monitoring method based on unmanned aerial vehicle remote sensing. A mine geological environment dynamic monitoring method based on unmanned aerial vehicle remote sensing comprises the following steps: S10, acquiring a panoramic remote sensing image of a target area, identifying and marking a vehicle road network in the panoramic remote sensing image, and determining segmentation sizes of different positions in the panoramic remote sensing image according to the vehicle road network density, the segmentation size is inversely proportional to the vehicle road network density, and segmenting the panoramic remote sensing image into a plurality of remote sensing sub-images according to the segmentation sizes of different positions; and S20, obtaining color information of each pixel point in the remote sensing sub-image, and extracting pixel texture features in the remote sensing sub-image according to the color information of each pixel point. According to the scheme, the problem that misinformation is easily generated due to the fact that slag dumping and landslide are easily confused in an existing unmanned aerial vehicle remote sensing technology is solved.

Description

Technical Field

[0001] This solution belongs to the field of remote sensing technology and ecological environment monitoring technology, and specifically involves a dynamic monitoring method of mine geological environment based on drone remote sensing. Background Art

[0002] As modern mining operations continue to expand, the risk of mine landslides has significantly increased. Mines are often located in complex mountainous and hilly terrain, where rock and soil are unstable. Mining activities cause the rock and soil to become fragmented and loose, and geological structural damage further weakens its integrity. Traditional manual survey methods, due to the complex terrain and vast scope of mining areas, are difficult to fully cover, making it easy to miss geological hazards. Furthermore, surveys are time-consuming and labor-intensive, with delayed results, making them unable to meet the real-time and accuracy requirements.

[0003] In recent years, drone remote sensing technology has been widely used in mine geological environment monitoring due to its high flexibility, fast data acquisition, and low cost. For example, the Chengdu Geological Survey Center uses drones equipped with LiDAR, oblique photography, and orthophoto sensors to achieve high-precision real-world 3D modeling, accurately extracting parameters such as area, earthwork volume, and slope angle of each type of land, supporting mine ecological restoration.

[0004] However, the locations where slag is dumped within some mines are uncertain, and the accumulation process is similar to a landslide on exposed land. Drone image recognition algorithms that rely solely on the mountain's outer contours to distinguish between the two can easily confuse the two, leading to frequent false alarms. Existing technologies use millimeter-wave radar penetrating imaging and thermal infrared surface temperature difference detection to compensate for the lack of optical data, but these technologies are costly and prohibitive for most mines due to economic constraints. Summary of the Invention

[0005] The purpose of this program is to provide a dynamic monitoring method for mine geological environment based on UAV remote sensing, so as to solve the problem that existing UAV remote sensing technology easily confuses slag dumping and landslides, resulting in false alarms.

[0006] To achieve the above objectives, this solution provides a method for dynamic monitoring of mine geological environment based on UAV remote sensing, which includes the following steps:

[0007] S10: Acquire a panoramic remote sensing image of the target area, identify and mark the vehicle road network in the panoramic remote sensing image, and determine a segmentation size at different positions in the panoramic remote sensing image according to the vehicle road network density, wherein the segmentation size is inversely proportional to the vehicle road network density, and segment the panoramic remote sensing image into a plurality of remote sensing sub-images according to the segmentation size at different positions;

[0008] S20: Acquire color information of each pixel in the remote sensing sub-image, extract pixel texture features in the remote sensing sub-image based on the color information of each pixel, and divide the pixel texture features into road network texture features and rock texture features;

[0009] S30: analyzing the movement vectors of the pixels of the rock texture feature in the remote sensing sub-image according to the acquisition time of the panoramic remote sensing image, and calculating the movement volume and movement vector of the rock texture feature according to the movement vectors;

[0010] S40: When the moving volume is greater than the preset landslide volume, a positional relationship between the moving vector and the road network texture feature is obtained, and a landslide warning is triggered according to the positional relationship.

[0011] The principle and technical effect of this solution are as follows: First, this solution divides the panoramic remote sensing image into several remote sensing sub-images by identifying and marking the vehicle road network in the panoramic remote sensing image and determining the segmentation size according to the vehicle road network density. This method can dynamically adjust the segmentation size according to the vehicle road network density to ensure the accuracy of image segmentation. Areas with high vehicle road network density usually mean frequent human activities, while areas with low vehicle road network density are closer to dumping areas. Areas with frequent human activities are more stable geologically due to the influence of vehicles, while areas with less human activities have looser accumulated slag and are more prone to landslides. In this way, different types of geological changes can be more accurately identified and analyzed, and the scope of this solution's focus on loose geological areas can be precisely narrowed, so that slag sliding can be obtained faster and more accurately, reducing the occurrence of false landslide alarms.

[0012] Furthermore, the more frequent human activity there is in an area, the more vehicles there are. When vehicles pass through the road network, the greater the resonance they bring to the mine, and the greater the probability of a mine landslide. At this time, expanding the scope of attention to the road network (i.e., expanding the segmentation size of the remote sensing sub-image) can obtain a wider range of environmental information around the road network, thereby discovering the connection between the road network and slag landslides earlier and more accurately, and generating landslide warning information more accurately. The areas with less human activity are closer to the dumping area and the geology is looser. Reducing the segmentation size of this part of the area can more accurately obtain the volume difference of slag landslides in various directions. When the difference in landslides in a certain direction is large, landslides are more likely to occur in this direction. When the volume of slag landslides in various directions is small and evenly distributed, the possibility of rockfall caused by the vibration caused by the passing of vehicles is greater. By reducing the segmentation size of this part, this solution can more accurately monitor the slag landslides near the dumping point, and thus more accurately generate landslide warning information.

[0013] In summary, this solution solves the problem that existing UAV remote sensing technology easily confuses slag dumping and landslides, leading to false alarms.

[0014] Furthermore, the calculation method for determining the segmentation size according to the vehicle road network density in step S10 includes the following steps:

[0015] S11: The road network density field is established at the pixel point position. The formula for establishing the road network density field is shown in the following formula (1):

[0016]

[0017] Among them, (x, y) is the image pixel coordinate, (x i ,y i ) represents the coordinates of the center point of the i-th road network segment, N is the total number of road network segments, A is the normalization coefficient, and σ is the preset Gaussian kernel bandwidth;

[0018] S12: Construct an adaptive segmentation size mapping model, as shown in the following formula (2):

[0019]

[0020] Among them, S(x,y) is the segmentation size at (x,y), S max 、S min denote the maximum and minimum segmentation sizes, ρ max , ρ min are the upper and lower limits of road network density;

[0021] S13: Use the quadtree segmentation algorithm to implement image partitioning, which satisfies the following formula (3):

[0022]

[0023] Among them, R k represents the kth segmented region, and η is the size variation coefficient.

[0024] The present invention effectively distinguishes the characteristic differences between slag dumping and landslides through an image segmentation method that is adaptive to vehicle-road network density. Since slag dumping is mostly concentrated in vehicle-dense areas, while natural landslides often occur in slope areas with sparse road networks, this method establishes a road network density field and dynamically adjusts the segmentation size, so that vehicle-dense areas use fine segmentation to capture slag transportation details, while sparse road network areas use larger segmentation sizes to highlight mountain deformation characteristics, thereby significantly reducing the risk of misjudgment. In terms of computational efficiency, the allocation of computing resources is optimized through intelligent segmentation size adjustment; in terms of multi-scale feature extraction, it can not only retain the fine motion trajectory of vehicle machinery, but also capture the macroscopic displacement of the mountain; in terms of environmental adaptability, it overcomes the limitations of traditional methods in complex mining environments; in terms of data association, it provides a reliable spatial constraint basis for subsequent vibration-landslide coupling analysis. This series of technical improvements has formed a complete monitoring optimization system, which not only solves the core misjudgment problem, but also comprehensively improves the overall performance of this solution.

[0025] Furthermore, the calculation of the moving volume and moving vector of the rock texture feature includes the following steps:

[0026] S31: For the panoramic remote sensing image I of adjacent time series t and I t+Δt , where Δt is the interval time for panoramic remote sensing image acquisition. The optical flow method is used to calculate the displacement vector field V(x,y) of the characteristic pixel points of rock texture. V(x,y)=(u(x,y),v(x,y)), where u(x,y) and v(x,y) represent the displacement components of the pixel point (x,y) in the horizontal and vertical directions, respectively. The calculation is performed by minimizing the energy function. The calculation formula is shown in the following formula (4):

[0027]

[0028] Among them, I x ,I y ,I t are the gradients of the image in the x, y, and t directions respectively, and λ is the smoothing coefficient;

[0029] S32: Based on the displacement vector field V(x,y), calculate the moving volume M of the rock texture feature. The calculation formula of M is shown in the following formula (5):

[0030]

[0031] Among them, V i is the displacement vector of the i-th pixel, w i is the weight factor, which is determined by the texture saliency and spatial position of the i-th pixel;

[0032] S33: Perform spatial clustering on the displacement vector field to extract the dominant movement direction D. The calculation formula of D is shown in the following formula (7):

[0033]

[0034] Among them, K is the number of clusters, n k is the number of pixels in the k-th cluster, is the displacement mean vector of the cluster;

[0035] S34: When the moving volume M exceeds the threshold M textth When the angle θ between the dominant movement direction D and the slope normal is less than 45°, a landslide warning is triggered, where:

[0036]

[0037] N is the geometric normal vector of the slope.

[0038] This solution achieves a more essential distinction at the mechanism level through the calculation of the displacement vector field and the moving volume of the characteristic pixel points of the rock texture. Specifically, slag dumping usually manifests as local, discrete short-range displacement (such as the growth of slag piles caused by the unloading of transport vehicles), while landslides present a continuous, coordinated large-scale displacement field. The displacement vector field calculated by the optical flow method can effectively capture the essential difference in this movement pattern: the displacement vectors in the slag area are disordered and have sudden changes in amplitude, while the displacement vectors in the landslide area have spatial continuity and directional consistency. Furthermore, by introducing texture saliency weights and displacement clustering analysis, not only the interference caused by local changes in the slag pile is suppressed, but also the identification of the overall movement trend of the landslide is enhanced.

[0039] Furthermore, the vehicle texture features are extracted from the road network texture features, and the position changes of the vehicles in the road network are analyzed according to the acquisition interval of the panoramic remote sensing image to obtain the vehicle's moving speed, moving direction and the number of vehicles in the road network; the resonance effect of the vehicle vibration on the road network is calculated according to the number of vehicles, moving speed and moving direction, wherein the resonance effect is calculated by the vibration transfer model, and the calculation formula is shown in the following formula (9):

[0040]

[0041] Among them, m i is the equivalent mass of the i-th vehicle, a i is the acceleration, d i is the distance from the vehicle to the potential landslide area, β is the vibration attenuation coefficient;

[0042] The distribution position of the vehicles is spatially associated with the movement vector of the rock texture feature to calculate the resonance-landslide coupling risk value R. The calculation formula of R is shown in the following formula (10):

[0043] R = α·M + (1-α)·F (10), where α is the geological vulnerability coefficient. When R exceeds the preset threshold and the angle between the rock movement direction and the main vibration direction is less than the set angle, the resonance landslide warning is triggered.

[0044] This solution uses real-time monitoring of key parameters such as the number, speed, and direction of transport vehicles in the mining area to establish a dynamic correlation model between vehicle activity and geological changes, enabling accurate identification of the two phenomena at the mechanistic level. Slag dumping areas exhibit high-intensity, but localized, vibration characteristics due to frequent vehicle activity, while potential landslide areas exhibit a continuously increasing and widespread vibration response pattern. This fundamental difference provides a reliable basis for accurate identification. Furthermore, by incorporating a geological environment adaptive adjustment mechanism, this solution enables the monitoring system to automatically optimize assessment parameters for different geotechnical conditions, significantly improving the model's applicability. Secondly, this solution integrates mechanical vibration effects with geological displacement characteristics for analysis, creating a more comprehensive risk assessment system. Thirdly, by setting a vibration direction consistency condition, random interference signals unrelated to landslides are effectively filtered out. These methods complement image segmentation techniques and displacement field analysis methods, forming a multidimensional monitoring network that spans from phenomenon observation to mechanistic analysis. This significantly reduces false alarm rates and improves early warning capabilities for potential disasters, providing a more reliable technical guarantee for mine safety.

[0045] Furthermore, when calculating the resonance-landslide coupling risk value, the weights of adjacent remote sensing sub-images are dynamically adjusted according to the size of the remote sensing sub-image, specifically including the following steps:

[0046] A10: Establish a weight distribution model for adjacent remote sensing sub-images. The weight calculation formula is shown in the following formula (11):

[0047]

[0048] Among them, w adj is the weight of the jth adjacent remote sensing sub-image to the current remote sensing sub-image i, A i is the segmentation size area of ​​the current remote sensing sub-image, d ij is the spatial distance between the center points of the two remote sensing sub-images, d max is the maximum impact distance threshold;

[0049] A20: Filter valid adjacent remote sensing sub-images according to weights, satisfying the following formula (12):

[0050]

[0051] Among them, N adj is the number of adjacent remote sensing sub-images finally selected, τ is the weight accumulation threshold, wmax is the maximum single weight value;

[0052] A30: Calculate Moving Balance Index B i , B i The calculation formula is shown in the following formula (13):

[0053]

[0054] in, is the moving vector of the rock texture feature of the jth adjacent remote sensing sub-image;

[0055] A40: Modify the landslide risk value R′. The correction formula of R′ is shown in the following formula (14):

[0056]

[0057] Among them, α is the balance factor, β is the attenuation coefficient, when B i When <0.3, a balance warning is triggered.

[0058] This scheme avoids the feature dilution caused by sparse road network in large-scale remote sensing sub-images and overcomes the disadvantage that small-scale remote sensing sub-images are easily affected by noise by dynamically adjusting the neighborhood analysis range. The corrected risk value R′ is more accurate in the early stage of motion imbalance (B i <0.3) can trigger an early warning, which is earlier than the resonance-landslide coupling model (which needs to meet the volume threshold and direction angle conditions). At the same time, if the vehicle vibration (resonance effect) only causes local non-coordinated displacement (such as slag spillage), it will be automatically suppressed by the balance index ( The term approaches zero. This approach upgrades the directional consistency test (angle θ) within a single remote sensing sub-image to a vector field coordination assessment across multiple remote sensing sub-images. While inheriting the concept of vibration-geology coupling, it achieves dynamic decoupling of mechanical vibration and natural landslides through a weight distribution model. This combination further reduces the probability of false alarms.

[0059] Furthermore, the mining area is divided into different risk areas according to the movement vector of the rock texture characteristics. When determining the recommended dumping direction, the area different from the risk area where the loose direction is located is preferred as the safe dumping area. The loose direction is calculated using the following formula (15):

[0060]

[0061] in, is the motion vector of the rock texture feature in the adjacent remote sensing sub-image j; when the motion volume of the rock texture feature of the adjacent remote sensing sub-image is obtained regularly, if the motion volume is inconsistent with the preset threshold, the current remote sensing sub-image is used as the relay monitoring area, and a remote sensing sub-image in a risk area different from the relay monitoring area is randomly selected as the verification area, and the motion vector verification data is sent to the verification area through the relay monitoring area; if the motion vector verification data received in the verification area matches the local monitoring data, the stability score of the relay monitoring area is reduced; if not, the relay monitoring area is marked as a high-risk area, and adjacent remote sensing sub-images in the same risk area as the relay monitoring area are selected one by one as auxiliary verification areas, and the motion vector verification data is sent to the verification area through the auxiliary verification area.

[0062] First, this scheme effectively distinguishes the random displacement of slag dumping from the coordinated displacement of landslides by introducing a loosely oriented spatial consistency verification mechanism, and combines the edge constraints of the road network to reduce the false alarm rate. Secondly, by using temporal behavior identification technology, the slope change rate detection and continuous tracking are used to further reduce temporal misjudgments. In addition, this scheme has the following synergistic advantages: on the one hand, its active risk control function directly converts the monitoring results into transportation path instructions, and improves the stability of the slope by optimizing the slag accumulation position; on the other hand, the multi-source data fusion design realizes the dynamic balance between road network accessibility and geological safety, and reduces computing power consumption through localized operations. It is particularly noteworthy that this method is complementary to the adjacent remote sensing sub-image weight distribution model: when the balance index B i ≥0.3, automatically start path planning, build a "monitoring-decision-execution" closed loop; and B i When it is <0.3, landslide warning is triggered first, and the two together constitute a graded response system.

[0063] Furthermore, the method further comprises the following steps:

[0064] S501: Perform cloud and fog detection on the current panoramic remote sensing image and calculate the cloud and fog coverage area ratio C. If C>0, it is determined that the current weather is rainy, and the image is divided into cloud-blocked areas and visible areas according to the cloud distribution;

[0065] S502: Acquire the most recent panoramic remote sensing image without cloud or fog as the first image, and use the currently acquired panoramic remote sensing image with cloud or fog as the second image. Extract the pixel color rendering of the visible area of ​​the second image and record it as the first color rendering S1. Extract the pixel color rendering of the same position in the first image and record it as the second color rendering S2. Calculate the color rendering difference ratio ΔS according to the following formula (16):

[0066]

[0067] For cloud-blocked areas, the humidity impact value H of the blocked area is obtained based on ΔS of the adjacent visible area and the cloud thickness estimated by infrared band penetration.

[0068] S503: Modify the weight distribution of the remote sensing sub-image according to the humidity impact value H, and predict the landslide risk value R of the visible area based on the adjusted weight and combined with the historical movement volume data. pred ;

[0069] S504: Calculate the real risk value R based on the actual collected visible area landslide data real , R real and the predicted value R pred Compare and obtain the error E. If E> the preset error threshold, use the back propagation algorithm to adjust the calculation parameters of the humidity impact value H until the error converges;

[0070] S505: Apply the optimized model to the cloud-covered area and output the predicted landslide risk value R final , according to R final A landslide warning is triggered in the area and it is marked as a high-risk area requiring manual verification.

[0071] This scheme first analyzes the historical movement patterns of cloud-obstructed areas (such as the periodic characteristics of slag dumping) and the random differences of natural landslides, combined with the correction effect of humidity impact values ​​(based on the quantification of color rendering differences), to effectively distinguish between human activities and geological disasters, thereby significantly reducing the false alarm rate. In addition, through dynamic prediction models and real-time data correction of visible areas, this scheme solves the problem of monitoring interruption caused by cloud obstruction and ensures monitoring continuity in severe weather. At the same time, based on the refined adjustment of humidity weights, this scheme can distinguish between high-humidity loose areas and low-humidity stable areas, achieve differentiated risk warnings, and avoid "one-size-fits-all" alarms. Finally, by marking areas where prediction errors exceed thresholds and resonance coupling risk areas, this scheme optimizes the allocation of manual verification resources, requiring only targeted inspections of high-risk target areas, significantly reducing the probability of false alarms.

[0072] Furthermore, in the road network texture features in the visible area, the target detection algorithm is used to identify and mark the vehicle, and the instantaneous speed v of the vehicle before entering the cloud-covered area is extracted. in , record the time t when the vehicle enters the foggy area in ; According to the road network topology and vehicle speed v in , predict the theoretical time T for a vehicle to pass through the cloud and fog obstruction area pred , Among them L cloud is the length of the road network in the area blocked by fog; the actual passing time t is calculated by the moment when the vehicle reappears in the visible area out , calculate the actual passing time Treal ,

[0073] T real =t out -t in ; Define the traffic difficulty coefficient P to reflect the abnormality of road conditions:

[0074]

[0075] If P is greater than the preset threshold, it is determined that there is a traffic obstruction in the foggy area; the equivalent passing speed v of the vehicle is dynamically corrected according to the P value. adj , v adj The calculation formula is shown in the following formula (18):

[0076] v adj =v in ·(1-ηP) (18),

[0077] Where η is the damping coefficient, 0<η<1; the difficulty coefficient P is incorporated into the landslide risk prediction model in the cloud-covered area: the corrected humidity influence value H new As shown in the following formula (19):

[0078] H new =H·(1+μP) (19),

[0079] Where μ is the coupling coefficient, which is calibrated by geological permeability; update the weight distribution model to increase the monitoring priority of high P value areas; when the corrected landslide risk value R final When the threshold is exceeded and P continues to rise, it is determined that the geological instability in the area has been aggravated by traffic obstruction, triggering a "vehicle-landslide" coupling warning.

[0080] This solution first calculates the difficulty coefficient of travel by comparing the actual travel time of vehicles in the fog-obstructed area with the predicted travel time. Since slag dumping usually presents regular travel characteristics, while landslides can cause abnormally long travel times, it can effectively distinguish between human activities and geological disasters. At the same time, cross-validation is performed in combination with humidity impact values. When high K values ​​and high H values ​​appear at the same time, it is confirmed that there is a real landslide risk, while a single high K value is not a real landslide risk.

[0081] The value may indicate other interference factors such as mechanical failure, further reducing the false alarm rate. In addition, this solution also uses vehicles as "mobile sensors" to detect hidden geological changes (such as potential ground fissures) early through traffic anomalies, which has higher sensitivity than traditional methods that rely on visual deformation detection. On this basis, the monitoring strategy is dynamically adjusted: the monitoring frequency is automatically increased for high K value areas (such as switching to SAR mode), while the scanning intensity is reduced for slag dumping areas (low K value), realizing the intelligent allocation of monitoring resources. Finally, by establishing a "vehicle-road network-geology" ternary coupling analysis system, this solution not only solves the technical problem of false alarms of slag dumping, but also achieves multiple technical effects such as early warning of hidden risks and optimal allocation of monitoring resources, significantly improving the reliability of mining geological disaster monitoring under cloud and fog conditions.

[0082] Further: Establish a comprehensive risk index R based on the rock movement volume M, resonance-landslide coupling risk value R, passage difficulty coefficient P and humidity impact value H total The weighted fusion model is shown in the following formula (20):

[0083]

[0084] Among them, ω1~ω4 are weight coefficients, which are determined by the analytic hierarchy process AHP and meet the following requirements: ), M th is the landslide volume threshold;

[0085] According to R total A multi-level warning system will be established, and different response measures will be formulated for different levels of warnings.

[0086] Furthermore, after the early warning is triggered, an emergency response plan is automatically generated based on the monitoring data and risk assessment results, which specifically includes the following steps:

[0087] B10: Based on the comprehensive risk value R total and geological vulnerability coefficient k to divide risk levels and achieve differentiated marking of high-risk areas;

[0088] B20: Combines the path planning algorithm with the resource scheduling model to generate personnel evacuation and rescue resource deployment plans;

[0089] B30: Dynamically adjust emergency plans through real-time data fusion and threshold judgment;

[0090] B40: Convert the plan into visual results and automatically push execution instructions.

[0091] In the monitoring of mine geological environment, this solution realizes dynamic assessment by integrating multi-dimensional risk factors, and dynamically adjusts weights by combining professional analysis methods, so that early warnings are more in line with actual risk changes and significantly improve the accuracy of early warnings. The early warning system of this solution is divided into multiple levels, and different levels correspond to differentiated response measures: from adjusting the monitoring frequency to launching emergency inspections, to automatically planning safe evacuation routes, implementing drone material delivery and regional linkage, forming a gradient emergency mode, effectively improving response efficiency and personnel evacuation safety. In addition, this solution differentiates and marks high-risk areas based on the results of comprehensive risk assessment to clarify rescue priorities; in personnel evacuation and resource scheduling, it optimizes path planning and scientifically deploys equipment and vehicles through intelligent algorithms, taking into account both safety and resource utilization efficiency; at the same time, it establishes a dynamic adjustment mechanism to timely refresh the plan according to real-time risk changes to ensure flexibility in responding to emergencies; with the help of visualization platforms and automated instructions, it improves command decision-making efficiency and reduces human operation deviations. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 The figure is a flow chart of a method for dynamic monitoring of mine geological environment based on UAV remote sensing in an embodiment of the present invention. DETAILED DESCRIPTION

[0093] The following will clearly and completely describe the concept and technical effects of the present invention in conjunction with the embodiments to fully understand the purpose, features and effects of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative work are all within the scope of protection of the present invention:

[0094] like Figure 1 As shown, a method for dynamic monitoring of mine geological environment based on UAV remote sensing includes the following steps:

[0095] S10: Acquire a panoramic remote sensing image of the target area, identify and mark the vehicle road network in the panoramic remote sensing image, and determine a segmentation size at different positions in the panoramic remote sensing image according to the vehicle road network density, wherein the segmentation size is inversely proportional to the vehicle road network density, and segment the panoramic remote sensing image into a plurality of remote sensing sub-images according to the segmentation size at different positions;

[0096] S20: Acquire color information of each pixel in the remote sensing sub-image, extract pixel texture features in the remote sensing sub-image based on the color information of each pixel, and divide the pixel texture features into road network texture features and rock texture features;

[0097] S30: analyzing the movement vectors of the pixels of the rock texture feature in the remote sensing sub-image according to the acquisition time of the panoramic remote sensing image, and calculating the movement volume and movement vector of the rock texture feature according to the movement vectors;

[0098] S40: When the moving volume is greater than the preset landslide volume, a positional relationship between the moving vector and the road network texture feature is obtained, and a landslide warning is triggered according to the positional relationship.

[0099] The calculation method for determining the segmentation size according to the vehicle road network density in step S10 includes the following steps:

[0100] S11: The road network density field is established at the pixel point position. The formula for establishing the road network density field is shown in the following formula (1):

[0101]

[0102] Among them, (x, y) is the image pixel coordinate, (x i ,y i ) represents the coordinates of the center point of the i-th road network segment, N is the total number of road network segments, A is the normalization coefficient, and σ is the preset Gaussian kernel bandwidth;

[0103] S12: Construct an adaptive segmentation size mapping model, as shown in the following formula (2):

[0104]

[0105] Among them, S(x,y) is the segmentation size at (x,y), S max 、S min denote the maximum and minimum segmentation sizes, ρ max , ρ min are the upper and lower limits of road network density;

[0106] S13: Use the quadtree segmentation algorithm to implement image partitioning, which satisfies the following formula (3):

[0107]

[0108] Among them, R k represents the kth segmented region, and η is the size variation coefficient.

[0109] The calculation of the moving volume and moving vector of the rock texture feature includes the following steps:

[0110] S31: For the panoramic remote sensing image I of adjacent time series t and I t+Δt, where Δt is the interval time for panoramic remote sensing image acquisition. The optical flow method is used to calculate the displacement vector field V(x,y) of the characteristic pixel points of rock texture. V(x,y)=(u(x,y),v(x,y)), where u(x,y) and v(x,y) represent the displacement components of the pixel point (x,y) in the horizontal and vertical directions, respectively. The calculation is performed by minimizing the energy function. The calculation formula is shown in the following formula (4):

[0111]

[0112] Among them, I x ,I y ,I t are the gradients of the image in the x, y, and t directions respectively, and λ is the smoothing coefficient;

[0113] S32: Based on the displacement vector field V(x,y), calculate the moving volume M of the rock texture feature. The calculation formula of M is shown in the following formula (5):

[0114]

[0115] Among them, V i is the displacement vector of the i-th pixel, w i is the weight factor, which is determined by the texture saliency and spatial position of the i-th pixel, w i The calculation formula is shown in the following formula (6):

[0116]

[0117] S i is the texture saliency (calculated based on local binary pattern LBP), d i is the Euclidean distance from the pixel to the potential starting point of the landslide, and σ is the spatial attenuation coefficient;

[0118] S33: Perform spatial clustering on the displacement vector field to extract the dominant movement direction D. The calculation formula of D is shown in the following formula (7):

[0119]

[0120] Among them, K is the number of clusters (determined adaptively based on the DBSCAN algorithm), n k is the number of pixels in the k-th cluster, is the displacement mean vector of the cluster;

[0121] S34: When the moving volume M exceeds the threshold M textth When the angle θ between the dominant movement direction D and the slope normal is less than 45°, a landslide warning is triggered, where:

[0122]

[0123] N is the geometric normal vector of the slope (calculated by the digital elevation model DEM).

[0124] The vehicle texture features are extracted from the road network texture features, and the position changes of the vehicles in the road network are analyzed according to the acquisition interval of the panoramic remote sensing image to obtain the vehicle's moving speed, moving direction and the number of vehicles in the road network; the resonance effect of the vehicle vibration on the road network is calculated according to the number of vehicles, moving speed and moving direction, wherein the resonance effect is calculated by the vibration transfer model, and the calculation formula is shown in the following formula (9):

[0125]

[0126] Among them, m i is the equivalent mass of the i-th vehicle, a i is the acceleration, d i is the distance from the vehicle to the potential landslide area, β is the vibration attenuation coefficient;

[0127] The distribution position of the vehicles is spatially associated with the movement vector of the rock texture feature to calculate the resonance-landslide coupling risk value R. The calculation formula of R is shown in the following formula (10):

[0128] R = α·M + (1-α)·F (10), where α is the geological vulnerability coefficient. When R exceeds the preset threshold and the angle between the rock movement direction and the main vibration direction is less than the set angle, the resonance landslide warning is triggered.

[0129] Specifically, when calculating the resonance-landslide coupling risk value, the weights of adjacent remote sensing sub-images are dynamically adjusted according to the size of the remote sensing sub-image, which specifically includes the following steps:

[0130] A10: Establish a weight distribution model for adjacent remote sensing sub-images. The weight calculation formula is shown in the following formula (11):

[0131]

[0132] Among them, w adj is the weight of the jth adjacent remote sensing sub-image to the current remote sensing sub-image i, A i is the segmentation size area of ​​the current remote sensing sub-image, d ij is the spatial distance between the center points of the two remote sensing sub-images, d max is the maximum impact distance threshold;

[0133] A20: Filter valid adjacent remote sensing sub-images according to weights, satisfying the following formula (12):

[0134]

[0135] Among them, N adj is the number of adjacent remote sensing sub-images finally selected, τ is the weight accumulation threshold (0.6~0.8), w max is the maximum single weight value;

[0136] A30: Calculate Moving Balance Index B i , B i The calculation formula is shown in the following formula (13):

[0137]

[0138] in, is the moving vector of the rock texture feature of the jth adjacent remote sensing sub-image;

[0139] A40: Modify the landslide risk value R′. The correction formula of R′ is shown in the following formula (14):

[0140]

[0141] Among them, α is the balance factor (1.2-1.5), β is the attenuation coefficient (0.8-1.2), when B i When <0.3, a balance warning is triggered.

[0142] The mining area is divided into different risk areas according to the moving vector of the rock texture characteristics. When determining the recommended dumping direction, the area different from the risk area where the loose direction is located is preferred as the safe dumping area. The loose direction is calculated using the following formula (15):

[0143]

[0144] in, is the moving vector of the rock texture feature in the adjacent remote sensing sub-image j.

[0145] When periodically acquiring the movement volume of rock texture features from adjacent remote sensing sub-images, if the movement volume is inconsistent with a preset threshold, the current remote sensing sub-image is used as the relay monitoring area. A remote sensing sub-image in a different risk area from the relay monitoring area is randomly selected as the verification area. Movement vector verification data is sent to the verification area via the relay monitoring area. If the movement vector verification data received in the verification area matches the local monitoring data, the stability score of the relay monitoring area is lowered. If not, the relay monitoring area is marked as a high-risk area, and adjacent remote sensing sub-images in the same risk area as the relay monitoring area are selected one by one as auxiliary verification areas. Movement vector verification data is sent to the verification area via the auxiliary verification areas.

[0146] In this embodiment, a method for dynamic monitoring of mine geological environment based on UAV remote sensing further includes the following steps:

[0147] S501: Perform cloud and fog detection on the current panoramic remote sensing image and calculate the cloud and fog coverage area ratio C. If C>0 (clouds and fog exist), it is determined to be rainy weather, and the image is divided into cloud-blocked areas (direct monitoring is limited) and visible areas (data can be collected normally) according to the distribution of clouds and fog.

[0148] S502: Acquire the most recent panoramic remote sensing image without cloud or fog as the first image, and use the currently acquired panoramic remote sensing image with cloud or fog as the second image. Extract the pixel color rendering of the visible area of ​​the second image and record it as the first color rendering S1. Extract the pixel color rendering of the same position in the first image and record it as the second color rendering S2. Calculate the color rendering difference ratio ΔS according to the following formula (16):

[0149]

[0150] For cloud-blocked areas, the humidity impact value H of the blocked area is obtained based on ΔS of the adjacent visible area and the cloud thickness estimated by infrared band penetration.

[0151] S503: Modify the weight distribution of remote sensing sub-images according to the humidity impact value H (the weight of high humidity areas is increased, and the weight of low humidity areas is reduced; the initial weight of cloud and fog obscured areas is set to 1.5 × H of the weight of the adjacent visible areas). Based on the adjusted weights and combined with historical movement volume data, predict the landslide risk value R of the visible area. pred ;

[0152] S504: Calculate the real risk value R based on the actual collected visible area landslide data (such as displacement and crack extension) real , R real and the predicted value R pred Compare and obtain the error E(E=|R pred -R real |), if E>the preset error threshold, the back propagation algorithm is used to adjust the calculation parameters of the humidity impact value H until the error converges;

[0153] S505: Apply the optimized model to the cloud-covered area and output the predicted landslide risk value R final , according to R final A landslide warning is triggered in the area and it is marked as a high-risk area requiring manual verification.

[0154] Among them, in the road network texture features in the visible area, the vehicle is identified and marked by the target detection algorithm, and the instantaneous speed v of the vehicle before entering the cloud and fog obscured area is extracted. in ; Record the time t when the vehicle enters the foggy area in ;

[0155] According to the road network topology and vehicle speed v in , predict the theoretical time T for a vehicle to pass through the cloud and fog obstruction area pred , Among them L cloud is the length of the road network in the area blocked by fog; the actual passing time t is calculated by the moment when the vehicle reappears in the visible area out , calculate the actual passing time T real , T real =t out -t in ;

[0156] The definition of the traffic difficulty coefficient P reflects the degree of abnormal road conditions:

[0157]

[0158] If P>P th (threshold), to determine whether there is a traffic obstruction (such as road collapse or water accumulation) in the cloud-covered area;

[0159] Dynamically correct the vehicle's equivalent passing speed v according to the P value adj , v adj The calculation formula is shown in the following formula (18):

[0160] v adj =v in ·(1-ηP) (18),

[0161] Where η is the damping coefficient (0<η<1);

[0162] Incorporating the access difficulty coefficient P into the landslide risk prediction model in fog-covered areas: Correcting the humidity impact value H new As shown in the following formula (19):

[0163] H new =H·(1+μP) (19),

[0164] where μ is the coupling coefficient, which is calibrated by geological permeability; the weight distribution model is updated to increase the monitoring priority of high P value areas;

[0165] When the corrected landslide risk value R final When the threshold is exceeded and P continues to rise, it is determined that the geological instability in the area has been aggravated by traffic obstruction, triggering a "vehicle-landslide" coupling warning.

[0166] Among them, the comprehensive risk index R is established based on the rock movement volume M, the resonance-landslide coupling risk value R, the passage difficulty coefficient P and the humidity impact value H. total The weighted fusion model is shown in the following formula (20):

[0167]

[0168] Among them, ω1~ω4 are weight coefficients, which are determined by the analytic hierarchy process AHP and meet the following requirements: ), M th is the landslide volume threshold;

[0169] According to R total A multi-level warning system will be established, and different response measures will be formulated for different levels of warnings.

[0170] Specifically, the multi-level warning system (four-level warning level division and response rules) is as follows:

[0171] General warning (Level I, 0.3≤R total <0.6):

[0172] Trigger condition: A single factor approaches a threshold (e.g. M ≥ 0.8M th or P ≥ 0.7P th );

[0173] Response measures: Increase the frequency of drone monitoring to once an hour, shorten the ground sensor data collection interval to 30 minutes, and generate a factor trend forecast report for the next 24 hours.

[0174] Severe warning (Level II, 0.6≤R total <0.8):

[0175] Trigger condition: At least two factors exceed the threshold (such as M≥M th and R≥0.9R th );

[0176] Response measures: Initiate drone emergency patrol routes (route density increased to 1.5 times that of standard routes), simultaneously trigger ground displacement meters and inclinometers for intensified monitoring (interval shortened to 10 minutes), and send early warning briefings to the mine safety department.

[0177] Serious warning (Level III, 0.8≤R total <0.95):

[0178] Trigger conditions: Multiple factors significantly exceed the standard (such as M≥1.2M th and P ≥ 1.1P th );

[0179] Response measures: Generate evacuation paths based on the Dijkstra algorithm. The path cost function is shown in the following formula (21):

[0180] C path =α·d+β·∑ (x,y)∈path R(x,y) (21),

[0181] Where d is the path distance, R(x,y) is the risk value of the path pixel, and α and β are the distance and risk weight coefficients (α + β = 1). The mine broadcast system is automatically activated to issue evacuation orders by region and send equipment dispatch requests to emergency material storage points.

[0182] Especially serious warning (Level IV, R total ≥0.95):

[0183] Trigger condition: M ≥ 1.5M th And R total Continuous ascent for more than 2 hours;

[0184] Response measures: Immediately activate the emergency evacuation plan for the entire mining area, and deploy emergency rescue supplies through a swarm of drones (the coordinates of the deployment points are determined by the particle swarm optimization algorithm (PSO), and the objective function is to minimize the coverage of blind spots); send an encrypted early warning data packet to the local emergency management department, containing a real-time 3D geological model, personnel location distribution, and predicted landslide boundary data; and establish an emergency joint defense mechanism with surrounding mining areas to share monitoring data and disposal resources.

[0185] After an early warning is triggered, an emergency response plan is automatically generated based on monitoring data and risk assessment results, specifically including the following steps:

[0186] B10: Based on the comprehensive risk value R total and geological vulnerability coefficient k to divide risk levels and achieve differentiated marking of high-risk areas;

[0187] B20: Combines the path planning algorithm with the resource scheduling model to generate personnel evacuation and rescue resource deployment plans;

[0188] B30: Dynamically adjust emergency plans through real-time data fusion and threshold judgment;

[0189] B40: Convert the plan into visual results and automatically push execution instructions.

[0190] Specifically, the detailed steps are as follows:

[0191] B10: According to the landslide risk value R total and geological vulnerability coefficient κ, the high-risk areas are divided into three levels, as shown in the following formula (22):

[0192]

[0193] Among them, the level I area is marked with a red polygon, and the boundary width is the same as R total Positive correlation; Level II areas are marked with yellow dashed lines, with the historical displacement vector field superimposed Level III areas are filled with translucent blue and marked with the difficulty coefficient P.

[0194] B20: Generate the optimal path based on the improved potential field-ant colony hybrid algorithm, where the risk repulsion F is defined risk and export gravity F exit , and formulate the rules for updating ant colony pheromones; F risk and F exit As shown in the following formula (23), the ant colony pheromone update rule is shown in the following formula (24):

[0195]

[0196] Among them, Q exit is the export attraction constant, d(x,y) is the distance from the point to the risk source; T k is the path time of the kth ant, C safe is the safety factor (with R total negative correlation).

[0197] A multi-objective integer linear programming model is constructed to optimize the rescue resource scheduling. The multi-objective integer linear programming model is shown in the following formula (25):

[0198]

[0199] Among them, the decision variable x i Indicates whether to dispatch equipment to the i-th material point, y j represents the dispatch quantity of the jth type of vehicle, and α=0.3 is the vehicle coverage coefficient.

[0200] B30: Update the risk value R every Δt minutes total , Kalman filtering is used to fuse the UAV and ground sensor data, as shown in the following formula (26):

[0201]

[0202] Where z(k) is the actual value of the sensor, K(k) is the Kalman gain, and H is the measurement matrix;

[0203] If any of the conditions shown in the following formula (27) is met, the emergency plan is regenerated:

[0204]

[0205] B40: Rendering evacuation path heat map, the color mapping function is where R path is the path risk value; the material dispatch route is displayed in pulse animation, and the pulse frequency priority (Unit: Hz); Send the GeoJSON data of the optimal path to the rescuer terminal, including the coordinate points (x i ,y i ) and risk level labels; push resource scheduling lists to the emergency command center, including equipment type, quantity and estimated arrival time.

[0206] In practice, at mines facing severe landslide risk, when weather forecasts warn of rain, the disaster prevention and mitigation system is rapidly activated. Just before the rain, technicians commission drones equipped with high-resolution visible and infrared sensors. As rain clouds approach, the drones fly along a pre-set route, collecting image data of the mining area, slag dumps, and surrounding mountains. Simultaneously, satellite signal receivers at ground monitoring stations near the mine track satellites and simultaneously receive remote sensing data covering topography, geological structure, and more.

[0207] After all the data are aggregated to the data processing center, the cloud and fog obstruction judgment model uses the Otsu algorithm to calculate the visible light and infrared band intensity thresholds based on environmental parameters such as lighting and climate, compares the data collected by the drone, and accurately identifies and marks the cloud and fog obstruction areas.

[0208] Next, the environmental risk factor fusion calculation module runs. It comprehensively assesses the impact of cloud cover, vegetation cover, and surface temperature changes on risk based on the ratio of cloud-obstructed areas to the total mine area, satellite remote sensing Normalized Difference Vegetation Index (NDVI), ground temperature sensors, and satellite thermal infrared data. During rainfall periods, relevant parameters are adjusted in real time based on rainfall duration and substituted into a multi-parameter risk weighting function to derive a dynamic risk weight.

[0209] The spatiotemporal memory prediction model then operates based on risk weights and historical rock texture feature movement data. Weights are calculated using the memory weight allocation rule and, combined with the gradient of the risk weight function, the predicted rock texture feature movement volume and movement vector are derived. A physical constraint correction mechanism then calculates thresholds based on gravitational acceleration, loose layer thickness, and internal friction angle to rationalize the predicted values.

[0210] Once the prediction is complete, a multi-scale validation mechanism begins to evaluate the model. This involves acquiring actual data from monitoring points across the mine, which serve as observations. This validation is performed at the pixel, regional, and global levels, and the error is calculated. If the model prediction deviates from the actual situation, an adaptive correction algorithm constructs the objective function and adjusts the model parameters using gradient descent.

[0211] Throughout the monitoring process, the dynamic risk warning system calculates landslide risk values ​​based on a time-varying risk model. Once the risk value exceeds the warning threshold, the monitoring system notifies management via text messages, internal broadcasts, and flashes red lights on the monitoring platform. Management then quickly identifies high-risk areas, sets up warning signs, and evacuates personnel and equipment, minimizing landslide losses.

[0212] The above is only an embodiment of the present invention, and the common knowledge such as the specific structure and characteristics of the scheme is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A method for dynamic monitoring of mine geological environment based on UAV remote sensing, characterized in that: The following steps are involved: S10: Acquire a panoramic remote sensing image of the target area, identify and mark the vehicle road network in the panoramic remote sensing image, and determine a segmentation size at different positions in the panoramic remote sensing image according to the vehicle road network density, wherein the segmentation size is inversely proportional to the vehicle road network density, and segment the panoramic remote sensing image into a plurality of remote sensing sub-images according to the segmentation size at different positions; S20: Acquire color information of each pixel in the remote sensing sub-image, extract pixel texture features in the remote sensing sub-image based on the color information of each pixel, and divide the pixel texture features into road network texture features and rock texture features; S30: analyzing the movement vectors of the pixels of the rock texture feature in the remote sensing sub-image according to the acquisition time of the panoramic remote sensing image, and calculating the movement volume and movement vector of the rock texture feature according to the movement vectors; S40: When the moving volume is greater than the preset landslide volume, a positional relationship between the moving vector and the road network texture feature is obtained, and a landslide warning is triggered according to the positional relationship.

2. The method for dynamic monitoring of mine geological environment based on UAV remote sensing according to claim 1, characterized in that: The calculation method for determining the segmentation size according to the vehicle road network density in step S10 includes the following steps: S11: The road network density field is established at the pixel point position. The formula for establishing the road network density field is shown in the following formula (1): Among them, (x, y) is the image pixel coordinate, (x i ,y i ) represents the coordinates of the center point of the i-th road network segment, N is the total number of road network segments, A is the normalization coefficient, and σ is the preset Gaussian kernel bandwidth; S12: Construct an adaptive segmentation size mapping model, as shown in the following formula (2): Among them, S(x,y) is the segmentation size at (x,y), S max 、S min denote the maximum and minimum segmentation sizes, ρ max , ρ min are the upper and lower limits of road network density; S13: Use the quadtree segmentation algorithm to implement image partitioning, which satisfies the following formula (3): Among them, R k represents the kth segmented region, and η is the size variation coefficient.

3. The method for dynamic monitoring of mine geological environment based on UAV remote sensing according to claim 2, characterized in that: Calculating the moving volume and moving vector of rock texture features involves the following steps: S31: For the panoramic remote sensing image I of adjacent time series t and I t+Δt , where Δt is the interval time for panoramic remote sensing image acquisition. The optical flow method is used to calculate the displacement vector field V(x,y) of the characteristic pixel points of rock texture. V(x,y)=(u(x,y),v(x,y)), where u(x,y) and v(x,y) represent the displacement components of the pixel point (x,y) in the horizontal and vertical directions, respectively. The calculation is performed by minimizing the energy function. The calculation formula is shown in the following formula (4): Among them, I x ,I y ,I t are the gradients of the image in the x, y, and t directions respectively, and λ is the smoothing coefficient; S32: Based on the displacement vector field V(x,y), calculate the moving volume M of the rock texture feature. The calculation formula of M is shown in the following formula (5): Among them, V i is the displacement vector of the i-th pixel, w i is the weight factor, which is determined by the texture saliency and spatial position of the i-th pixel; S33: Perform spatial clustering on the displacement vector field to extract the dominant movement direction D. The calculation formula of D is shown in the following formula (7): Among them, K is the number of clusters, n k is the number of pixels in the k-th cluster, is the displacement mean vector of the cluster; S34: When the moving volume M exceeds the threshold M textth When the angle θ between the dominant movement direction D and the slope normal is less than 45°, a landslide warning is triggered, where: N is the geometric normal vector of the slope.

4. The method for dynamic monitoring of mine geological environment based on UAV remote sensing according to claim 3, characterized in that: The vehicle texture features are extracted from the road network texture features, and the position change of the vehicle in the road network is analyzed according to the acquisition interval of the panoramic remote sensing image to obtain the vehicle's moving speed, moving direction and the number of vehicles in the road network; the resonance effect of the vehicle vibration on the road network is calculated according to the number of vehicles, moving speed and moving direction, wherein the resonance effect is calculated by a vibration transfer model, and the calculation formula is as follows: Formula (9) As shown: Among them, m i is the equivalent mass of the i-th vehicle, a i is the acceleration, d i is the distance from the vehicle to the potential landslide area, β is the vibration attenuation coefficient; The distribution position of the vehicles is spatially associated with the movement vector of the rock texture feature to calculate the resonance-landslide coupling risk value R. The calculation formula of R is shown in the following formula (10): R=α·M+(1-α)·F(10), Among them, α is the geological vulnerability coefficient; when R exceeds the preset threshold and the angle between the rock movement direction and the main vibration direction is less than the set angle, the resonance landslide warning is triggered.

5. The method for dynamic monitoring of mine geological environment based on UAV remote sensing according to claim 4, characterized in that: When calculating the resonance-landslide coupling risk value, the weights of adjacent remote sensing sub-images are dynamically adjusted according to the size of the remote sensing sub-image. Specifically, the following steps are included: A10: Establish a weight distribution model for adjacent remote sensing sub-images. The weight calculation formula is shown in the following formula (11): Among them, w adj is the weight of the jth adjacent remote sensing sub-image to the current remote sensing sub-image i, A i is the segmentation size area of ​​the current remote sensing sub-image, d ij is the spatial distance between the center points of the two remote sensing sub-images, d max is the maximum impact distance threshold; A20: Filter valid adjacent remote sensing sub-images according to weights, satisfying the following formula (12): Among them, N adj is the number of adjacent remote sensing sub-images finally selected, τ is the weight accumulation threshold, w max is the maximum single weight value; A30: Calculate Moving Balance Index B i , B i The calculation formula is shown in the following formula (13): in, is the moving vector of the rock texture feature of the jth adjacent remote sensing sub-image; A40: Modify the landslide risk value R′. The correction formula of R′ is shown in the following formula (14): Among them, α is the balance factor, β is the attenuation coefficient, when B i When <0.3, a balance warning is triggered.

6. The method for dynamic monitoring of mine geological environment based on UAV remote sensing according to claim 5, characterized in that: The mining area is divided into different risk areas according to the moving vector of the rock texture characteristics. When determining the recommended dumping direction, the area different from the risk area where the loose direction is located is preferred as the safe dumping area. The loose direction is calculated using the following formula (15): in, is the moving vector of the rock texture feature in the adjacent remote sensing sub-image j; When regularly acquiring the movement volume of rock texture features of adjacent remote sensing sub-images, if the movement volume is inconsistent with the preset threshold, the current remote sensing sub-image is used as the relay monitoring area, and a remote sensing sub-image in a risk area different from the relay monitoring area is randomly selected as the verification area. The movement vector verification data is sent to the verification area through the relay monitoring area. If the motion vector verification data received in the verification area matches the local monitoring data, the stability score of the relay monitoring area is lowered; if it does not match, the relay monitoring area is marked as a high-risk area, and adjacent remote sensing sub-images with the same risk area as the relay monitoring area are selected one by one as auxiliary verification areas, and the motion vector verification data is sent to the verification area through the auxiliary verification areas.

7. The method for dynamic monitoring of mine geological environment based on UAV remote sensing according to claim 6, characterized in that: The following steps are also included: S501: Perform cloud and fog detection on the current panoramic remote sensing image and calculate the cloud and fog coverage area ratio C. If C>0, it is determined that the current weather is rainy, and the image is divided into cloud-blocked areas and visible areas according to the cloud distribution; S502: Acquire the most recent panoramic remote sensing image without cloud or fog as the first image, and use the currently acquired panoramic remote sensing image with cloud or fog as the second image. Extract the pixel color rendering of the visible area of ​​the second image and record it as the first color rendering S1. Extract the pixel color rendering of the same position in the first image and record it as the second color rendering S2. Calculate the color rendering difference ratio ΔS according to the following formula (16): For cloud-blocked areas, the humidity impact value H of the blocked area is obtained based on ΔS of the adjacent visible area and the cloud thickness estimated by infrared band penetration. S503: Modify the weight distribution of the remote sensing sub-image according to the humidity impact value H, and predict the landslide risk value R of the visible area based on the adjusted weight and combined with the historical movement volume data. pred ; S504: Calculate the real risk value R based on the actual collected visible area landslide data real , R real and the predicted value R pred Compare and obtain the error E. If E> the preset error threshold, use the back propagation algorithm to adjust the calculation parameters of the humidity impact value H until the error converges; S505: Apply the optimized model to the cloud-covered area and output the predicted landslide risk value R final , according to R final A landslide warning is triggered in the area and it is marked as a high-risk area requiring manual verification.

8. The method for dynamic monitoring of mine geological environment based on UAV remote sensing according to claim 7, characterized in that: In the road network texture features in the visible area, the target detection algorithm is used to identify and mark the vehicle, and the instantaneous speed v of the vehicle before entering the cloud-covered area is extracted. in , record the time t when the vehicle enters the foggy area in ; According to the road network topology and vehicle speed v in , predict the theoretical time T for a vehicle to pass through the cloud and fog obstruction area pred , Among them L cloud is the length of the road network in the area blocked by fog; the actual passing time t is calculated by the moment when the vehicle reappears in the visible area out , calculate the actual passing time T real , T real =t out -t in ; The definition of the traffic difficulty coefficient P reflects the degree of abnormal road conditions: If P is greater than the preset threshold, it is determined that there is a traffic obstruction in the foggy area; the equivalent passing speed v of the vehicle is dynamically corrected according to the P value. adj , v adj The calculation formula is shown in the following formula (18): v adj =v in ·(1-ηP) (18), Where η is the damping coefficient, 0<η<1; Incorporating the access difficulty coefficient P into the landslide risk prediction model in fog-covered areas: Correcting the humidity impact value H new As shown in the following formula (19): H new =H·(1+μP) (19), where μ is the coupling coefficient, calibrated by geological permeability; Update the weight allocation model to increase the monitoring priority of high P value areas; when the revised landslide risk value R final When the threshold is exceeded and P continues to rise, it is determined that the geological instability in the area has been aggravated by traffic obstruction, triggering a "vehicle-landslide" coupled warning.

9. The method for dynamic monitoring of mine geological environment based on UAV remote sensing according to claim 8, characterized in that: The comprehensive risk index R is established based on the rock movement volume M, the resonance-landslide coupling risk value R, the travel difficulty coefficient P and the humidity impact value H. total The weighted fusion model is shown in the following formula (20): Among them, ω1~ω4 are weight coefficients, which are determined by the analytic hierarchy process AHP and meet the following requirements: ), M th is the landslide volume threshold; According to R total A multi-level warning system will be established, and different response measures will be formulated for different levels of warnings.

10. The method for dynamic monitoring of mine geological environment based on UAV remote sensing according to claim 9, characterized in that: After an early warning is triggered, an emergency response plan is automatically generated based on monitoring data and risk assessment results. The specific steps include: B10: Based on the comprehensive risk value R total and geological vulnerability coefficient k to divide risk levels and achieve differentiated marking of high-risk areas; B20: Combines the path planning algorithm with the resource scheduling model to generate personnel evacuation and rescue resource deployment plans; B30: Dynamically adjust emergency plans through real-time data fusion and threshold judgment; B40: Convert the plan into visual results and automatically push execution instructions.

Images