Intelligent rainwater accumulation early warning method and system for strip mine pit based on image recognition
By constructing a high-precision DEM model and improved water fall energy loss analysis, combining image recognition and hydrodynamic models, and introducing intelligent risk warning, the accuracy and intelligence problems of rainwater accumulation monitoring and warning in open-pit mines have been solved, achieving more accurate and efficient risk management and emergency response.
Patent Information
- Application Number
- CN202510995743.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-18
- Publication Date
- 2025-09-16
AI Technical Summary
In existing technologies, digital elevation models are distorted, the simulation of water flow physical processes is rough, the prediction results are inaccurate, and there is a lack of intelligent risk warnings, resulting in inaccurate and unreliable monitoring and warning of rainwater accumulation in open-pit mines.
By constructing a high-precision DEM model, combining the improved Saint-Venant equation and MineSegNet algorithm, dynamically coupling image recognition and hydrodynamic model, and introducing the intelligent risk warning model MAPGRPO algorithm, a multi-level and multi-dimensional intelligent risk warning strategy is generated.
It achieves accurate prediction and timely warning of rainwater accumulation in open-pit mines, provides more accurate and efficient risk management and emergency response, and can generate flexible early warning strategies based on multiple factors, improving the reliability of prediction results and the timeliness of emergency decision-making.
Smart Images
Figure CN120656288A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of intelligent early warning technology, and in particular relates to an open-pit mine intelligent rainwater accumulation early warning method and system based on image recognition. Background Art
[0002] Open-pit mines are prone to waterlogging during the rainy season or sudden heavy rainfall. This not only disrupts normal production operations but can also cause serious safety incidents such as slope instability and debris flows, threatening life and property. Therefore, monitoring and early warning of rainwater accumulation in open-pit mines has become an important research area. With the development of drones and artificial intelligence (AI), the use of image recognition technology for waterlogging monitoring has become possible. However, how to effectively integrate image information into hydrodynamic models and combine it with intelligent decision-making mechanisms to achieve accurate early warnings remains an urgent challenge.
[0003] The existing technology has the following defects: 1) Digital elevation model distortion: The digital elevation model constructed by existing methods has limited accuracy, which leads to distortion of the foundation for subsequent hydrodynamic model construction, and the simulation results deviate significantly from the actual situation; 2) Rough simulation of water flow physics: Traditional hydrodynamic models often simplify the flow of water over complex mine terrain (such as steps), particularly ignoring the energy loss of water as it falls. This makes it difficult to accurately predict the distribution, depth, and flow rate of accumulated water, especially in areas near steps. 3) Rough prediction results: Existing technologies often use image recognition (such as drone aerial photography) and hydrodynamic models separately, lacking an effective information fusion mechanism. While image recognition can provide intuitive waterlogging information, it is difficult to quantify. Hydrodynamic models can provide dynamic simulations, but they may deviate from reality. The failure to effectively combine these two information results in inaccurate and unreliable predictions (such as changes in waterlogged area, depth, range, and location). 4) Lack of intelligent risk warning: Existing early warning systems are mostly based on simple threshold judgments or single model outputs. The warning strategies generated are often "one-size-fits-all" and lack specificity and flexibility. They usually only issue water accumulation alarms, but fail to intelligently generate early warning strategies with specific action plans (such as personnel evacuation routes, equipment transfer locations, and reinforcement measures for key areas) based on multiple factors such as water accumulation prediction results, mine layout, and personnel and equipment distribution. Summary of the Invention
[0004] In order to solve the problems of digital elevation model distortion, rough simulation of water flow physical processes, rough prediction results and lack of intelligent risk warning in the existing technology, the purpose of the present invention is to provide an open-pit mine intelligent rainwater accumulation warning method and system based on image recognition.
[0005] The technical solution adopted in the present invention is: An intelligent rainwater accumulation early warning method for open-pit mines based on image recognition includes the following steps: Collect topographic data of the open pit, build a DEM model based on the topographic data, and define the water fall energy loss based on the DEM model; Collect image data and meteorological data of the open pit, and use the image recognition model to perform image recognition on the image data to obtain image recognition results; A hydrodynamic analysis model was constructed based on the open pit's meteorological data, DEM model, water flow energy loss, and image recognition results; Start the hydrodynamic analysis model to simulate the rainwater accumulation in the open pit at a future time point and obtain the rainwater accumulation prediction results; According to the rainwater accumulation prediction results, the intelligent risk warning model is used to generate intelligent risk warnings, and the obtained intelligent risk warning strategies are executed.
[0006] Furthermore, the terrain data of the open pit is collected, a DEM model is constructed based on the terrain data, and the corresponding water flow fall energy loss is obtained based on the DEM model, including the following steps: Use a drone cluster equipped with a 3D laser scanner to collect terrain data of the open pit and pre-process the terrain data to obtain pre-processed terrain data; Based on the pre-processed terrain data, a DEM model of the terraced terrain of the open pit is constructed. The DEM model is then annotated with key information and gridded to obtain several computational grids. According to the DEM model of the open pit, the improved Saint-Venant equation is used to obtain the corresponding water fall energy loss.
[0007] Furthermore, based on the DEM model of the open pit, the improved Saint-Venant equation is used to define the water fall energy loss, which includes the following steps: Use the curvature analysis algorithm to locate the edge of the steps in the terraced terrain of the mine in the DEM model and obtain the corresponding vertical drop of the steps; According to the vertical drop of the steps, the empirical loss coefficient and the instantaneous momentum loss term are defined, and an improved momentum equation is constructed based on the empirical loss coefficient and the instantaneous momentum loss term; The improved momentum equation is used to characterize the energy loss of water falling in an open pit.
[0008] Furthermore, the image recognition model is built based on the improved MineSegNet algorithm, and the main network architecture of the image recognition model includes an encoder built based on the ResNet algorithm, a decoder built based on the Deconvolution-U-Net algorithm, and a classifier built based on the Sigmoid activation function, which are connected in sequence. The decoder is connected to an attention module built based on the SE Block-CBAM algorithm. The input end of the attention module is also connected to a terrain feature extraction module built based on the CNN algorithm and a ripple feature extraction module built based on the FT-Daubechies algorithm. The image recognition model is set with a loss function based on the Saint-Venant equation constraints.
[0009] Furthermore, image data and meteorological data of the open pit are collected, and an image recognition model is used to perform image recognition on the image data to obtain an image recognition result, including the following steps: Using drone swarms equipped with high-definition cameras to collect image data from the open-pit mine, and collecting meteorological data from the open-pit mine from external IoT rain gauges in the mining area; Preprocessing the image data and the meteorological data to obtain preprocessed image data and preprocessed meteorological data; Use the decoder of the image recognition model to extract multi-level image features of the preprocessed image data; Use the terrain feature extraction module of the image recognition model to extract the terrain features of the DEM model; Use the ripple feature extraction module of the image recognition model to extract the ripple features of the preprocessed image data; According to the dynamic attention weight, the attention module of the image recognition model is used to perform weighted fusion on multi-level image features, terrain features, and ripple features to obtain weighted fusion features; Based on the weighted fusion features, the decoder of the image recognition model is used to generate a segmentation mask to obtain a segmentation mask of the same size as the preprocessed image data; Based on the segmentation mask and pre-processed image data, the classifier of the image recognition model is used to perform classification, and image recognition results including waterlogging areas and runoff areas are obtained; The pixel coordinates of the waterlogged area in the image recognition result are mapped to the grid coordinates of the calculation grid corresponding to the DEM model.
[0010] Furthermore, based on the open pit meteorological data, DEM model, water flow drop energy loss, and image recognition results, a hydrodynamic analysis model was constructed, which includes the following steps: Combining the continuity equation with the improved momentum equation for water drop energy loss, a two-dimensional shallow water equation is obtained. The computational grid of the DEM model is spatially discretized to obtain the initial hydrodynamic analysis model. According to the height difference information of the DEM model and the coordinates of the water-logged pixels in the water-logged area in the image recognition results, the initial water depth field of the calculation grid of the initial hydrodynamic analysis model is set; According to the initial water depth field and the runoff area in the image recognition results, the initial flow velocity field of the computational grid of the initial hydrodynamic analysis model is set; According to the meteorological data and image recognition results of the open pit, the boundary conditions of the initial hydrodynamic analysis model are set to obtain the final hydrodynamic analysis model.
[0011] Furthermore, a hydrodynamic analysis model is started to simulate the rainwater accumulation situation in the open pit at a future time point to obtain a rainwater accumulation prediction result, including the following steps: Start the hydrodynamic analysis model and continuously collect simulation results of rainwater accumulation in the open pit at future time points. The simulation results include the water depth field prediction results and flow velocity field prediction results of each calculation grid in the hydrodynamic analysis model, as well as the scour risk heat map. The simulation results of rainwater accumulation are post-processed to obtain rainwater accumulation prediction results; the rainwater accumulation prediction results include the prediction results of the accumulation area, the prediction results of the change of the accumulation depth, range and position over time corresponding to the accumulation area prediction results, the prediction results of the water flow path in the mine, and the prediction results of the change of the slope infiltration line.
[0012] Furthermore, the intelligent risk warning model is constructed based on the MAPGRPO algorithm, and the intelligent risk warning model includes a central-level intelligent risk warning generation layer, a coordination layer, and a system-level intelligent risk warning generation layer connected in sequence. The central-level intelligent risk warning generation layer is provided with a first multi-optimization target set and a central-level intelligent agent, and the system-level intelligent risk warning generation layer is provided with a second multi-optimization target set and several parallel system-level intelligent agents.
[0013] Furthermore, based on the rainwater accumulation prediction results, the intelligent risk warning model is used to generate an intelligent risk warning, and the obtained intelligent risk warning strategy is executed, including the following steps: Based on the rainwater accumulation prediction results, the central-level intelligent agent of the central-level intelligent risk warning strategy generation layer of the intelligent risk warning model is used to generate an intelligent risk warning strategy to obtain a central-level intelligent risk warning strategy; Through the coordination layer of the intelligent risk warning model, the rainwater accumulation prediction results and the central-level intelligent risk warning strategy are input into several parallel system-level intelligent agents in the system-level intelligent risk warning strategy generation layer; Based on the rainwater accumulation prediction results and the center-level intelligent risk warning strategy, several system-level intelligent agents in the system-level intelligent risk warning strategy generation layer are used to generate intelligent risk warning strategies, and several system-level intelligent risk warning strategies are obtained; Integrate the center-level intelligent risk warning strategy and several system-level intelligent risk warning strategies to obtain the intelligent risk warning strategy, publish the intelligent risk warning strategy to all execution systems, and execute the intelligent risk warning strategy based on the execution system.
[0014] An open pit intelligent rainwater accumulation early warning system based on image recognition is used to implement an open pit intelligent rainwater accumulation early warning method. The system includes a DEM model construction unit, an image recognition unit, a hydrodynamic analysis model construction unit, a rainwater accumulation simulation unit and an intelligent risk early warning unit connected in sequence.
[0015] The beneficial effects of the present invention are: The present invention provides an intelligent rainwater accumulation warning method and system for open-pit mines based on image recognition. By using a cluster of drones equipped with three-dimensional laser scanners to collect terrain data, a high-precision DEM model is constructed, which can accurately depict the complex features of the stepped terrain of the mine. At the same time, the energy loss of water flow is incorporated into the model, making the hydrodynamic model closer to physical reality, overcoming the problem of simplification and distortion of traditional models, and laying a solid foundation for subsequent predictions; based on the improved Saint-Venant equation, the step edge is located through curvature analysis and the energy loss term is defined, so that the model can more realistically simulate the falling, energy loss and diffusion process of water flow at the step, thereby improving the understanding of the water flow movement law and the simulation accuracy; the image recognition model based on the improved MineSegNet algorithm is dynamically coupled with the hydrodynamic model, and image recognition can provide real-time water accumulation areas, runoff areas, etc., and the hydrodynamic model provides The dynamic and continuous evolution of water accumulation, and the fusion of information from the two make the prediction results (water accumulation area, depth, range, location change, water flow path, slope infiltration line, etc.) more accurate and reliable; an intelligent risk warning model based on the MAPGRPO algorithm is introduced, which can generate multi-level and multi-dimensional intelligent risk warning strategies (such as evacuation of specific areas, equipment transfer, temporary drainage point setting, etc.) based on accurate rainwater accumulation prediction results and comprehensive consideration of multiple optimization goals such as personnel safety, equipment protection, and production efficiency. This avoids the "one-size-fits-all" problem of traditional warnings and achieves more accurate and efficient risk management and emergency response; by using meteorological data as the input of the hydrodynamic model and combining it with high-precision DEM models and image recognition results, the hydrodynamic model can respond to weather changes more dynamically and sensitively, especially when dealing with sudden heavy rainfall, and can provide more timely and reliable warnings and decision-making support.
[0016] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 It is a flowchart of the open-pit mine intelligent rainwater accumulation early warning method based on image recognition in the present invention.
[0018] Figure 2 This is a structural block diagram of the open-pit mine intelligent rainwater accumulation early warning system based on image recognition in the present invention. DETAILED DESCRIPTION
[0019] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.
[0020] Example 1: like Figure 1 As shown, this embodiment provides an open-pit mine intelligent rainwater accumulation early warning method based on image recognition, including the following steps: S1: Collect the topographic data of the open pit and build a digital elevation model (DEM) based on the topographic data. Then, define the water fall energy loss based on the DEM model. The steps include: S1-1: Use a drone cluster equipped with a 3D laser scanner to collect terrain data of the open pit and pre-process the terrain data to obtain pre-processed terrain data; S1-2: Based on the pre-processed terrain data, a DEM model of the terraced terrain of the open pit is constructed. Key information is annotated and the DEM model is meshed to obtain several computational grids. Key information includes road slope elevation, step edge location, step vertical drop, existing drainage ditch location, and step edge curvature, which are crucial for subsequent prediction of step edges and calculation of energy losses. Use unstructured grids for meshing, and the computational grid should be able to finely depict key areas such as step edges and drainage ditches; S1-3: Based on the DEM model of the open pit, the modified Saint-Venant equation is used to obtain the corresponding water drop energy loss, including the following steps: S1-3-1: Use the curvature analysis algorithm to locate the edge of the steps in the terraced terrain of the mine in the DEM model and obtain the corresponding vertical drop of the steps; The formula is:
[0021] Where, for point The curvature change at , usually corresponds to the area with drastic curvature change, such as the edge of a step; for point the height of the place; For height Coordinates The second-order partial derivative reflects the Changes in curvature of direction; For height Coordinates The second-order partial derivative reflects the Changes in curvature of direction; S1-3-2: Define the empirical loss coefficient and instantaneous momentum loss term based on the vertical drop of the steps, and construct an improved momentum equation based on the empirical loss coefficient and instantaneous momentum loss term; The formula is:
[0022] Where, is the experience loss coefficient; is the vertical drop of the steps (m); is the incoming flow velocity (m / s); is the depth of incoming water (m); is the acceleration due to gravity; is the hyperbolic tangent function;
[0023] Where, For stairs The instantaneous momentum loss term; is the step indication amount; For stairs Experience loss coefficient; is the amount of water passing through the section per unit time; is the cross-sectional area of water flow; is the Dirac function; is the hyperbolic tangent function; is the coordinate variable; For stairs location;
[0024] Where, is the amount of water passing through the section per unit time About time The first partial derivative of ; For water depth; is the slope of the riverbed; is the friction slope; is the total number of step edges; is the total energy loss of the step; Water depth Coordinates The first partial derivative of ; is the time indicator; S1-3-3: Characterize the energy loss of water falling in open pits using the improved momentum equation; S2: Collect image data and meteorological data of the open pit, and use the image recognition model to perform image recognition on the image data to obtain image recognition results; The image recognition model is built based on the improved Mine Segmentation Network (MineSegNet) algorithm. The main network architecture of the image recognition model includes an encoder built based on the Residual Neural Network (ResNet) algorithm, a decoder built based on the Transposed Convolution (Deconvolution)-U-Net Convolutional Network (U-Net) algorithm, and a classifier built based on the Sigmoid activation function. The decoder is connected to an attention module built based on the Squeeze-and-Excitation Block (SE Block)-Convolutional Block Attention Module (CBAM) algorithm. The input of the attention module is also connected to a terrain feature extraction module built based on the Convolutional Neural Network (CNN) algorithm and a ripple feature extraction module built based on the Fourier Transform (FT)-Daubechies algorithm. The encoder is used to extract multi-level features of the image, and the decoder includes an upsampling module based on the Deconvolution algorithm and a jump connection module based on the U-Net algorithm to combine the high-level semantic features extracted by the encoder with the low-level spatial detail features, and finally generate a segmentation mask of the same size as the input image; the attention module includes a channel attention layer based on the SEBlock algorithm and a spatial attention layer based on the CBAM algorithm, while considering the spatial distribution of multi-level image features, terrain features and ripple features for weighted fusion; the terrain feature extraction module is used to extract terrain features in the DEM model, including local slope, curvature, and distance from drainage ditches; The ripple feature extraction module is used to extract edge and texture information features of the image at different scales. The Fourier transform module based on the FT algorithm is used to perform Fourier transform on local areas of the image and analyze the energy distribution in a specific frequency range. Waterlogged surfaces usually have high energy in a specific frequency range. The edge and texture analysis module constructed using the Daubechies algorithm is then used to analyze the edge and texture information of the image at different scales. Ripples are usually obvious at specific scales. Finally, the image gradient is calculated, the structural tensor is analyzed, and features related to surface roughness are extracted to obtain texture features. The classifier usually uses one output channel (two-category classification: waterlogged / non-waterlogged) and is connected to a Sigmoid activation function to output the probability of each pixel belonging to waterlogged areas. The image recognition model is set with a loss function based on the Saint-Venant equation constraint; The formula is:
[0025] Where, is the total loss function; are the cross entropy loss function, boundary loss function, terrain consistency constraint term, confluence consistency constraint term, and ripple consistency constraint term; are the first weight coefficient, the second weight coefficient, the third weight coefficient, the fourth weight coefficient, and the fifth weight coefficient;
[0026] Where, is the pixel coordinate; is the segmentation mask of the waterlogged area; Pixel DEM model data; Pixel local slope; It is a confluence consistency penalty function that outputs a larger value when the pixel is at a high place or on a steep slope. It encourages water accumulation areas to appear mainly in depressions or gentle slopes rather than steep slopes or high places.
[0027] Where, Pixels distance to the nearest drain; It is a confluence consistency penalty function that outputs a larger value when the pixel is far away from the drainage ditch; it encourages the spatial relationship between the waterlogged area, drainage ditch, and natural water collection area to conform to physical laws;
[0028] Where, Pixels The ripple characteristics of is the ripple consistency penalty function, which outputs a larger value when the ripple feature intensity of the pixel is high. The negative sign indicates that this is a loss term that needs to be minimized. It encourages areas where ripple features are detected in the image to be identified as water accumulation. Collecting image data and meteorological data from an open pit mine, and using an image recognition model to perform image recognition on the image data to obtain image recognition results, includes the following steps: S2-1: Use a swarm of drones equipped with high-definition cameras to collect image data from the open-pit mine, and collect meteorological data from the open-pit mine from an external IoT rain gauge in the mining area; S2-2: Preprocessing the image data and the meteorological data to obtain preprocessed image data and preprocessed meteorological data; S2-3: Use the decoder of the image recognition model to extract multi-level image features of the preprocessed image data; S2-4: Use the terrain feature extraction module of the image recognition model to extract the terrain features of the DEM model; S2-5: Use the ripple feature extraction module of the image recognition model to extract the ripple features of the preprocessed image data; S2-6: Based on the dynamic attention weights, the attention module of the image recognition model is used to perform weighted fusion on the multi-level image features, terrain features, and ripple features to obtain weighted fusion features. S2-7: Based on the weighted fusion features, use the decoder of the image recognition model to generate a segmentation mask to obtain a segmentation mask of the same size as the preprocessed image data; S2-8: Using the classifier of the image recognition model to perform classification based on the segmentation mask and the preprocessed image data, an image recognition result including the waterlogged area and the runoff area is obtained; S2-9: Mapping the pixel coordinates of the waterlogged area in the image recognition result to the grid coordinates of the calculation grid corresponding to the DEM model; S3: Based on the open pit meteorological data, DEM model, water flow energy loss, and image recognition results, a hydrodynamic analysis model is constructed, including the following steps: S3-1: Combining the continuity equation and the improved momentum equation for water drop energy loss, a two-dimensional shallow water equation is obtained. The computational grid of the DEM model is spatially discretized to obtain the initial hydrodynamic analysis model. The formula is:
[0029] Where, is the total water depth; for Horizontal flow velocity in the direction; for Horizontal flow velocity in the direction; Total water depth About time The first partial derivative of ; For coordinates Total water depth Product coordinates The first partial derivative of ; For coordinates Total water depth Product coordinates The first partial derivative of ; S3-2: Based on the height difference information of the DEM model and the pixel coordinates of the waterlogged area in the image recognition results, set the initial water depth field of the calculation grid of the initial hydrodynamic analysis model; Based on the height difference information of the DEM model and the pixel coordinates of the waterlogged area in the image recognition result, an estimate is made to obtain the estimated waterlogged depth, and the estimated waterlogged depth is set as the initial water depth of the calculation grid of the initial hydrodynamic analysis model; for areas where waterlogging is not recognized in the image and the DEM shows flat land or gentle slopes, the initial water depth is set to 0 or a small value; for areas where waterlogging is not recognized in the image but the DEM shows depressions or possible waterlogging areas, the initial water depth can be set to a non-zero value based on the terrain and possible waterlogging conditions; S3-3: According to the initial water depth field and the runoff area in the image recognition results, set the initial flow velocity field of the calculation grid of the initial hydrodynamic analysis model; The initial flow velocity is usually set to 0. If the image identifies obvious runoff (such as a stream flowing down a step), the initial flow velocity is set at the boundary or source of the corresponding area; S3-4: Based on the meteorological data and image recognition results of the open pit, the boundary conditions of the initial hydrodynamic analysis model are set to obtain the final hydrodynamic analysis model; Rainfall intensity is set based on meteorological data (e.g., Chicago rainfall patterns) and used as the boundary condition for rainfall inflow. Boundary conditions such as free outflow or fixed water levels are set at locations such as the mine exit and the end of the drainage ditch. Mine slopes, step edges (non-flow side), and buildings are set as non-slip solid wall boundaries. The boundaries of the waterlogged areas identified by image recognition are used as known boundary conditions at the initial moment. S4: Start the hydrodynamic analysis model to simulate the rainwater accumulation situation in the open pit at a future time point to obtain the rainwater accumulation prediction result, including the following steps: S4-1: Select an appropriate numerical solver, set the time step and total simulation time, start the hydrodynamic analysis model, and continuously collect simulation results of rainwater accumulation in the open-pit mine at future time points. The simulation results of rainwater accumulation include the water depth field prediction results and flow velocity field prediction results of each calculation grid in the hydrodynamic analysis model, as well as the scour risk heat map. The water depth prediction result is the water depth of each calculation grid at each future time step (e.g., every minute, every 5 minutes); the flow velocity prediction result is the magnitude and direction of the water flow velocity in each calculation grid; the scour risk heat map is the distribution of the scour risk index calculated based on flow velocity, water depth, terrain slope, etc. S4-2: Post-process the simulation results of rainwater accumulation to obtain rainwater accumulation prediction results. The rainwater accumulation prediction results include the prediction results of the accumulation area, the prediction results of the change of the accumulation depth, range and location of the accumulation area over time, the prediction results of the water flow path in the mine, and the prediction results of the change of the slope infiltration line. S5: Based on the rainwater accumulation prediction results, use the intelligent risk warning model to generate intelligent risk warnings and execute the obtained intelligent risk warning strategies; The intelligent risk warning model is constructed based on the Multi-Agent Parallel Group Relative Policy Optimization (MAPGRPO) algorithm. The intelligent risk warning model includes a central-level intelligent risk warning generation layer, a coordination layer, and a system-level intelligent risk warning generation layer, which are connected in sequence. The central-level intelligent risk warning generation layer is provided with a first set of multiple optimization objectives and central-level agents, and the system-level intelligent risk warning generation layer is provided with a second set of multiple optimization objectives and several parallel system-level agents. The central-level intelligent risk warning generation layer is the top layer of the model. It is responsible for setting preliminary risk warning targets and generating a macro-strategy framework from a global perspective, focusing on the overall risk situation. The coordination layer is responsible for transmitting information, coordinating actions, and resolving potential conflicts between central-level and system-level agents to ensure the coherence and consistency of strategies at all levels. The system-level intelligent risk warning generation layer is the bottom layer of the model and contains multiple parallel system-level agents. Each agent is responsible for focusing on a specific area or type of risk within the mine (for example, the stability of a slope, the flooding risk of a certain area, etc.), and generates specific, localized warning strategies based on central-level instructions and local information. The first set of multiple optimization objectives is global, such as minimizing overall safety risks, ensuring the safe evacuation of personnel, protecting critical equipment, and optimizing the initial deployment of resources (such as pumping equipment). Based on these objectives and input information (waterlogging prediction results), the central-level agent generates preliminary, macro-level strategies (central-level intelligent risk warning strategies). The second multi-optimization objective set has more specific and localized goals, for example: the infiltration line of a specific slope must not exceed a safety threshold, the water depth in a specific area must not exceed a certain value, and obstacles in a specific water flow path must be cleared promptly. Several parallel system-level agents each generate specific and executable risk warning strategies (system-level intelligent risk warning strategies) based on their assigned goals and received information (waterlogging prediction results + central-level strategies). S5-1: Based on the rainwater accumulation prediction results, use the intelligent risk warning model to generate intelligent risk warnings and implement the obtained intelligent risk warning strategy, including the following steps: S5-2: Based on the rainwater accumulation prediction results, the central-level intelligent agent in the central-level intelligent risk warning strategy generation layer of the intelligent risk warning model generates intelligent risk warning strategies, such as "focus on the slopes and central depression areas in the northwest region", "prepare to deploy pumping equipment in the southeast region", and "notify relevant departments to enter the second-level warning state". S5-3: Through the coordination layer of the intelligent risk warning model, the rainwater accumulation prediction results and the central-level intelligent risk warning strategy are input into several parallel system-level intelligent agents in the system-level intelligent risk warning strategy generation layer; S5-4: Based on the rainwater accumulation prediction results and the center-level intelligent risk warning strategy, several system-level intelligent agents in the system-level intelligent risk warning strategy generation layer are used to generate intelligent risk warning strategies. For example, the agent responsible for the northwest slope generates a strategy: "Monitor the infiltration line at point A on the slope. If it exceeds threshold X, immediately issue a red alert and notify the reinforcement team." The agent responsible for the central depression generates a strategy: "It is expected that the water accumulation depth at point B will reach Y meters. Start pumping station C in advance and issue an orange alert to the workers." S5-5: Integrate the center-level intelligent risk warning strategy and several system-level intelligent risk warning strategies to obtain the intelligent risk warning strategy, publish the intelligent risk warning strategy to all execution systems, and execute the intelligent risk warning strategy based on the execution system.
[0030] Example 2: like Figure 2 As shown, this embodiment provides an open-pit mine intelligent rainwater accumulation early warning system based on image recognition, which is used to implement an open-pit mine intelligent rainwater accumulation early warning method. The system includes a DEM model construction unit, an image recognition unit, a hydrodynamic analysis model construction unit, a rainwater accumulation simulation unit, and an intelligent risk early warning unit connected in sequence; The DEM model building unit is used to collect terrain data of the open pit, build a DEM model based on the terrain data, and define the water flow fall energy loss based on the DEM model; An image recognition unit is used to collect image data and meteorological data of the open pit, and use an image recognition model to perform image recognition on the image data to obtain an image recognition result; A hydrodynamic analysis model building unit is used to build a hydrodynamic analysis model based on the open pit meteorological data, DEM model, water flow drop energy loss and image recognition results; The rainwater accumulation simulation unit is used to start the hydrodynamic analysis model, simulate the rainwater accumulation situation in the open pit at a future time point, and obtain the rainwater accumulation prediction result; The intelligent risk warning unit is used to generate intelligent risk warnings based on rainwater accumulation prediction results using an intelligent risk warning model and to execute the obtained intelligent risk warning strategy.
[0031] The present invention provides an intelligent rainwater accumulation warning method and system for open-pit mines based on image recognition. By using a cluster of drones equipped with three-dimensional laser scanners to collect terrain data, a high-precision DEM model is constructed, which can accurately depict the complex features of the stepped terrain of the mine. At the same time, the energy loss of water flow is incorporated into the model, making the hydrodynamic model closer to physical reality, overcoming the problem of simplification and distortion of traditional models, and laying a solid foundation for subsequent predictions; based on the improved Saint-Venant equation, the step edge is located through curvature analysis and the energy loss term is defined, so that the model can more realistically simulate the falling, energy loss and diffusion process of water flow at the step, thereby improving the understanding of the water flow movement law and the simulation accuracy; the image recognition model based on the improved MineSegNet algorithm is dynamically coupled with the hydrodynamic model, and image recognition can provide real-time water accumulation areas, runoff areas, etc., and the hydrodynamic model provides The dynamic and continuous evolution of water accumulation, and the fusion of information from the two make the prediction results (water accumulation area, depth, range, location change, water flow path, slope infiltration line, etc.) more accurate and reliable; an intelligent risk warning model based on the MAPGRPO algorithm is introduced, which can generate multi-level and multi-dimensional intelligent risk warning strategies (such as evacuation of specific areas, equipment transfer, temporary drainage point setting, etc.) based on accurate rainwater accumulation prediction results and comprehensive consideration of multiple optimization goals such as personnel safety, equipment protection, and production efficiency. This avoids the "one-size-fits-all" problem of traditional warnings and achieves more accurate and efficient risk management and emergency response; by using meteorological data as the input of the hydrodynamic model and combining it with high-precision DEM models and image recognition results, the hydrodynamic model can respond to weather changes more dynamically and sensitively, especially when dealing with sudden heavy rainfall, and can provide more timely and reliable warnings and decision-making support.
[0032] The present invention is not limited to the above optional embodiments. Anyone can derive various other forms of products based on the teachings of the present invention. The above specific embodiments should not be construed as limiting the scope of protection of the present invention. The scope of protection of the present invention shall be based on the scope defined in the claims, and the description can be used to interpret the claims.
Claims
1. An intelligent rainwater accumulation early warning method for open-pit mines based on image recognition, characterized by: The steps include: Collect topographic data of the open pit, build a DEM model based on the topographic data, and define the water fall energy loss based on the DEM model; Collect image data and meteorological data of the open pit, and use the image recognition model to perform image recognition on the image data to obtain image recognition results; A hydrodynamic analysis model was constructed based on the open pit's meteorological data, DEM model, water flow energy loss, and image recognition results; Start the hydrodynamic analysis model to simulate the rainwater accumulation in the open pit at a future time point and obtain the rainwater accumulation prediction results; According to the rainwater accumulation prediction results, the intelligent risk warning model is used to generate intelligent risk warnings, and the obtained intelligent risk warning strategies are executed.
2. The method for early warning of rainwater accumulation in open-pit mines based on image recognition according to claim 1, characterized in that: Collect the topographic data of the open pit, build a DEM model based on the topographic data, and obtain the corresponding water flow drop energy loss based on the DEM model, including the following steps: Use a drone cluster equipped with a 3D laser scanner to collect terrain data of the open pit and pre-process the terrain data to obtain pre-processed terrain data; Based on the pre-processed terrain data, a DEM model of the terraced terrain of the open pit is constructed. The DEM model is then annotated with key information and gridded to obtain several computational grids. According to the DEM model of the open pit, the improved Saint-Venant equation is used to obtain the corresponding water fall energy loss.
3. The method for early warning of rainwater accumulation in open-pit mines based on image recognition according to claim 2, characterized in that: Based on the DEM model of the open pit, the modified Saint-Venant equation is used to define the energy loss of water flow, which includes the following steps: Use the curvature analysis algorithm to locate the edge of the steps in the terraced terrain of the mine in the DEM model and obtain the corresponding vertical drop of the steps; According to the vertical drop of the steps, the empirical loss coefficient and the instantaneous momentum loss term are defined, and an improved momentum equation is constructed based on the empirical loss coefficient and the instantaneous momentum loss term; The improved momentum equation is used to characterize the energy loss of water falling in an open pit.
4. The method for early warning of rainwater accumulation in open-pit mines based on image recognition according to claim 3, characterized in that: The image recognition model is constructed based on an improved MineSegNet algorithm, and the main network architecture of the image recognition model includes an encoder constructed based on the ResNet algorithm, a decoder constructed based on the Deconvolution-U-Net algorithm, and a classifier constructed based on the Sigmoid activation function, which are connected in sequence. The decoder is connected to an attention module constructed based on the SE Block-CBAM algorithm, and the input end of the attention module is further connected to a terrain feature extraction module constructed based on the CNN algorithm and a ripple feature extraction module constructed based on the FT-Daubechies algorithm. The image recognition model is set with a loss function based on the Saint-Venant equation constraints.
5. The method for early warning of rainwater accumulation in open-pit mines based on image recognition according to claim 4, characterized in that: Collecting image data and meteorological data from an open pit mine, and using an image recognition model to perform image recognition on the image data to obtain image recognition results, includes the following steps: Using drone swarms equipped with high-definition cameras to collect image data from the open-pit mine, and collecting meteorological data from the open-pit mine from external IoT rain gauges in the mining area; Preprocessing the image data and the meteorological data to obtain preprocessed image data and preprocessed meteorological data; Use the decoder of the image recognition model to extract multi-level image features of the preprocessed image data; Use the terrain feature extraction module of the image recognition model to extract the terrain features of the DEM model; Use the ripple feature extraction module of the image recognition model to extract the ripple features of the preprocessed image data; According to the dynamic attention weight, the attention module of the image recognition model is used to perform weighted fusion on multi-level image features, terrain features, and ripple features to obtain weighted fusion features; Based on the weighted fusion features, the decoder of the image recognition model is used to generate a segmentation mask to obtain a segmentation mask of the same size as the preprocessed image data; Based on the segmentation mask and pre-processed image data, the classifier of the image recognition model is used to perform classification, and image recognition results including waterlogging areas and runoff areas are obtained; The pixel coordinates of the waterlogged area in the image recognition result are mapped to the grid coordinates of the calculation grid corresponding to the DEM model.
6. The method for early warning of rainwater accumulation in open-pit mines based on image recognition according to claim 5, characterized in that: Based on the open pit meteorological data, DEM model, water flow energy loss, and image recognition results, a hydrodynamic analysis model was constructed, including the following steps: Combining the continuity equation with the improved momentum equation for water drop energy loss, a two-dimensional shallow water equation is obtained. The computational grid of the DEM model is spatially discretized to obtain the initial hydrodynamic analysis model. According to the height difference information of the DEM model and the coordinates of the water-logged pixels in the water-logged area in the image recognition results, the initial water depth field of the calculation grid of the initial hydrodynamic analysis model is set; According to the initial water depth field and the runoff area in the image recognition results, the initial flow velocity field of the computational grid of the initial hydrodynamic analysis model is set; According to the meteorological data and image recognition results of the open pit, the boundary conditions of the initial hydrodynamic analysis model are set to obtain the final hydrodynamic analysis model.
7. The method for early warning of rainwater accumulation in open-pit mines based on image recognition according to claim 6, characterized in that: The hydrodynamic analysis model is started to simulate the rainwater accumulation in the open pit at a future time point to obtain the rainwater accumulation prediction results, including the following steps: Starting a hydrodynamic analysis model to continuously collect simulation results of rainwater accumulation in the open pit at future time points; the simulation results of rainwater accumulation include water depth field prediction results, flow velocity field prediction results, and scour risk heat map for each calculation grid in the hydrodynamic analysis model; The simulation results of rainwater accumulation are post-processed to obtain rainwater accumulation prediction results; the rainwater accumulation prediction results include the prediction results of the accumulation area, the prediction results of the change of the accumulation depth, range and position of the accumulation area corresponding to the accumulation area prediction results over time, the prediction results of the water flow path in the mine pit and the prediction results of the change of the slope infiltration line.
8. The method for early warning of rainwater accumulation in open-pit mines based on image recognition according to claim 7, characterized in that: The intelligent risk warning model is constructed based on the MAPGRPO algorithm, and the intelligent risk warning model includes a central-level intelligent risk warning generation layer, a coordination layer, and a system-level intelligent risk warning generation layer connected in sequence. The central-level intelligent risk warning generation layer is provided with a first multi-optimization target set and a central-level intelligent agent, and the system-level intelligent risk warning generation layer is provided with a second multi-optimization target set and several parallel system-level intelligent agents.
9. The method for early warning of rainwater accumulation in open-pit mines based on image recognition according to claim 8, characterized in that: Based on the rainwater accumulation prediction results, the intelligent risk warning model is used to generate intelligent risk warnings and execute the obtained intelligent risk warning strategy, including the following steps: Based on the rainwater accumulation prediction results, the central-level intelligent agent of the central-level intelligent risk warning strategy generation layer of the intelligent risk warning model is used to generate an intelligent risk warning strategy to obtain a central-level intelligent risk warning strategy; Through the coordination layer of the intelligent risk warning model, the rainwater accumulation prediction results and the central-level intelligent risk warning strategy are input into several parallel system-level intelligent agents in the system-level intelligent risk warning strategy generation layer; Based on the rainwater accumulation prediction results and the center-level intelligent risk warning strategy, several system-level intelligent agents in the system-level intelligent risk warning strategy generation layer are used to generate intelligent risk warning strategies, and several system-level intelligent risk warning strategies are obtained; Integrate the center-level intelligent risk warning strategy and several system-level intelligent risk warning strategies to obtain the intelligent risk warning strategy, publish the intelligent risk warning strategy to all execution systems, and execute the intelligent risk warning strategy based on the execution system.
10. An open pit mine intelligent rainwater accumulation early warning system based on image recognition, used to implement the open pit mine intelligent rainwater accumulation early warning method according to any one of claims 1 to 9, characterized in that: The system comprises a DEM model building unit, an image recognition unit, a hydrodynamic analysis model building unit, a rainwater accumulation simulation unit and an intelligent risk warning unit which are connected in sequence.
Citation Information
Cited By
Dental caries identification method and system based on deep learning
CN121505399A