Method for monitoring safety of water accumulation area of tailings pond based on deep learning

Abstract

This invention discloses a method for monitoring the safety of tailings dam waterlogged areas based on deep learning, comprising: Step 1, acquiring panoramic images of the waterlogged area and close-up images of the drainage outlet, preprocessing the images, and extracting basic features; Step 2, calculating the initial image texture entropy, and calculating the actual water level change based on the basic features; Step 3, calculating the real-time turbidity based on the basic features and the initial image texture entropy; Step 4, calculating the water flow state quantification value, and calculating the siltation coefficient based on the water flow state quantification value to determine the degree of siltation; Step 5, calculating the comprehensive hazard index based on the basic features, the actual water level change, the real-time turbidity, and the siltation coefficient, determining the risk level, and identifying the hazard type. This invention solves the problems of traditional machine vision monitoring algorithms having poor adaptability to dynamic interference in waterlogged areas, water quality assessment not being correlated with particle settling characteristics, water flow analysis not being coupled with hydrostatic resistance, and hazard judgment relying on a single indicator, which is prone to misjudgment.

Description

Technical Field

[0001] This invention belongs to the field of closed reservoir drainage monitoring technology, specifically involving a method for monitoring the safety of tailings dam water accumulation areas based on deep learning. Background Technology

[0002] As a core supporting facility for mining production, tailings dams require drainage engineering before closure, which is a crucial prerequisite for safe closure. my country currently has nearly 12,000 tailings dams. The water accumulation areas of many wet-discharge tailings dams have long received extremely fine-grained tailings mud (particle size <0.075mm, content exceeding 60%), generally facing complex conditions such as high turbidity, stratified sedimentation, silt disturbance, and dynamic water flow. Furthermore, the poor self-compacting properties and high porosity of tailings mud pose a severe challenge to drainage safety monitoring.

[0003] During reservoir closure and drainage, it is crucial to accurately control the rate of water level decline, water turbidity, water flow status, and potential siltation / leakage hazards; otherwise, it can easily lead to safety accidents such as beach instability and environmental pollution. However, current monitoring technologies have significant shortcomings: traditional contact sensors are easily clogged and corroded by silt, resulting in data distortion and high maintenance costs, making it difficult to achieve full coverage; manual inspections have high safety risks and limited coverage, failing to meet the needs of continuous monitoring around the clock, and the identification of hidden hazards relies on experience, making misjudgments easy; existing general-purpose machine vision algorithms are not adapted to special working conditions, requiring reliance on external models, which are complex to operate and have long deployment cycles; at the same time, many hazards are judged based on a single indicator, without establishing multi-dimensional parameter coupling correlations, resulting in insufficient accuracy in hazard inference. Summary of the Invention

[0004] The purpose of this invention is to provide a method for monitoring the safety of tailings dam water accumulation areas based on deep learning. This method solves the problems of traditional machine vision monitoring algorithms, such as poor adaptability to dynamic interference in water accumulation areas, lack of correlation between water quality assessment and particle settling characteristics, lack of coupling of water flow analysis with hydrostatic resistance, and reliance on a single indicator for hazard identification, which can easily lead to misjudgment.

[0005] The technical solution adopted in this invention is a method for monitoring the safety of tailings dam water accumulation areas based on deep learning, comprising the following steps: Step 1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets, preprocess the images, and extract basic features; Step 2: Calculate the initial image texture entropy, and based on the basic features, calculate the actual water level change of the silt-texture coupling. Step 3: Calculate the real-time turbidity based on the basic features and the initial image texture entropy; Step 4: Calculate the quantitative value of water flow state based on the close-up image of the drain outlet, and calculate the siltation coefficient based on the quantitative value of water flow state to determine the degree of siltation. Step 5: Calculate the comprehensive hazard index based on basic characteristics, actual water level changes, real-time turbidity, and siltation coefficient. Then, determine the risk level and identify the type of hazard based on the comprehensive hazard index.

[0006] The invention is further characterized by: Step 1 is as follows: Step 1.1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets; Step 1.2: Suppress image noise using bilateral filtering algorithm, preserve tailings mud particle texture and water flow characteristics, use Retinex enhancement algorithm to eliminate water surface reflection interference in image, improve contrast between water area and background, separate water area in image using OTSU threshold segmentation algorithm, generate water mask and contour, and obtain preprocessed image; Step 1.3: Based on the preprocessed image, extract the core basic features of the water body area in the image; The core basic features include average grayscale value, real-time texture entropy, and dark pixel ratio.

[0007] Step 2 is as follows: Step 2.1: Extract stable feature points from the edge of the drainage outlet and the fixed marker post, perform feature point matching on consecutive frame images, and obtain the average pixel offset of the feature points in the vertical direction. Step 2.2: Calculate the initial image texture entropy, determine the perturbation coefficient through field test fitting, and calculate the silt perturbation rate according to equation (1); (1); in, The silt disturbance rate; Real-time texture entropy; c is the initial image texture entropy; c is the perturbation coefficient; Step 2.3: Obtain pixel equivalent through on-site calibration, measure initial turbidity, real-time turbidity, and monitoring interval time, and calculate the actual water level change of silt-texture coupling according to formula (2); (2); in, represents the actual water level change in the silt-texture coupling, with a positive value indicating a decrease; k is the pixel equivalent. Initial turbidity; Real-time turbidity; For monitoring interval time; This represents the average pixel offset of the feature point in the vertical direction.

[0008] The specific process for calculating the initial image texture entropy in step 2.2 is as follows: Step 2.2.1: Acquire several frames of images of the relatively still water accumulation area with silt; Step 2.2.2: Based on the image of a relatively static water accumulation area with silt in a single frame, calculate the entropy value of the gray-level co-occurrence matrix at each angle, and take the average value as the texture entropy of the single frame image; Step 2.2.3: Calculate the texture entropy of several frames of images, take the average value, and obtain the initial image texture entropy.

[0009] Step 3 specifically involves: Step 3.1: Calculate the change in texture entropy based on the real-time texture entropy and the initial image texture entropy according to equation (3); (3); in, Texture entropy change; Real-time texture entropy; The initial image texture entropy; Step 3.2: Obtain the average gray value of the clear water area. The first fitting coefficient, the second fitting coefficient, and the third fitting coefficient are obtained by calibration using on-site measured data. Based on the average gray value and the proportion of dark pixels, the real-time turbidity is calculated according to formula (4). (4); in, Real-time turbidity; This represents the average grayscale value. This represents the average grayscale value of the clear water area. Texture entropy change; For monitoring interval time; denoted as the percentage of dark pixels; a represents the first fitting coefficient, b represents the second fitting coefficient, and d represents the third fitting coefficient. Step 3.3: Based on the real-time turbidity, obtain the turbidity classification of the water accumulation area during the tailings dam closure and drainage period.

[0010] The turbidity classification in step 3.3 is as follows: ≤100 NTU indicates low turbidity; <100 NTU indicates low turbidity. A value ≤300 indicates moderate turbidity. A value greater than 300 NTU indicates high turbidity.

[0011] Step 4 is as follows: Step 4.1: Use the Lucas-Kanade optical flow method to calculate the average optical flow velocity and the standard deviation of the optical flow velocity on the close-up image of the drain outlet; Step 4.2: Obtain the optical flow velocity corresponding to the camera frame rate and standard drainage flow rate, calculate the water flow state quantization value according to formula (5), and determine the water flow state; (5); in, This is a quantification value for the water flow state; The average optical flow velocity is k; k is the pixel equivalent. For camera frame rate; The optical flow velocity corresponding to the standard drainage flow rate; Real-time turbidity; Step 4.3: Obtain the initial water flow state quantification value and calculate the siltation coefficient according to formula (6); (6); in, This is the siltation coefficient; This is a quantification value for the water flow state; This is the quantized value of the initial water flow state; Step 4.4: Determine the degree of blockage based on the blockage coefficient.

[0012] In step 4.2, the water flow state is determined as follows: When <0.4, it is considered normal flow; 0.4≤ When the flow rate is ≤0.8, it is considered a slow flow. When the value is greater than 0.8, it is considered a turbulent flow.

[0013] Step 4.4 specifically involves: A value greater than 0.6 indicates severe clogging; a value less than or equal to 0.3 indicates severe clogging. When the concentration is ≤0.6, it is considered moderate sludge. When the value is less than 0.3, it indicates mild stagnation.

[0014] Step 5 specifically involves: Step 5.1: Obtain the allowable water level change rate, allowable turbidity, allowable optical flow standard deviation, first weighting coefficient, second weighting coefficient, third weighting coefficient, and fourth weighting coefficient. Calculate the comprehensive index of hidden dangers according to formula (7) based on the actual water level change, real-time turbidity, siltation coefficient, and optical flow velocity standard deviation. (7); in, The comprehensive index of potential hazards; This represents the actual water level change in the silt-texture coupling. To allow for the rate of change of water level; Real-time turbidity; To allow for turbidity; This is the siltation coefficient; The standard deviation of optical flow velocity; To allow for the standard deviation of optical flow; w1 is the first weighting coefficient; w2 is the second weighting coefficient; w3 is the third weighting coefficient; w4 is the fourth weighting coefficient; Step 5.2: Determine the risk level based on the comprehensive hazard index; The risk level is: A value greater than 1.6 indicates a high-risk status; a value less than or equal to 1.1 indicates a low-risk status. A value ≤1.6 indicates a medium-risk status. When the value is less than 1.1, it is considered a low-risk condition. Step 5.3: Identify the types of hazards based on the risk level; Specifically, under high-risk conditions, if >0.6, judged as a potential clogging risk; if > and > If it is determined to be a potential leakage hazard; under medium-risk conditions, if If the value is greater than 0.5, it is considered a potential hazard of abnormal particle settling, and corresponding handling suggestions are provided. Recommendations include cleaning the drainage outlet when there is blockage; checking the slope when there is leakage; and increasing the monitoring frequency when there is abnormal particle settlement.

[0015] The beneficial effects of this invention are: 1) Strong dynamic adaptability and high monitoring accuracy: The formula incorporates the specific dynamic characteristics of waterlogged areas, such as silt disturbance, particle settling, and hydrostatic resistance, effectively correcting the environmental interference errors of traditional algorithms; 2) Clear quantitative logic and strong practicality: All key parameters (water level, turbidity, water flow state, siltation coefficient) are quantified and calculated through formulas to avoid subjective judgment bias, and the output results are directly related to the type of hidden danger and treatment suggestions, providing clear guidance for on-site operators; 3) Easy to operate and adaptable to on-site needs: Only an industrial camera is needed to collect images. No additional sensors or monitoring equipment are required. The algorithm is simple and efficient, adaptable to the complex on-site operating environment and rapid management needs during the tailings dam closure and drainage period, and has broad application prospects. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method for monitoring the safety of water accumulation areas in tailings ponds based on deep learning, according to the present invention. Detailed Implementation

[0017] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0018] The method for monitoring the safety of tailings dam water accumulation areas based on deep learning provided by this invention includes the following steps: Step 1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets, preprocess the images, and extract basic features; Specifically: Step 1.1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets; Step 1.2: Suppress image noise using bilateral filtering algorithm, preserve tailings mud particle texture and water flow characteristics, use Retinex enhancement algorithm to eliminate water surface reflection interference in image, improve contrast between water area and background, separate water area in image using OTSU threshold segmentation algorithm, generate water mask and contour, and obtain preprocessed image; Step 1.3: Based on the preprocessed image, extract the core basic features of the water body area in the image; Core basic features include average grayscale value, real-time texture entropy, and dark pixel ratio; Step 2: Calculate the initial image texture entropy, and based on the basic features, calculate the actual water level change of the silt-texture coupling. Specifically: Step 2.1: Extract stable feature points from the edge of the drainage outlet and the fixed marker post, perform feature point matching on consecutive frame images, and obtain the average pixel offset of the feature points in the vertical direction. Step 2.2: Calculate the initial image texture entropy, determine the perturbation coefficient through field test fitting, and calculate the silt perturbation rate according to equation (1); (1); in, The silt disturbance rate; Real-time texture entropy; c is the initial image texture entropy; c is the perturbation coefficient; The specific process for calculating the initial image texture entropy is as follows: Step 2.2.1: Acquire several frames of images of the relatively still water accumulation area with silt; Step 2.2.2: Based on the image of a relatively static water accumulation area with silt in a single frame, calculate the entropy value of the gray-level co-occurrence matrix at each angle, and take the average value as the texture entropy of the single frame image; Step 2.2.3: Calculate the texture entropy of several frames of images, take the average value, and obtain the initial image texture entropy; Step 2.3: Obtain pixel equivalent through on-site calibration, measure initial turbidity, real-time turbidity, and monitoring interval time, and calculate the actual water level change of silt-texture coupling according to formula (2); (2); in, represents the actual water level change in the silt-texture coupling, with a positive value indicating a decrease; k is the pixel equivalent. Initial turbidity; Real-time turbidity; For monitoring interval time; This represents the average pixel offset of the feature point in the vertical direction. Step 3: Calculate the real-time turbidity based on the basic features and the initial image texture entropy; Specifically: Step 3.1: Calculate the change in texture entropy based on the real-time texture entropy and the initial image texture entropy according to equation (3); (3); in, Texture entropy change; Real-time texture entropy; The initial image texture entropy; Step 3.2: Obtain the average gray value of the clear water area. The first fitting coefficient, the second fitting coefficient, and the third fitting coefficient are obtained by calibration using on-site measured data. Based on the average gray value and the proportion of dark pixels, the real-time turbidity is calculated according to formula (4). (4); in, Real-time turbidity; This represents the average grayscale value. This represents the average grayscale value of the clear water area. Texture entropy change; For monitoring interval time; denoted as the percentage of dark pixels; a represents the first fitting coefficient, b represents the second fitting coefficient, and d represents the third fitting coefficient. Step 3.3: Based on the real-time turbidity, obtain the turbidity classification of the water accumulation area during the tailings dam closure and drainage period; Turbidity is classified as follows: ≤100 NTU indicates low turbidity; <100 NTU indicates low turbidity. A value ≤300 indicates moderate turbidity. A value >300 NTU indicates high turbidity; Step 4: Calculate the quantitative value of water flow state based on the close-up image of the drain outlet, and calculate the siltation coefficient based on the quantitative value of water flow state to determine the degree of siltation. Specifically: Step 4.1: Use the Lucas-Kanade optical flow method to calculate the average optical flow velocity and the standard deviation of the optical flow velocity on the close-up image of the drain outlet; Step 4.2: Obtain the optical flow velocity corresponding to the camera frame rate and standard drainage flow rate, calculate the water flow state quantization value according to formula (5), and determine the water flow state; (5); in, This is a quantification value for the water flow state; The average optical flow velocity is k; k is the pixel equivalent. For camera frame rate; The optical flow velocity corresponding to the standard drainage flow rate; Real-time turbidity; The water flow state is determined as follows: When <0.4, it is considered normal flow; 0.4≤ When the flow rate is ≤0.8, it is considered a slow flow. When the value is greater than 0.8, it is considered a turbulent flow; Step 4.3: Obtain the initial water flow state quantification value and calculate the siltation coefficient according to formula (6); (6); in, This is the siltation coefficient; This is a quantification value for the water flow state; This is the quantized value of the initial water flow state; Step 4.4: Determine the degree of blockage based on the blockage coefficient; Specifically: A value greater than 0.6 indicates severe clogging; a value less than or equal to 0.3 indicates severe clogging. When the concentration is ≤0.6, it is considered moderate sludge. When the concentration is less than 0.3, it indicates mild sludge buildup. Step 5: Calculate the comprehensive hazard index based on basic characteristics, actual water level changes, real-time turbidity, and siltation coefficient; and determine the risk level and identify the hazard type based on the comprehensive hazard index. Specifically: Step 5.1: Obtain the allowable water level change rate, allowable turbidity, allowable optical flow standard deviation, first weighting coefficient, second weighting coefficient, third weighting coefficient, and fourth weighting coefficient. Calculate the comprehensive index of hidden dangers according to formula (7) based on the actual water level change, real-time turbidity, siltation coefficient, and optical flow velocity standard deviation. (7); in, The comprehensive index of potential hazards; This represents the actual water level change in the silt-texture coupling. To allow for the rate of change of water level; Real-time turbidity; To allow for turbidity; This is the siltation coefficient; The standard deviation of optical flow velocity; To allow for the standard deviation of optical flow; w1 is the first weighting coefficient; w2 is the second weighting coefficient; w3 is the third weighting coefficient; w4 is the fourth weighting coefficient; Step 5.2: Determine the risk level based on the comprehensive hazard index; The risk level is: A value greater than 1.6 indicates a high-risk status; a value less than or equal to 1.1 indicates a low-risk status. A value ≤1.6 indicates a medium-risk status. When the value is less than 1.1, it is considered a low-risk condition. Step 5.3: Identify the types of hazards based on the risk level; Specifically, under high-risk conditions, if >0.6, judged as a potential clogging risk; if > and > If it is determined to be a potential leakage hazard; under medium-risk conditions, if If the value is greater than 0.5, it is considered a potential hazard of abnormal particle settling, and corresponding handling suggestions are provided. Recommendations include cleaning the drainage outlet when there is blockage; checking the slope when there is leakage; and increasing the monitoring frequency when there is abnormal particle settlement.

[0019] Example 1 The method for monitoring the safety of tailings dam water accumulation areas based on deep learning proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets, preprocess the images, and extract basic features; Step 2: Calculate the initial image texture entropy, and based on the basic features, calculate the actual water level change of the silt-texture coupling. Step 3: Calculate the real-time turbidity based on the basic features and the initial image texture entropy; Step 4: Calculate the quantitative value of water flow state based on the close-up image of the drain outlet, and calculate the siltation coefficient based on the quantitative value of water flow state to determine the degree of siltation. Step 5: Calculate the comprehensive hazard index based on basic characteristics, actual water level changes, real-time turbidity, and siltation coefficient. Then, determine the risk level and identify the type of hazard based on the comprehensive hazard index.

[0020] Example 2 The method for monitoring the safety of tailings dam water accumulation areas based on deep learning proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets, preprocess the images, and extract basic features; Specifically: Step 1.1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets; Step 1.2: Suppress image noise using bilateral filtering algorithm, preserve tailings mud particle texture and water flow characteristics, use Retinex enhancement algorithm to eliminate water surface reflection interference in image, improve contrast between water area and background, separate water area in image using OTSU threshold segmentation algorithm, generate water mask and contour, and obtain preprocessed image; Step 1.3: Based on the preprocessed image, extract the core basic features of the water body area in the image; Core basic features include average grayscale value, real-time texture entropy, and dark pixel ratio; Step 2: Calculate the initial image texture entropy, and based on the basic features, calculate the actual water level change of the silt-texture coupling. Step 3: Calculate the real-time turbidity based on the basic features and the initial image texture entropy; Step 4: Calculate the quantitative value of water flow state based on the close-up image of the drain outlet, and calculate the siltation coefficient based on the quantitative value of water flow state to determine the degree of siltation. Step 5: Calculate the comprehensive hazard index based on basic characteristics, actual water level changes, real-time turbidity, and siltation coefficient. Then, determine the risk level and identify the type of hazard based on the comprehensive hazard index.

[0021] Example 3 The method for monitoring the safety of tailings dam water accumulation areas based on deep learning proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets, preprocess the images, and extract basic features; Specifically: Step 1.1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets; Step 1.2: Suppress image noise using bilateral filtering algorithm, preserve tailings mud particle texture and water flow characteristics, use Retinex enhancement algorithm to eliminate water surface reflection interference in image, improve contrast between water area and background, separate water area in image using OTSU threshold segmentation algorithm, generate water mask and contour, and obtain preprocessed image; Step 1.3: Based on the preprocessed image, extract the core basic features of the water body area in the image; Core basic features include average grayscale value, real-time texture entropy, and dark pixel ratio; Step 2: Calculate the initial image texture entropy, and based on the basic features, calculate the actual water level change of the silt-texture coupling. Specifically: Step 2.1: Extract stable feature points from the edge of the drainage outlet and the fixed marker post, perform feature point matching on consecutive frame images, and obtain the average pixel offset of the feature points in the vertical direction. Step 2.2: Calculate the initial image texture entropy, determine the perturbation coefficient through field test fitting, and calculate the silt perturbation rate according to equation (1); (1); in, The silt disturbance rate; Real-time texture entropy; c is the initial image texture entropy; c is the perturbation coefficient; The specific process for calculating the initial image texture entropy is as follows: Step 2.2.1: Acquire several frames of images of the relatively still water accumulation area with silt; Step 2.2.2: Based on the image of a relatively static water accumulation area with silt in a single frame, calculate the entropy value of the gray-level co-occurrence matrix at each angle, and take the average value as the texture entropy of the single frame image; Step 2.2.3: Calculate the texture entropy of several frames of images, take the average value, and obtain the initial image texture entropy; Step 2.3: Obtain pixel equivalent through on-site calibration, measure initial turbidity, real-time turbidity, and monitoring interval time, and calculate the actual water level change of silt-texture coupling according to formula (2); (2); in, represents the actual water level change in the silt-texture coupling, with a positive value indicating a decrease; k is the pixel equivalent. Initial turbidity; Real-time turbidity; For monitoring interval time; This represents the average pixel offset of the feature point in the vertical direction. Step 3: Calculate the real-time turbidity based on the basic features and the initial image texture entropy; Step 4: Calculate the quantitative value of water flow state based on the close-up image of the drain outlet, and calculate the siltation coefficient based on the quantitative value of water flow state to determine the degree of siltation. Step 5: Calculate the comprehensive hazard index based on basic characteristics, actual water level changes, real-time turbidity, and siltation coefficient. Then, determine the risk level and identify the type of hazard based on the comprehensive hazard index.

[0022] Example 4 The method for monitoring the safety of tailings dam water accumulation areas based on deep learning proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets, preprocess the images, and extract basic features; Specifically: Step 1.1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets; Step 1.2: Suppress image noise using bilateral filtering algorithm, preserve tailings mud particle texture and water flow characteristics, use Retinex enhancement algorithm to eliminate water surface reflection interference in image, improve contrast between water area and background, separate water area in image using OTSU threshold segmentation algorithm, generate water mask and contour, and obtain preprocessed image; Step 1.3: Based on the preprocessed image, extract the core basic features of the water body area in the image; Core basic features include average grayscale value, real-time texture entropy, and dark pixel ratio; Step 2: Calculate the initial image texture entropy, and based on the basic features, calculate the actual water level change of the silt-texture coupling. Specifically: Step 2.1: Extract stable feature points from the edge of the drainage outlet and the fixed marker post, perform feature point matching on consecutive frame images, and obtain the average pixel offset of the feature points in the vertical direction. Step 2.2: Calculate the initial image texture entropy, determine the perturbation coefficient through field test fitting, and calculate the silt perturbation rate according to equation (1); (1); in, The silt disturbance rate; Real-time texture entropy; c is the initial image texture entropy; c is the perturbation coefficient; The specific process for calculating the initial image texture entropy is as follows: Step 2.2.1: Acquire several frames of images of the relatively still water accumulation area with silt; Step 2.2.2: Based on the image of a relatively static water accumulation area with silt in a single frame, calculate the entropy value of the gray-level co-occurrence matrix at each angle, and take the average value as the texture entropy of the single frame image; Step 2.2.3: Calculate the texture entropy of several frames of images, take the average value, and obtain the initial image texture entropy; Step 2.3: Obtain pixel equivalent through on-site calibration, measure initial turbidity, real-time turbidity, and monitoring interval time, and calculate the actual water level change of silt-texture coupling according to formula (2); (2); in, represents the actual water level change in the silt-texture coupling, with a positive value indicating a decrease; k is the pixel equivalent. Initial turbidity; Real-time turbidity; For monitoring interval time; This represents the average pixel offset of the feature point in the vertical direction. Step 3: Calculate the real-time turbidity based on the basic features and the initial image texture entropy; Specifically: Step 3.1: Calculate the change in texture entropy based on the real-time texture entropy and the initial image texture entropy according to equation (3); (3); in, Texture entropy change; Real-time texture entropy; The initial image texture entropy; Step 3.2: Obtain the average gray value of the clear water area. The first fitting coefficient, the second fitting coefficient, and the third fitting coefficient are obtained by calibration using on-site measured data. Based on the average gray value and the proportion of dark pixels, the real-time turbidity is calculated according to formula (4). (4); in, Real-time turbidity; This represents the average grayscale value. This represents the average grayscale value of the clear water area. Texture entropy change; For monitoring interval time; denoted as the percentage of dark pixels; a represents the first fitting coefficient, b represents the second fitting coefficient, and d represents the third fitting coefficient. Step 3.3: Based on the real-time turbidity, obtain the turbidity classification of the water accumulation area during the tailings dam closure and drainage period; Turbidity is classified as follows: ≤100 NTU indicates low turbidity; <100 NTU indicates low turbidity. A value ≤300 indicates moderate turbidity. A value >300 NTU indicates high turbidity; Step 4: Calculate the quantitative value of water flow state based on the close-up image of the drain outlet, and calculate the siltation coefficient based on the quantitative value of water flow state to determine the degree of siltation. Step 5: Calculate the comprehensive hazard index based on basic characteristics, actual water level changes, real-time turbidity, and siltation coefficient. Then, determine the risk level and identify the type of hazard based on the comprehensive hazard index.

[0023] Example 5 The method for monitoring the safety of tailings dam water accumulation areas based on deep learning proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets, preprocess the images, and extract basic features; Specifically: Step 1.1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets; Step 1.2: Suppress image noise using bilateral filtering algorithm, preserve tailings mud particle texture and water flow characteristics, use Retinex enhancement algorithm to eliminate water surface reflection interference in image, improve contrast between water area and background, separate water area in image using OTSU threshold segmentation algorithm, generate water mask and contour, and obtain preprocessed image; Step 1.3: Based on the preprocessed image, extract the core basic features of the water body area in the image; Core basic features include average grayscale value, real-time texture entropy, and dark pixel ratio; Step 2: Calculate the initial image texture entropy, and based on the basic features, calculate the actual water level change of the silt-texture coupling. Specifically: Step 2.1: Extract stable feature points from the edge of the drainage outlet and the fixed marker post, perform feature point matching on consecutive frame images, and obtain the average pixel offset of the feature points in the vertical direction. Step 2.2: Calculate the initial image texture entropy, determine the perturbation coefficient through field test fitting, and calculate the silt perturbation rate according to equation (1); (1); in, The silt disturbance rate; Real-time texture entropy; c is the initial image texture entropy; c is the perturbation coefficient; The specific process for calculating the initial image texture entropy is as follows: Step 2.2.1: Acquire several frames of images of the relatively still water accumulation area with silt; Step 2.2.2: Based on the image of a relatively static water accumulation area with silt in a single frame, calculate the entropy value of the gray-level co-occurrence matrix at each angle, and take the average value as the texture entropy of the single frame image; Step 2.2.3: Calculate the texture entropy of several frames of images, take the average value, and obtain the initial image texture entropy; Step 2.3: Obtain pixel equivalent through on-site calibration, measure initial turbidity, real-time turbidity, and monitoring interval time, and calculate the actual water level change of silt-texture coupling according to formula (2); (2); in, represents the actual water level change in the silt-texture coupling, with a positive value indicating a decrease; k is the pixel equivalent. Initial turbidity; Real-time turbidity; For monitoring interval time; This represents the average pixel offset of the feature point in the vertical direction. Step 3: Calculate the real-time turbidity based on the basic features and the initial image texture entropy; Specifically: Step 3.1: Calculate the change in texture entropy based on the real-time texture entropy and the initial image texture entropy according to equation (3); (3); in, Texture entropy change; Real-time texture entropy; The initial image texture entropy; Step 3.2: Obtain the average gray value of the clear water area. The first fitting coefficient, the second fitting coefficient, and the third fitting coefficient are obtained by calibration using on-site measured data. Based on the average gray value and the proportion of dark pixels, the real-time turbidity is calculated according to formula (4). (4); in, Real-time turbidity; This represents the average grayscale value. This represents the average grayscale value of the clear water area. Texture entropy change; For monitoring interval time; denoted as the percentage of dark pixels; a represents the first fitting coefficient, b represents the second fitting coefficient, and d represents the third fitting coefficient. Step 3.3: Based on the real-time turbidity, obtain the turbidity classification of the water accumulation area during the tailings dam closure and drainage period; Turbidity is classified as follows: ≤100 NTU indicates low turbidity; <100 NTU indicates low turbidity. A value ≤300 indicates moderate turbidity. A value >300 NTU indicates high turbidity; Step 4: Calculate the quantitative value of water flow state based on the close-up image of the drain outlet, and calculate the siltation coefficient based on the quantitative value of water flow state to determine the degree of siltation. Specifically: Step 4.1: Use the Lucas-Kanade optical flow method to calculate the average optical flow velocity and the standard deviation of the optical flow velocity on the close-up image of the drain outlet; Step 4.2: Obtain the optical flow velocity corresponding to the camera frame rate and standard drainage flow rate, calculate the water flow state quantization value according to formula (5), and determine the water flow state; (5); in, This is a quantification value for the water flow state; The average optical flow velocity is k; k is the pixel equivalent. For camera frame rate; The optical flow velocity corresponding to the standard drainage flow rate; Real-time turbidity; The water flow state is determined as follows: When <0.4, it is considered normal flow; 0.4≤ When the flow rate is ≤0.8, it is considered a slow flow. When the value is greater than 0.8, it is considered a turbulent flow; Step 4.3: Obtain the initial water flow state quantification value and calculate the siltation coefficient according to formula (6); (6); in, This is the siltation coefficient; This is a quantification value for the water flow state; This is the quantized value of the initial water flow state; Step 4.4: Determine the degree of blockage based on the blockage coefficient; Specifically: A value greater than 0.6 indicates severe clogging; a value less than or equal to 0.3 indicates severe clogging. When the concentration is ≤0.6, it is considered moderate sludge. When the concentration is less than 0.3, it indicates mild sludge buildup. Step 5: Calculate the comprehensive hazard index based on basic characteristics, actual water level changes, real-time turbidity, and siltation coefficient. Then, determine the risk level and identify the type of hazard based on the comprehensive hazard index.

[0024] Example 6 The method for monitoring the safety of tailings dam water accumulation areas based on deep learning proposed in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets, preprocess the images, and extract basic features; Specifically: Step 1.1: Use a 1080P industrial camera (30fps) and fix it on the bank of the waterlogged area or the inspection robot to simultaneously collect panoramic images of the waterlogged area and close-up images of the drainage outlet to ensure that feature points and water flow details are clearly distinguishable. Step 1.2: Suppress image noise using a bilateral filtering algorithm, preserve tailings mud particle texture and water flow features, use the Retinex enhancement algorithm to eliminate water surface reflection interference in the image, improve the contrast between the water area and the background, separate the water area in the image using the OTSU threshold segmentation algorithm, generate water mask and contour, lay the foundation for subsequent feature extraction, and obtain the preprocessed image. Step 1.3: Based on the preprocessed image, extract the core basic features of the water body area in the image; Core basic features include average grayscale value, real-time texture entropy, and dark pixel ratio (the proportion of pixels with a grayscale value < 50). Step 2: Calculate the initial image texture entropy, and based on the basic features, calculate the actual water level change of the silt-texture coupling. Specifically: Step 2.1: Extract stable feature points from the edge of the drainage outlet and the fixed marker post, perform feature point matching on consecutive frame images, and obtain the average pixel offset (pixels) of the feature points in the vertical direction. Step 2.2: Calculate the initial image texture entropy (collected when the silt is still), determine the disturbance coefficient through field test fitting, and calculate the silt disturbance rate according to formula (1); (1); in, The silt disturbance rate; Real-time texture entropy; c is the initial image texture entropy; c is the perturbation coefficient; The specific process for calculating the initial image texture entropy is as follows: Step 2.2.1: Acquire 3-5 frames of images of the relatively still water accumulation area with silt; Step 2.2.2: Based on the image of a relatively static water accumulation area with silt in a single frame, calculate the entropy value of the gray-level co-occurrence matrix at each angle, and take the average value as the texture entropy of the single frame image; Step 2.2.3: Calculate the texture entropy of 3 to 5 frames of images, take the average value, and obtain the initial image texture entropy; Step 2.3: Obtain the pixel equivalent (cm / pixel, obtained through on-site calibration), measure the initial turbidity (measured in the initial stage of drainage), real-time turbidity (calculated by subsequent water quality assessment algorithm), and monitoring interval time, and calculate the actual water level change of silt-texture coupling according to formula (2); (2); in, represents the actual water level change in the silt-texture coupling, with a positive value indicating a decrease; k is the pixel equivalent. Initial turbidity; Real-time turbidity; For monitoring interval time; This represents the average pixel offset of the feature point in the vertical direction. Step 3: Calculate the real-time turbidity based on the basic features and the initial image texture entropy; Specifically: Step 3.1: Calculate the change in texture entropy based on the real-time texture entropy and the initial image texture entropy according to equation (3); (3); in, Texture entropy change; Real-time texture entropy; The initial image texture entropy; Step 3.2: Obtain the average gray value of the clear water area (calibrated to 220). The first fitting coefficient, the second fitting coefficient, and the third fitting coefficient were obtained by calibration using on-site measured data (calibrated by on-site measured data, least squares method). Based on the average gray value and the proportion of dark pixels, the real-time turbidity was calculated according to formula (4). (4); in, Real-time turbidity; This represents the average grayscale value. This represents the average grayscale value of the clear water area. Texture entropy change; For monitoring interval time; denoted as the percentage of dark pixels; a represents the first fitting coefficient, b represents the second fitting coefficient, and d represents the third fitting coefficient. Step 3.3: Based on the real-time turbidity, obtain the turbidity classification of the water accumulation area during the tailings dam closure and drainage period; Turbidity is classified as follows: ≤100 NTU indicates low turbidity; <100 NTU indicates low turbidity. A value ≤300 indicates moderate turbidity. When the turbidity is >300 NTU, it is considered high turbidity. The determination is based on the suspension characteristics of tailings mud particles in the water accumulation area, the environmental protection requirements for drainage treatment, the engineering safety control threshold, and the adaptability of machine vision monitoring algorithms. Step 4: Calculate the quantitative value of water flow state based on the close-up image of the drain outlet, and calculate the siltation coefficient based on the quantitative value of water flow state to determine the degree of siltation. Specifically: Step 4.1: Use the Lucas-Kanade optical flow method to calculate the average optical flow velocity (pixels / frame) and the standard deviation of the optical flow velocity (reflecting the degree of flow turbulence) on the close-up image of the drain outlet. Step 4.2: Obtain the camera frame rate (fps) and the optical flow velocity (pixels / frame, calibrated by more than 100 frames of images under rated flow conditions), calculate the water flow state quantization value (unitless) according to formula (5), and determine the water flow state; (5); in, This is a quantification value for the water flow state; The average optical flow velocity is k; k is the pixel equivalent. For camera frame rate; The optical flow velocity corresponding to the standard drainage flow rate; Real-time turbidity; The water flow state is determined as follows: When <0.4, it is considered normal flow; 0.4≤ When the flow rate is ≤0.8, it is considered a slow flow. When the value is greater than 0.8, it is considered a turbulent flow; Step 4.3: Obtain the initial water flow state quantification value (calculated in the initial stage of drainage), and calculate the siltation coefficient according to formula (6); (6); in, This is the siltation coefficient; This is a quantification value for the water flow state; This is the quantized value of the initial water flow state; Step 4.4: Determine the degree of blockage based on the blockage coefficient; Specifically: A value greater than 0.6 indicates severe clogging; a value less than or equal to 0.3 indicates severe clogging. When the concentration is ≤0.6, it is considered moderate sludge. When the concentration is less than 0.3, it indicates mild sludge buildup. Step 5: Calculate the comprehensive hazard index based on basic characteristics, actual water level changes, real-time turbidity, and siltation coefficient; and determine the risk level and identify the hazard type based on the comprehensive hazard index. Specifically: Step 5.1: Obtain the allowable water level change rate, allowable turbidity, allowable optical flow standard deviation, first weight coefficient, second weight coefficient, third weight coefficient, and fourth weight coefficient (calibrated according to the weight of the impact of the hidden danger). Calculate the comprehensive index of the hidden danger according to formula (7) based on the actual water level change, real-time turbidity, siltation coefficient, and optical flow velocity standard deviation. (7); in, The comprehensive index of potential hazards; This represents the actual water level change in the silt-texture coupling. To allow for the rate of change of water level; Real-time turbidity; To allow for turbidity; This is the siltation coefficient; The standard deviation of optical flow velocity; To allow for the standard deviation of optical flow; w1 is the first weighting coefficient; w2 is the second weighting coefficient; w3 is the third weighting coefficient; w4 is the fourth weighting coefficient; The weighting of potential hazards follows the principles of "prioritizing risk contribution, adhering to regulatory requirements, and adapting to engineering practices." It references the "Technical Specification for Risk Assessment of Natural Disasters in Metal and Non-metal Mines and Tailings Dams" and the "Technical Regulations for Classification and Grading of Environmental Supervision of Tailings Dams (Trial)." By quantifying the actual impact of each hazard indicator on the safety of the tailings dam (e.g., siltation directly leads to drainage failure, with the highest risk contribution; turbulent water flow is an indirect early warning signal, with a lower contribution), and combining this with the control priorities clearly defined in the regulations, the proportion of each coefficient is determined to ensure that the weighting is consistent with the "hazard-risk" correlation logic.

[0025] Step 5.2: Determine the risk level based on the comprehensive hazard index; The risk level is: A value greater than 1.6 indicates a high-risk status; a value less than or equal to 1.1 indicates a low-risk status. A value ≤1.6 indicates a medium-risk status. When the value is less than 1.1, it is considered a low-risk condition. Step 5.3: Identify the types of hazards based on the risk level; Specifically, under high-risk conditions, if >0.6, judged as a potential clogging risk; if > and > If it is determined to be a potential leakage hazard; under medium-risk conditions, if If the value is greater than 0.5, it is considered a potential hazard of abnormal particle settling, and corresponding handling suggestions are provided. Recommendations include cleaning the drainage outlet when there is blockage; checking the slope when there is leakage; and increasing the monitoring frequency when there is abnormal particle settlement.

Claims

1. A method for monitoring the safety of tailings dam water accumulation areas based on deep learning, characterized in that, Includes the following steps: Step 1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets, preprocess the images, and extract basic features; Step 2: Calculate the initial image texture entropy, and based on the basic features, calculate the actual water level change of the silt-texture coupling. Step 3: Calculate the real-time turbidity based on the basic features and the initial image texture entropy; Step 4: Calculate the quantitative value of water flow state based on the close-up image of the drain outlet, and calculate the siltation coefficient based on the quantitative value of water flow state to determine the degree of siltation. Step 5: Calculate the comprehensive hazard index based on basic characteristics, actual water level changes, real-time turbidity, and siltation coefficient. Then, determine the risk level and identify the type of hazard based on the comprehensive hazard index.

2. The method for monitoring the safety of tailings dam water accumulation areas based on deep learning according to claim 1, characterized in that, Step 1 specifically involves: Step 1.1: Collect panoramic images of the waterlogged area and close-up images of the drainage outlets; Step 1.2: Suppress image noise using bilateral filtering algorithm, preserve tailings mud particle texture and water flow characteristics, use Retinex enhancement algorithm to eliminate water surface reflection interference in image, improve contrast between water area and background, separate water area in image using OTSU threshold segmentation algorithm, generate water mask and contour, and obtain preprocessed image; Step 1.3: Based on the preprocessed image, extract the core basic features of the water body area in the image; The core basic features include average grayscale value, real-time texture entropy, and dark pixel ratio.

3. The method for monitoring the safety of tailings dam water accumulation areas based on deep learning according to claim 2, characterized in that, Step 2 specifically involves: Step 2.1: Extract stable feature points from the edge of the drainage outlet and the fixed marker post, perform feature point matching on consecutive frame images, and obtain the average pixel offset of the feature points in the vertical direction. Step 2.2: Calculate the initial image texture entropy, determine the perturbation coefficient through field test fitting, and calculate the silt perturbation rate according to equation (1); （1）； in, The silt disturbance rate; Real-time texture entropy; c is the initial image texture entropy; c is the perturbation coefficient; Step 2.3: Obtain pixel equivalent through on-site calibration, measure initial turbidity, real-time turbidity, and monitoring interval time, and calculate the actual water level change of silt-texture coupling according to formula (2); （2）； in, represents the actual water level change in the silt-texture coupling, with a positive value indicating a decrease; k is the pixel equivalent. Initial turbidity; Real-time turbidity; For monitoring interval time; This represents the average pixel offset of the feature point in the vertical direction.

4. The method for monitoring the safety of tailings dam water accumulation areas based on deep learning according to claim 3, characterized in that, The specific process for calculating the initial image texture entropy in step 2.2 is as follows: Step 2.2.1: Acquire several frames of images of the relatively still water accumulation area with silt; Step 2.2.2: Based on the image of a relatively static water accumulation area with silt in a single frame, calculate the entropy value of the gray-level co-occurrence matrix at each angle, and take the average value as the texture entropy of the single frame image; Step 2.2.3: Calculate the texture entropy of several frames of images, take the average value, and obtain the initial image texture entropy.

5. The method for monitoring the safety of tailings dam water accumulation areas based on deep learning according to claim 4, characterized in that, Step 3 specifically involves: Step 3.1: Calculate the change in texture entropy based on the real-time texture entropy and the initial image texture entropy according to equation (3); （3）； in, Texture entropy change; Real-time texture entropy; The initial image texture entropy; Step 3.2: Obtain the average gray value of the clear water area. The first fitting coefficient, the second fitting coefficient, and the third fitting coefficient are obtained by calibration using on-site measured data. Based on the average gray value and the proportion of dark pixels, the real-time turbidity is calculated according to formula (4). （4）； in, Real-time turbidity; This represents the average grayscale value. This represents the average grayscale value of the clear water area. Texture entropy change; For monitoring interval time; denoted as the percentage of dark pixels; a represents the first fitting coefficient, b represents the second fitting coefficient, and d represents the third fitting coefficient. Step 3.3: Based on the real-time turbidity, obtain the turbidity classification of the water accumulation area during the tailings dam closure and drainage period.

6. The method for monitoring the safety of tailings dam water accumulation areas based on deep learning according to claim 5, characterized in that, The turbidity classification described in step 3.3 is as follows: ≤100 NTU indicates low turbidity; <100 NTU indicates low turbidity. A value ≤300 indicates moderate turbidity. A value greater than 300 NTU indicates high turbidity.

7. The method for monitoring the safety of tailings dam water accumulation areas based on deep learning according to claim 6, characterized in that, Step 4 specifically involves: Step 4.1: Use the Lucas-Kanade optical flow method to calculate the average optical flow velocity and the standard deviation of the optical flow velocity on the close-up image of the drain outlet; Step 4.2: Obtain the optical flow velocity corresponding to the camera frame rate and standard drainage flow rate, calculate the water flow state quantization value according to formula (5), and determine the water flow state; （5）； in, This is a quantification value for the water flow state; The average optical flow velocity is k; k is the pixel equivalent. For camera frame rate; The optical flow velocity corresponding to the standard drainage flow rate; Real-time turbidity; Step 4.3: Obtain the initial water flow state quantification value and calculate the siltation coefficient according to formula (6); （6）； in, This is the siltation coefficient; This is a quantification value for the water flow state; This is the quantized value of the initial water flow state; Step 4.4: Determine the degree of blockage based on the blockage coefficient.

8. The method for monitoring the safety of tailings dam water accumulation areas based on deep learning according to claim 7, characterized in that, The determination of water flow state in step 4.2 is as follows: When <0.4, it is considered normal flow; 0.4≤ When the flow rate is ≤0.8, it is considered a slow flow. When the value is greater than 0.8, it is considered a turbulent flow.

9. The method for monitoring the safety of tailings dam water accumulation areas based on deep learning according to claim 8, characterized in that, Step 4.4 specifically involves: A value greater than 0.6 indicates severe clogging; 0.3≤ When the concentration is ≤0.6, it is considered moderate sludge. When the value is less than 0.3, it indicates mild stagnation.

10. The method for monitoring the safety of tailings dam water accumulation areas based on deep learning according to claim 9, characterized in that, Step 5 specifically involves: Step 5.1: Obtain the allowable water level change rate, allowable turbidity, allowable optical flow standard deviation, first weighting coefficient, second weighting coefficient, third weighting coefficient, and fourth weighting coefficient. Calculate the comprehensive index of hidden dangers according to formula (7) based on the actual water level change, real-time turbidity, siltation coefficient, and optical flow velocity standard deviation. （7）； in, The comprehensive index of potential hazards; This represents the actual water level change in the silt-texture coupling. To allow for the rate of change of water level; Real-time turbidity; To allow for turbidity; This is the siltation coefficient; The standard deviation of optical flow velocity; To allow for the standard deviation of optical flow; w1 is the first weighting coefficient; w2 is the second weighting coefficient; w3 is the third weighting coefficient; w4 is the fourth weighting coefficient; Step 5.2: Determine the risk level based on the comprehensive hazard index; The risk level is: A value greater than 1.6 indicates a high-risk status; a value less than or equal to 1.1 indicates a low-risk status. A value ≤1.6 indicates a medium-risk status. When the value is less than 1.1, it is considered a low-risk condition. Step 5.3: Identify the types of hazards based on the risk level; Specifically, under high-risk conditions, if >0.6, judged as a potential clogging risk; if > and > If it is determined to be a potential leakage hazard; under medium-risk conditions, if If the value is greater than 0.5, it is considered a potential hazard of abnormal particle settling, and corresponding handling suggestions are provided. The proposed solutions include cleaning the drainage outlet when there is blockage; checking the slope when there is leakage; and increasing the monitoring frequency when there is abnormal particle settlement.

Images