A mine safety risk early warning method based on air-ground integration

Abstract

The present invention discloses a mine safety risk early warning method based on air-space-ground integration, and relates to the field of risk early warning technology. The present invention comprises the following steps: respectively acquiring satellite remote sensing data, aerial monitoring data, and ground monitoring data; transmitting the acquired data information to a ground data center; performing format unification and denoising processing on the acquired data; extracting mine safety risk-related features from the pre-processed data; constructing a mine risk assessment model using machine learning; and issuing early warning information based on the output of the mine risk assessment model. The present invention respectively acquires satellite remote sensing data, aerial monitoring data, and ground monitoring data through air-space-ground technical means, and constructs a mine risk assessment model for the acquired data using machine learning, thereby achieving real-time safety early warning for mines from air-space-ground, improving the level of mine safety management, and reducing the probability of mine accidents.

Description

Technical Field

[0001] The present invention belongs to the technical field of risk warning, and in particular relates to a mine safety risk warning method based on air-ground-space integration. Background Art

[0002] With the continued development of the global mining industry, the scale of mining continues to expand, and the operating environment becomes increasingly complex. Mine safety issues have always been one of the key factors restricting the sustainable development of the industry. Traditional mine safety monitoring and early warning methods mainly rely on limited ground sensor networks and manual inspections, which have many limitations.

[0003] In terms of ground monitoring, although sensors can provide real-time monitoring of key parameters such as mine pressure and hazardous gas concentrations, their placement is limited by the geographical environment and the mining layout, making it difficult to achieve comprehensive coverage of the entire mining area. For example, in some mountainous mines with complex terrain or large open-pit mines, it is difficult to deploy sensors to all potentially dangerous areas, resulting in safety hazards in some areas not being discovered in a timely manner. Moreover, data from a single ground sensor can only reflect localized, specific types of safety information, lacking comprehensive awareness of safety risks arising from changes in the mine's overall geological structure, large-scale surface deformation, and interactions with the surrounding environment.

[0004] Manual inspections, a traditional auxiliary monitoring method, are inefficient and highly subjective. Factors such as the inspector's professionalism, work experience, and physical and mental state can significantly influence test results. Furthermore, manual inspections struggle to provide real-time, continuous monitoring, with significant time lags. Safety incidents can occur between inspections, leaving them undetected and unpredictable.

[0005] Satellite remote sensing technology has been increasingly used in mine monitoring in recent years. It can capture surface information across large areas of a mine, enabling macroscopic monitoring of the mine's overall topography, mining area, and waste dumping conditions. However, the resolution of satellite remote sensing data is relatively limited, making it difficult to accurately capture detailed information within mines, such as the operating status of small equipment and localized rock cracks. Furthermore, satellite data acquisition is affected by factors such as weather and the satellite's orbital period, and the data update frequency sometimes fails to meet the needs of real-time mine safety warnings.

[0006] The rise of aerial monitoring, particularly drone technology, has provided a new means for mine safety monitoring. Drones can fly at low altitudes and carry a variety of high-precision sensors, such as high-resolution cameras, thermal imagers, and lidar. These drones can obtain more detailed images and data of localized mine areas, partially addressing the shortcomings of satellite remote sensing and ground-based monitoring. However, drones have limited endurance, a small coverage area per flight, and limited flight operations in complex weather conditions (such as strong winds and heavy rain). This makes it difficult for them to independently undertake the long-term, stable safety monitoring of an entire mine.

[0007] In summary, existing mine safety monitoring methods all have their limitations and are unable to provide comprehensive, real-time, and accurate early warning of mine safety risks. Therefore, there is an urgent need for an innovative method that integrates the advantages of multiple monitoring methods in the air, space, and ground to achieve comprehensive, multi-level, and timely monitoring and early warning of mine safety risks. This approach can effectively improve mine safety management, reduce the probability of mine accidents, and protect the lives and property of mine workers and the stability of the ecological environment surrounding mines. Summary of the Invention

[0008] The purpose of the present invention is to provide a mine safety risk early warning method based on the integration of air, space and land. Through the integrated safety monitoring means of satellite remote sensing technology, aerial monitoring technology and ground monitoring technology, mine safety risks are warned, solving the problem that the existing limited ground sensor network and manual inspections have limitations and cannot be detected and warned in time.

[0009] To solve the above technical problems, the present invention is achieved through the following technical solutions:

[0010] The present invention is a mine safety risk early warning method based on air-ground integration, comprising the following steps:

[0011] Step S1, data acquisition: obtaining satellite remote sensing data, aerial monitoring data and ground monitoring data respectively;

[0012] Step S2, data transmission: transmitting the collected data information to the ground data center;

[0013] Step S3, data processing: formatting and denoising the collected data;

[0014] Step S4, feature extraction: extracting mine safety risk related features from the preprocessed data;

[0015] Step S5, model construction: using machine learning to construct a mine risk assessment model;

[0016] Step S6: issuing warnings: judging the warning level according to the output of the mine risk assessment model, and issuing warning information.

[0017] As a preferred technical solution, in step S1, the satellite remote sensing data is high-definition images of the mining area, collected by high-resolution optical satellites and synthetic aperture radar satellites, identifying static information such as the mining area and waste dump location. The aerial monitoring data is images of the mining operations and equipment and facilities, collected by drones and their onboard visible light cameras, thermal imagers, and lidar. The visible light camera is used to capture images of the mining operations, equipment, and facilities, enabling timely detection of equipment damage and personnel violations. Thermal imagers can detect heat sources within the mine. For example, in mines prone to spontaneous combustion, thermal imaging can detect abnormally high temperatures in ore piles in advance, thereby providing early warning of fire risks. LiDAR can construct a three-dimensional model of the mine and accurately measure changes in the mine's terrain, such as changes in slope gradient.

[0018] The ground monitoring data include the slope inclination angle, roof pressure and gas concentration in the mine, which are collected by tilt sensors, pressure sensors and gas sensors installed at different locations in the mine.

[0019] As a preferred technical solution, the satellite remote sensing data is transmitted to the ground data monitoring center via a satellite communication link; the aviation monitoring data is transmitted to the ground data monitoring center in real time via wireless communication; and the ground monitoring data is transmitted to the ground data monitoring center via wireless communication technology or a wired network.

[0020] As a preferred technical solution, in step S3, when the data format is unified, a template covering the key elements of the data is pre-set, specifically including the data acquisition time, data type, geographic location information, and data content, and the data is stored in JSON data format; when the data is denoised, the corresponding image data is processed by mean filtering, and the center pixel value is replaced by the average value of the pixel values ​​in the neighborhood around the pixel; the sensor data is subjected to Kalman filtering to dynamically calculate the optimal estimate based on the previous state estimate and the current measurement value to remove noise.

[0021] As a preferred technical solution, in step S4, the specific process of extracting mine safety risk-related features from image data is as follows:

[0022] Step SY41: Selecting a window size and a step size for the pre-processed image data;

[0023] Step SY42: Calculate texture features. For each window, calculate the statistical gray-level co-occurrence matrix according to the selected direction and distance. Let the image be I. The gray value of the pixel (x, y) is I(x, y). Then calculate the GLCM of the pixel with a distance of d pixels, where i = I(x, y) and j = (x, y + d). Count the number of (x, y) pairs that meet this condition in all windows to obtain the GLCM.

[0024] In the texture feature, the contrast calculation formula is as follows:

[0025]

[0026] The correlation calculation formula is as follows:

[0027]

[0028] Where L is the gray level, μ i 、μ j is the mean value in the i and j directions, σ i , σ j is the standard deviation;

[0029] Step SY43: For each central pixel, compare it with its surrounding adjacent pixels and count the frequencies of different LBP values ​​in the image as texture features;

[0030] Step SY44: Calculate shape features and perform Gaussian filtering. Let the processed Gaussian filtered image be G(x, y). Then use the Sobel operator to calculate the horizontal gradient G x and vertical gradient G y , direction θ=arctan(G y / G x ), perform non-maximum suppression and double threshold detection to obtain the edge;

[0031] Step SY45: Boundary point sequence for extraction (x n ,y n ),n=0,1,...,N-1, convert it into complex form z n =x n +iy n , perform discrete Fourier transform Take the first M (M<N) sequences as Fourier descriptors.

[0032] As a preferred technical solution, in step S4, the specific process of extracting mine safety risk-related features from sensor data is as follows:

[0033] Step SC41: For the time series x1, x2, ..., x n , calculate the mean variance By looping through the data points;

[0034] Step SC42: Calculate the conversion rate based on the data of adjacent points Calculate the rate of change for each time interval by looping through the time series data;

[0035] Step SC43: Assume that the time series of the sensor vibration signal is x(t), and calculate the frequency domain signal X(k);

[0036] Step SC44: Divide the signal into several segments, perform FFT on each segment, and then calculate the average power spectrum;

[0037] Step SC45: Assume signal γ(t), window function w(t), window length L, and overlapping length M. The PSD estimation formula for each signal segment is:

[0038]

[0039] Where U is the normalization factor.

[0040] As a preferred technical solution, in step S5, the specific process of using machine learning to construct a mine risk assessment model is as follows:

[0041] Step S51: Collect the mine data after preprocessing and feature extraction, label these data, and determine the safety risk level corresponding to each data sample; Collect the mine data after preprocessing and feature extraction. These data should contain various features related to safety risks, such as statistical features of sensor data, texture and shape features of image data, etc. At the same time, these data need to be labeled to determine the safety risk level corresponding to each data sample. Risk levels can be divided into low risk, medium risk and high risk, etc. The labeling is usually completed by mine safety experts based on historical accident data, on-site inspection conditions and relevant safety standards. For example, for gas concentration data, if the gas concentration is within the safety threshold range, it is labeled as low risk; if it is close to or exceeds the warning value, it is labeled as medium risk or high risk;

[0042] Step S52: Divide the collected labeled data into a training set, a validation set, and a test set. Generally speaking, the training set is used to train the model, the validation set is used to adjust the model's hyperparameters, and the test set is used to evaluate the model's final performance. A common division ratio is 70% for training, 15% for validation, and 15% for testing. However, the specific division ratio can be adjusted based on the amount of data and the complexity of the model.

[0043] Step S53: Use the rotational support vector machine as the classification model. For classification problems like mine safety risk assessment (determining risk levels), a variety of machine learning models can be used. If the relationship between data features and risk levels is complex, a neural network (such as a multilayer perceptron) is a good choice. It can automatically learn complex nonlinear relationships in the data. For example, a neural network with multiple hidden layers can learn the complex mapping between different combinations of sensor and image features and different risk levels.

[0044] Support vector machines (SVMs) are also a commonly used classification model, particularly when the data feature space is high-dimensional but the number of samples is relatively small. SVMs classify data into different risk categories by finding an optimal hyperplane. They work well for linearly separable data. For nonlinear cases, kernel functions (such as Gaussian kernels) can be used to map the data into a high-dimensional space, making it linearly separable.

[0045] Decision tree models (such as C4.5 and CART) can also be used for mine safety risk assessment. Decision trees classify data based on a series of conditional judgments (such as whether the gas concentration is above a certain threshold, whether the slope angle exceeds the safe range, etc.). Their advantage is that the model is highly interpretable and can intuitively demonstrate the decision-making process of risk assessment.

[0046] Step S54: Initialize the parameters of the model according to the selected model; for example, for a neural network, it is necessary to initialize the connection weights and bias terms between neurons. Generally, a random initialization method is adopted, such as randomly generating initial weights and biases within a certain range; for different models, there are different training algorithms. For example, a neural network can use the backpropagation algorithm to update parameters. At the same time, it is necessary to select a suitable loss function to measure the difference between the model prediction result and the true label. For multi-classification problems, the commonly used loss function is the cross entropy loss function. Taking the neural network as an example, in each training iteration, the prediction result is calculated by forward propagation, and then the value of the loss function is calculated, and then the model parameters are updated according to the gradient of the loss function through the backpropagation algorithm to gradually reduce the loss function;

[0047] Step S55: Use the back propagation algorithm to update the parameters and select a suitable loss function to measure the difference between the model prediction result and the true label;

[0048] Step S56: Use evaluation metrics to measure the performance of the model on the validation set and test set;

[0049] Step S57: Optimize model performance by adjusting the model's hyperparameters; Optimize model performance by adjusting the model's hyperparameters. Hyperparameters are parameters that need to be manually set before model training, such as the number of hidden layers, the number of neurons in each layer, the learning rate, etc. of the neural network, the penalty parameters and kernel function parameters of the SVM, etc. Grid search, random search, or more advanced Bayesian optimization methods can be used to find the optimal hyperparameter combination. For example, in grid search, a value range and step size are set for each hyperparameter, and then all possible hyperparameter combinations are trained and evaluated to find the combination with the best performance;

[0050] Step S58: Deploy the evaluated and optimized model into the mine safety risk early warning system. Ensure that the model can receive new data inputs in real time and output risk assessment results. This may require integrating the model into existing software systems or developing specialized interfaces to enable data transmission and feedback. As the mining environment changes, new data accumulates, and technology develops, the model needs to be regularly updated and maintained. For example, when the mining depth increases, the mining method changes, or new monitoring equipment is installed, the distribution and characteristics of the data may change. In this case, it is necessary to recollect data, retrain the model, or fine-tune the existing model to ensure its accuracy and effectiveness.

[0051] As a preferred technical solution, in step S6, when issuing warning information, a three-dimensional mine model is combined to perform a visual warning of risk anomalies, and the location information of the warning information is displayed on the three-dimensional mine model; the modeling formula of the three-dimensional mine model is as follows:

[0052]

[0053] Where ρ is the line of sight distance between the mine and the satellite, θ is the angle between the line of sight between the mine and the satellite and the local horizontal plane, η is the angle between the line of sight between the mine and the satellite and the satellite's radial direction, and R e is the effective radius of the earth, h is the altitude of the satellite, and φ is the angle between the sub-satellite point and the center of the earth.

[0054] As a preferred technical solution, the coverage area of ​​the mine three-dimensional model is According to the satellite's position S(x s ,y s ,z s ), the mine's geographical longitude and latitude location And the range of the mine elevation angle θ:

[0055]

[0056] Where Ω is the right ascension of the satellite's ascending node, and S is the Greenwich sidereal time at time t0.

[0057] The present invention has the following beneficial effects:

[0058] (1) The present invention obtains satellite remote sensing data, aerial monitoring data and ground monitoring data through air-space-ground technical means, and uses machine learning to construct a mine risk assessment model for the collected data, thereby achieving real-time safety warnings for mines from air-space-ground, improving the level of mine safety management, and reducing the probability of mine accidents.

[0059] (2) The present invention constructs a three-dimensional mine model through satellite remote sensing technology, and displays the collected satellite remote sensing data, aerial monitoring data and ground monitoring data on the three-dimensional mine model in real time. Once the mine risk assessment model outputs judgment warning information and level, it is displayed at the corresponding position of the three-dimensional mine model, which is convenient for administrators to quickly locate abnormal positions and perform maintenance in time.

[0060] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0062] Figure 1 This is a flow chart of a mine safety risk early warning method based on air-ground integration of the present invention. DETAILED DESCRIPTION

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0064] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0065] To make the purpose, technical solutions and advantages of this application clearer, Figure 1 The implementation methods of this application are described in further detail.

[0066] Before introducing the embodiments of the present application, the air-ground-space integration is first described.

[0067] Air-space-ground integration refers to the deep fusion of the sky (satellites, drones, and other aerial resources), the ground (various sensing devices), and space (remote sensing satellites) to form a new multi-layered, multi-access converged architecture. This architecture, through collaborative work, enables accurate information acquisition, rapid processing, and efficient transmission. It is widely used in smart city management, autonomous driving, vehicle-road collaboration, urban emergency logistics, emergency firefighting, and emergency rescue.

[0068] The specific components of air-ground-space integration include:

[0069] Sky layer: This includes low-orbit remote sensing satellites and low-altitude drones. These devices are responsible for data collection and transmission. Through the collaborative work of satellites and drones, real-time monitoring and data collection of ground conditions are achieved.

[0070] Ground layer: It mainly consists of various ground perception devices, such as cameras and sensors. These devices are responsible for data processing and analysis, providing real-time data support for decision-making.

[0071] Space layer: mainly remote sensing satellites, responsible for obtaining large amounts of geographic information data from space and providing basic data support for the entire system.

[0072] Application scenarios of air-space-ground integration:

[0073] Smart city management: Through satellite image comparison and drone on-site verification, accurate identification and management of illegal urban buildings can be achieved, improving the efficiency and accuracy of urban management.

[0074] Autonomous driving and vehicle-road collaboration: By integrating air and ground data, the environmental perception capabilities of autonomous vehicles are improved, enabling safer autonomous driving.

[0075] Emergency logistics and emergency rescue: When a disaster occurs, an integrated air-space-ground rapid response mechanism is used to achieve rapid deployment of emergency supplies and efficient execution of rescue operations.

[0076] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1 It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0077] See also Figure 1 As shown, the present invention is a mine safety risk early warning method based on air-ground integration, comprising the following steps:

[0078] Step S1, data acquisition: obtaining satellite remote sensing data, aerial monitoring data and ground monitoring data respectively;

[0079] Step S2, data transmission: transmitting the collected data information to the ground data center;

[0080] Step S3, data processing: formatting and denoising the collected data;

[0081] Step S4, feature extraction: extracting mine safety risk related features from the preprocessed data;

[0082] Step S5, model construction: using machine learning to construct a mine risk assessment model;

[0083] Step S6: issuing warnings: judging the warning level according to the output of the mine risk assessment model, and issuing warning information.

[0084] In step S1, the satellite remote sensing data is a high-definition image of the mining area, collected by high-resolution optical satellites and synthetic aperture radar satellites, which can identify static information such as the mining area and the location of waste slag piles. For example, by analyzing optical images from different periods, it can be determined whether the waste slag pile has abnormal signs of sliding or whether the mining area has exceeded the specified range. Synthetic aperture radar satellites are not restricted by weather and lighting conditions and can penetrate clouds to obtain data. They are very effective for monitoring subtle deformations of the mine surface. Through interferometric measurement technology, ground subsidence can be monitored with millimeter accuracy, which is extremely important for early warning of surface collapse that may be caused by underground mining.

[0085] The aerial monitoring data, which includes images of mining operations and equipment and facilities, is collected by drones equipped with visible light cameras, thermal imagers, and lidar. Visible light cameras capture images of mining operations, equipment, and facilities, enabling timely detection of equipment damage and personnel violations. Thermal imagers can detect heat sources within mines. For example, in mines prone to spontaneous combustion, thermal imaging can detect abnormally high temperatures in ore piles, providing early warning of fire risks. LiDAR can construct a three-dimensional model of the mine and accurately measure changes in the mine's terrain, such as changes in slope gradient.

[0086] The ground monitoring data are the slope inclination angle, roof pressure and gas concentration in the mine, which are collected by tilt sensors, pressure sensors and gas sensors installed at different locations in the mine. For example, by installing a tilt sensor on the slope of a mine, the slope's inclination angle can be monitored in real time. When the inclination angle exceeds a set threshold, it may indicate the risk of slope instability. Pressure sensors can be placed in underground tunnels to monitor roof pressure. Once the pressure changes abnormally, it may indicate that the roof is in danger of collapsing. In addition, gas sensors are installed in the mine to detect the concentration of harmful gases such as methane and carbon monoxide, and timely alarms are issued when the concentration exceeds the standard.

[0087] Satellite remote sensing data is transmitted to the ground data monitoring center via a satellite communication link. This method can ensure long-distance transmission of data and enable effective data return even in remote mining areas. The aerial monitoring data is transmitted to the ground data monitoring center in real time via wireless communication methods such as 4G / 5G networks. The ground monitoring data is transmitted to the ground data monitoring center via wireless communication technologies such as Zigbee and LoRa or wired networks.

[0088] In step S3, when the data is formatted, a template is pre-set that covers key data elements, including data acquisition time, data type, geographic location information, and data content. The data is then stored in JSON format, such as {"time":"2024-12-23 10:00:00","source":"ground_sensor_001","type":"pressure","location":{"longitude":116.3,"latitude":39.9,"altitude":500},"value":100.5}. Alternatively, specialized data conversion software or programming language libraries can be used for format conversion. For satellite remote sensing data, many remote sensing image processing software (such as ENVI and ErdasImagine) provide data format conversion functions. For example, raw image data from different satellite formats (such as Landsat's TIFF format and Sentinel's SAFE format) can be converted to the universal GeoTIFF format, which effectively preserves geospatial information. In terms of programming, Python's GDAL (Geospatial Data Abstraction Library) library is a powerful tool that can read and convert a variety of remote sensing data formats.

[0089] When denoising data, the image data is processed using a mean filter. This method replaces the central pixel value by calculating the average value of the pixels in the surrounding neighborhood. For example, for a 3×3 pixel neighborhood, the 9 pixel values ​​in the neighborhood are added together and divided by 9, and the average value is used as the new value of the central pixel. This method can effectively remove salt and pepper noise (black and white point noise in the image).

[0090] The mean filter method used above can also be used for denoising using a median filter. This method essentially sorts the pixel values ​​within a pixel neighborhood and then replaces the central pixel value with the median value. This method is very effective for removing impulse noise (such as isolated bright or dark spots in an image). For example, in mining images, if there are isolated abnormal bright spots caused by sensor failure, median filtering can effectively eliminate these noise points while preserving the image's edges and details.

[0091] Kalman filtering is used on sensor data to dynamically calculate the optimal estimate based on previous state estimates and current measurements, thereby removing noise. Kalman filtering is a more complex but more effective filtering method, based on state-space models and probability statistics. For pressure sensor data in underground mine tunnels, which is subject to various interference factors, Kalman filtering can dynamically calculate the optimal estimate based on previous state estimates and current measurements, effectively removing noise while also enabling prediction and correction of data.

[0092] In step S4, the specific process of extracting mine safety risk-related features from image data is as follows:

[0093] Step SY41: Select the window size and step size for the pre-processed image data; for satellite remote sensing images or drone visible light images of mines, GLCM can be used to describe the spatial distribution relationship of pixel gray levels in the image. To calculate GLCM, you need to first determine the direction (such as 0°, 45°, 90°, 135°) and distance (pixel interval). For example, calculate the GLCM with a direction of 0° and a distance of 1 pixel, and count the frequency of each gray value pair appearing in this direction and distance. Texture features such as contrast, correlation, energy, and entropy can be extracted from GLCM. Contrast reflects the severity of local brightness changes in the image. In mine images, high-contrast areas may indicate changes in terrain undulations or the boundaries between mined and unmined areas, which are related to slope stability and changes in mining range; correlation reflects the linear relationship between gray values ​​in space, which is helpful for analyzing the arrangement and structural integrity of mining facilities;

[0094] Step SY42: Calculate texture features. For each window, calculate the statistical gray-level co-occurrence matrix according to the selected direction and distance. Let the image be I. The gray value of the pixel (x, y) is I(x, y). Then calculate the GLCM of the pixel with a distance of d pixels, where i = I(x, y) and j = (x, y + d). Count the number of (x, y) pairs that meet this condition in all windows to obtain the GLCM.

[0095] In the texture feature, the contrast calculation formula is as follows:

[0096]

[0097] The correlation calculation formula is as follows:

[0098]

[0099] Where L is the gray level, μ i 、μ j is the mean value in the i and j directions, σ i , σ j is the standard deviation;

[0100] Step SY43: For each central pixel, compare it with the surrounding adjacent pixels and count the frequency of different LBP values ​​in the image as texture features; LBP is an operator used to describe the local texture features of an image. It generates a binary code to represent the local texture pattern by comparing the central pixel with its neighboring pixels. In mining images, LBP can be used to identify specific texture changes on the mine surface, such as changes in the shape of ore piles or damage to the road surface. For example, when the shape of an ore pile changes due to natural factors or mining activities, the LBP features on its surface will change, and this change can be discovered by comparing the LBP features before and after.

[0101] For example, Canny edge detection. Its specific algorithm steps include Gaussian filtering (for removing noise), calculating image gradient (through Sobel operator, etc.), non-maximum suppression (for refining edges) and double threshold detection (for determining true edges);

[0102] Step SY44: Calculate shape features and perform Gaussian filtering. Let the processed Gaussian filtered image be G(x, y). Then use the Sobel operator to calculate the horizontal gradient G x and vertical gradient G y , direction θ=arctan(G y / G x), perform non-maximum suppression and double threshold detection to obtain the edge; for some objects with obvious shapes in the mine (such as waste slag piles, buildings, etc.), their boundaries must be extracted first. Edge detection algorithms can be used, such as the Canny edge detection algorithm. The Canny algorithm determines the edge of the object by looking for pixels with drastic changes in grayscale intensity in the image. After extracting the boundary, the shape descriptor can be used to describe the shape characteristics of the object. For example, the Fourier descriptor is used to represent the shape of the object boundary. It regards the boundary curve as a periodic function and converts it to the frequency domain through Fourier transform. The low-frequency part represents the overall shape of the object, and the high-frequency part represents the details of the shape. For monitoring the stability of the waste slag pile, changes in shape (such as from an approximately circular shape to an irregular shape) may indicate the risk of sliding or collapse of the waste slag pile;

[0103] Step SY45: Boundary point sequence for extraction (x n ,y n ),n=0,1,...,N-1, convert it into complex form z n =x n +iy n , perform discrete Fourier transform Take the first M (M<N) sequences as Fourier descriptors.

[0104] In step S4, the specific process of extracting mine safety risk-related features from sensor data is as follows:

[0105] Step SC41: For the time series x1, x2, ..., x n , calculate the mean variance By looping through the data points;

[0106] Step SC42: Calculate the conversion rate based on the data of adjacent points Calculate the rate of change for each time interval by looping through the time series data;

[0107] Step SC43: Assume that the time series of the sensor vibration signal is x(t), and calculate the frequency domain signal X(k);

[0108] Step SC44: Divide the signal into several segments, perform FFT on each segment, and then calculate the average power spectrum;

[0109] Step SC45: Assume signal γ(t), window function w(t), window length L, and overlapping length M. The PSD estimation formula for each signal segment is:

[0110]

[0111] Where U is a normalization factor related to the window function. Finally, the PSD estimates for all segments are averaged to obtain the final PSD. In actual calculations, functions from signal processing libraries (such as scipy.signal.welch in Python) can also be used.

[0112] In step S5, the specific process of using machine learning to build a mine risk assessment model is as follows:

[0113] Step S51: collecting the pre-processed and feature-extracted mining data, labeling the data, and determining the safety risk level corresponding to each data sample;

[0114] Step S52: Divide the collected labeled data into a training set, a validation set, and a test set;

[0115] Step S53: rotating the support vector machine as a classification model;

[0116] Step S54: Initialize the parameters of the model according to the selected model;

[0117] Step S55: Use the back propagation algorithm to update the parameters and select a suitable loss function to measure the difference between the model prediction result and the true label;

[0118] Step S56: Use evaluation metrics to measure the performance of the model on the validation set and test set;

[0119] Step S57: Optimizing model performance by adjusting model hyperparameters;

[0120] Step S58: Deploy the evaluated and optimized model to the mine safety risk early warning system.

[0121] In step S6, when issuing warning information, a visual warning of risk anomalies is performed in combination with the three-dimensional mine model, and the location information of the warning information is displayed on the three-dimensional mine model; the modeling formula of the three-dimensional mine model is as follows:

[0122]

[0123] Where ρ is the line of sight distance between the mine and the satellite, θ is the angle between the line of sight between the mine and the satellite and the local horizontal plane, η is the angle between the line of sight between the mine and the satellite and the satellite's radial direction, and R e is the effective radius of the earth, h is the altitude of the satellite, and φ is the angle between the sub-satellite point and the center of the earth.

[0124] As a preferred technical solution, the coverage area of ​​the mine three-dimensional model is According to the satellite's position S(xs ,y s ,z s ), the mine's geographical longitude and latitude location And the range of the mine elevation angle θ:

[0125]

[0126] Where Ω is the right ascension of the satellite's ascending node, and S is the Greenwich sidereal time at time t0.

[0127] It is worth noting that in the above system embodiment, the various units included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0128] In addition, those skilled in the art will appreciate that all or part of the steps in the above-mentioned embodiments can be accomplished by instructing related hardware through a program, and the corresponding program can be stored in a computer-readable storage medium.

[0129] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A mine safety risk early warning method based on air-ground integration, characterized in that: The steps include: Step S1, data acquisition: obtaining satellite remote sensing data, aerial monitoring data and ground monitoring data respectively; Step S2, data transmission: transmitting the collected data information to the ground data center; Step S3, data processing: formatting and denoising the collected data; Step S4, feature extraction: extracting mine safety risk related features from the preprocessed data; Step S5, model construction: using machine learning to construct a mine risk assessment model; Step S6, warning release: judging the warning level according to the output of the mine risk assessment model and releasing warning information; In step S6, when issuing warning information, a visual warning of risk anomalies is performed in combination with the three-dimensional mine model, and the location information of the warning information is displayed on the three-dimensional mine model; the modeling formula of the three-dimensional mine model is as follows: ； Where, is the line-of-sight distance between the mine and the satellite, is the angle between the line of sight between the mine and the satellite and the local horizontal plane, is the angle between the line of sight between the mine and the satellite and the satellite's radial direction, is the effective radius of the Earth, is the satellite altitude, is the geocentric angle between the subsatellite point and the mine; The coverage area of ​​the mine 3D model is , according to satellite Position in the Earth-fixed coordinate system at any moment , the mine's geographical longitude and latitude location , and the elevation angle of the mine Scope: ； Where, is the right ascension of the satellite ascending node, S is Greenwich Mean Sidereal Time.

2. The mine safety risk early warning method based on air-ground integration according to claim 1 is characterized in that: In step S1, the satellite remote sensing data is a high-definition image of the mining area, which is collected by high-resolution optical satellites and synthetic aperture radar satellites; the aerial monitoring data is the mining operation images and equipment and facility images of the mine, which are collected by drones and visible light cameras, thermal imagers and lidars carried by drones; the ground monitoring data is the slope inclination angle, roof pressure and gas concentration in the mine, which are collected by tilt sensors, pressure sensors and gas sensors installed at different locations on the mine.

3. The mine safety risk early warning method based on air-ground integration according to claim 1 is characterized in that: In step S2, the satellite remote sensing data is transmitted to the ground data monitoring center via a satellite communication link; the aviation monitoring data is transmitted to the ground data monitoring center in real time via wireless communication; and the ground monitoring data is transmitted to the ground data monitoring center via wireless communication technology or a wired network.

4. The mine safety risk early warning method based on air-ground integration according to claim 1 is characterized in that: In step S3, when the data is formatted in a unified manner, a template covering key elements of the data is pre-set, specifically including data acquisition time, data type, geographic location information, and data content, and the data is stored in JSON data format; when the data is denoised, the corresponding image data is processed using mean filtering, and the center pixel value is replaced by the average value of the pixel values ​​in the neighborhood around the pixel; the sensor data is subjected to Kalman filtering to dynamically calculate the optimal estimate based on the previous state estimate and the current measurement value to remove noise.

5. The mine safety risk early warning method based on air-ground integration according to claim 1 is characterized in that: In step S4, the specific process of extracting mine safety risk-related features from image data is as follows: Step SY41: Selecting a window size and a step size for the pre-processed image data; Step SY42: Calculate texture features. For each window, calculate the statistical gray-level co-occurrence matrix according to the selected direction and distance. Suppose the image is , then the pixel The gray value is , then the calculated distance is The GLCM of the pixel, where , count all windows that meet this condition For the number of occurrences, GLCM is obtained; In the texture feature, the contrast calculation formula is as follows: ； The correlation calculation formula is as follows: ； Where, is the grayscale level, for The mean of the direction, is the standard deviation; Step SY43: For each central pixel, compare it with its surrounding adjacent pixels and count the frequencies of different LBP values ​​in the image as texture features; Step SY44: Calculate shape features and perform Gaussian filtering. Let the processed Gaussian filtered image be , then the Sobel operator calculates the horizontal gradient and vertical gradient ,direction , perform non-maximum suppression and double threshold detection to obtain the edge; Step SY45: Boundary point sequence for extraction , convert it to its plural form , perform discrete Fourier transform , take the front The sequence is used as the Fourier descriptor.

6. The mine safety risk early warning method based on air-ground integration according to claim 1 is characterized in that: In step S4, the specific process of extracting mine safety risk-related features from sensor data is as follows: Step SC41: For time series , calculate the mean ,variance , by looping through the data points; Step SC42: Calculate the conversion rate based on the data of adjacent points , by looping through the time series data to calculate the rate of change for each time interval; Step SC43: Assume that the sensor vibration signal time series is , calculate the frequency domain signal ; Step SC44: Divide the signal into several segments, perform FFT on each segment, and then calculate the average power spectrum; Step SC45: Set signal , the window function is , the window length is The length of the overlapping part is , then the PSD estimation formula for each signal segment is: ； Where, is the normalization factor.

7. The mine safety risk early warning method based on air-ground integration according to claim 1 is characterized in that: In step S5, the specific process of using machine learning to build a mine risk assessment model is as follows: Step S51: collecting the pre-processed and feature-extracted mining data, labeling the data, and determining the safety risk level corresponding to each data sample; Step S52: Divide the collected labeled data into a training set, a validation set, and a test set; Step S53: rotating the support vector machine as a classification model; Step S54: Initialize the parameters of the model according to the selected model; Step S55: Use the back propagation algorithm to update the parameters and select a suitable loss function to measure the difference between the model prediction result and the true label; Step S56: Use evaluation metrics to measure the performance of the model on the validation set and test set; Step S57: Optimizing model performance by adjusting model hyperparameters; Step S58: Deploy the evaluated and optimized model to the mine safety risk early warning system.

Images