Underground coal mining safety protection method and system

Abstract

The invention discloses an underground coal mining safety protection method and system, and relates to the technical field of underground safety protection, and the method comprises the steps: installing a high-definition camera and a sensor, and collecting the real-time video and data of a safety channel; extracting images from the real-time video frame by frame, and preprocessing the images; inputting the preprocessed image into a YOLOv7 model, and detecting obstacles and wall cracks in the image; analyzing the real-time data, and judging whether an abnormal condition exists or not; the detection condition and the abnormal condition are transmitted to a security channel monitoring center for early warning; when emergency evacuation is needed underground, an underground positioning system is used for determining the position of a person, an instant notice is sent through an application program, and a safe channel without problems is indicated to the person. According to the invention, early warning of abnormal conditions in the safety channel is carried out to ensure safe passage of the safety channel in emergency; when emergency evacuation is needed underground, navigation guidance is provided for personnel, so that the personnel can quickly arrive at a safety channel.

Description

Technical Field

[0001] The present invention relates to the technical field of underground safety protection, and in particular to an underground coal mining safety protection method and system. Background Art

[0002] At present, with the advancement of intelligent construction of coal mines, more and more high-tech means are applied to coal mine safety production, among which video monitoring technology and automated data analysis have become one of the important means to improve production efficiency and ensure personnel safety. However, in the complex and changeable mine environment, how to achieve efficient and reliable environmental monitoring and personnel safety management is still an urgent problem to be solved.

[0003] Traditional safety channel monitoring mostly relies on manual inspections or simple video recordings, lacking effective automated analysis tools to identify potential risk points. Existing cameras can only record images but cannot automatically analyze whether there are safety hazards; although some more advanced systems can achieve preliminary data collection, they face a high risk of false alarms or missed alarms under complex working conditions.

[0004] These traditional methods generally have problems such as slow response speed and low accuracy. Especially when emergencies occur, it is difficult to take effective response measures in a timely manner, which can easily lead to greater losses. Summary of the invention

[0005] The purpose of the present invention is to provide an underground coal mining safety protection method and system to solve the problems raised in the prior art.

[0006] To achieve the above object, the present invention provides the following technical solution: a safety protection method for underground coal mining, the method comprising:

[0007] S1. Install high-definition cameras and sensors to collect real-time video and real-time data of the safety channel;

[0008] S2, extracting images frame by frame from the real-time video and preprocessing the extracted images;

[0009] S3, input the preprocessed image into the YOLOv7 model to detect obstacles and wall cracks in the image;

[0010] S4. Analyze real-time data to determine whether there are any abnormal conditions in the safety channel;

[0011] S5, transmit the detection situation and abnormal situation to the safety channel monitoring center for early warning;

[0012] S6. When an emergency evacuation situation occurs underground, the underground positioning system is used to determine the location of personnel, and an instant notification is sent through the application to indicate a safe passage without problems.

[0013] According to the above scheme, in step S1:

[0014] The installation of high-definition cameras includes installing high-definition cameras at the turns and intersections of the safety passage, so that the high-definition cameras can capture the entire safety passage;

[0015] The sensors include pressure sensors, displacement sensors, dust sensors, methane sensors, oxygen sensors and carbon dioxide sensors;

[0016] The pressure sensor is installed on the bracket; the displacement sensor is installed on the wall of the safety passage and on the ceiling of the safety passage; the dust sensor, oxygen sensor and carbon dioxide sensor are installed in the travel route of the safety passage; the methane sensor is installed on the ceiling and floor of the safety passage;

[0017] The real-time data includes air pressure data in the safety passage, inclination or displacement data of the walls and ceiling of the safety passage, and dust concentration, methane concentration, oxygen concentration and carbon dioxide concentration in the safety passage;

[0018] Installing a variety of sensors and collecting a variety of data can comprehensively detect the environmental parameters within the mine and help complete more comprehensive safety monitoring.

[0019] According to the above scheme, in step S2:

[0020] The preprocessing includes image unification, image denoising, image enhancement and image normalization;

[0021] The image unification is to scale the image to a uniform size to facilitate model training and detection; the image denoising is to remove noise in the image using non-local mean denoising, which helps to more accurately detect targets in the image; the image enhancement is to apply histogram equalization to different areas of the image using adaptive histogram equalization to enhance local contrast and make the image clearer; the image normalization is to normalize the image pixel values ​​to between 0 and 1 to make image data from different sources consistent.

[0022] According to the above scheme, in step S3, the YOLOv7 model includes:

[0023] The historical safety passage images were randomly divided into training sets and test sets. The YOLOv7 model was trained with the training set to test its ability to detect obstacles and wall cracks. The detection ability of the trained YOLOv7 model was tested with the test set.

[0024] The YOLOv7 model includes an input layer, a feature extraction layer, a feature fusion layer, a target detection layer and a result output layer;

[0025] The YOLOv7 model is a target detection model that can quickly detect obstacles and wall cracks in real-time video with high accuracy and low latency.

[0026] The training set is input into the input layer, where the image is subjected to data enhancement to increase the diversity of the data and the generalization ability of the YOLOv7 model; the data enhancement includes randomly flipping, cropping, scaling and color transformation of the image to obtain feature images of different directions, different positions and different colors;

[0027] The image processed by the input layer enters the feature extraction layer; the feature extraction layer uses a convolutional neural network; the convolutional neural network gradually extracts low-level to high-level features of the image through a combination of convolutional layers and pooling layers; and through residual connections, information is effectively transmitted in the convolutional neural network, enhancing information flow and helping the YOLOv7 model learn deeper features;

[0028] The feature fusion layer includes a feature pyramid and a path aggregation network;

[0029] The feature map output by the feature extraction layer is input into the feature pyramid; the feature pyramid integrates feature maps from different levels from top to bottom, the high-level feature map has a larger receptive field and is suitable for detecting larger targets; the low-level feature map has a higher resolution and is suitable for detecting smaller targets; and through the integration of the feature pyramid, multi-scale targets are captured;

[0030] The path aggregation network transmits context information through lateral connections, interacts information between feature maps at different levels, and ensures that small targets can also be detected;

[0031] The combination of feature pyramid and path aggregation network can capture multi-scale targets;

[0032] The target detection layer includes anchor frame generation, position regression and classification;

[0033] The anchor frame generation generates a predefined anchor frame for each feature map. The anchor frame is a set of rectangular frames with different sizes and aspect ratios, covering the area where the target appears. The classification classifies the features in the anchor frame, and the model predicts the object category of each anchor frame and its specific position in the image to determine the bounding box of the target. The position regression adjusts the anchor frame position by learning the offset between the anchor frame and the true target bounding box, and outputs the category probability, which indicates the possibility that the target in the anchor frame belongs to each category.

[0034] The result output layer includes non-maximum suppression, threshold screening and visualization result output;

[0035] The non-maximum suppression removes candidate boxes with high overlap in the model output and only retains the boxes with the highest confidence to reduce redundant detection; the threshold screening screens the final detection results according to the set confidence threshold; the visualization result output draws the detected target box, classification information and confidence on the original image;

[0036] In the process of training the YOLOv7 model, an optimization loss function is used to adjust the parameters of the model so that the prediction results of the model are close to the true value; the loss function includes classification loss, position regression loss and confidence loss; the classification loss is expressed as the prediction accuracy of the model for the target category; the position regression loss is expressed as the prediction accuracy of the model for the target bounding box position; the confidence loss is expressed as the prediction accuracy of the model for whether the target exists or not.

[0037] According to the above scheme, after the YOLOv7 model training is completed, the test set is input into the YOLOv7 model for testing to obtain the performance index;

[0038] The performance indicators include precision, recall and average precision;

[0039] The accuracy is the ratio of the number of samples that are correctly classified to the total number of samples. The formula is as follows:

[0040]

[0041] Among them, P represents the accuracy; TP represents the number of positive samples correctly predicted by the model as positive samples; FP represents the number of negative samples incorrectly predicted by the model as positive samples;

[0042] The recall rate is the ratio of the number of correctly detected samples among all positive samples to the total number of positive samples. The formula is as follows:

[0043]

[0044] Among them, R represents the recall rate; FN represents the number of positive samples that the model incorrectly predicts as negative samples;

[0045] The average precision is a balance indicator between precision and recall, and the formula is as follows:

[0046]

[0047] Among them, AP represents the average precision; N represents the number of detection results; R k Expressed as the recall rate of the kth threshold, P k It is expressed as the precision of the kth threshold.

[0048] According to the above scheme, the preprocessed image is input into the YOLOv7 model, and the YOLOv7 model draws a rectangular frame of the detected obstacles and wall cracks on the image, indicates that the detected target is an obstacle or a wall crack, and indicates the confidence level of the detection result; the image of the detected obstacle and wall crack is transmitted to the safety channel monitoring center.

[0049] According to the above scheme, in step S4:

[0050] Preprocessing the real-time data includes data denoising, missing value processing and normalization processing; the data denoising uses a Kalman filter to remove noise and improve data accuracy; the missing value processing is used to fill or delete missing values ​​to ensure data integrity and avoid misjudgment due to missing data; the normalization processing normalizes the data to the same scale for easy comparison and analysis;

[0051] According to historical data, set the upper and lower limits of the normal range for the measurement value of each sensor; and compare the real-time data collected by the sensor with the set upper and lower limits of the normal range in real time; when the real-time data collected by a sensor exceeds the upper and lower limits of the set normal range, it is judged as an abnormal situation.

[0052] According to the above scheme, in step S5:

[0053] The detection conditions include the presence of obstacles in the safety passage and cracks in the walls of the safety passage;

[0054] The abnormal conditions include abnormal pressure in the safety passage, tilt or displacement of the wall or ceiling in the safety passage, excessive dust concentration in the safety passage, excessive methane concentration in the safety passage, too low or too high oxygen concentration in the safety passage, and too high carbon dioxide concentration in the safety passage;

[0055] The warning will display the detection situation, abnormal situation, images where the YOLOv7 model detects obstacles or wall cracks, and abnormal data values ​​in the abnormal situation.

[0056] According to the above scheme, in step S6:

[0057] According to the current emergency situation and the location of the personnel, the location of the safe passage closest to the personnel is analyzed; the safe passage includes no obstacles in the safe passage, no cracks in the wall, normal pressure, no tilt or displacement of the wall or ceiling, normal dust concentration, normal methane concentration, normal oxygen concentration and normal carbon dioxide concentration, otherwise it is an abnormal safe passage;

[0058] The application sends an instant notification including the location of the emergency and safe passage, displays the safe passage on a map, and provides navigation guidance to personnel by means of arrow indications and voice prompts.

[0059] Underground coal mining safety protection system, which includes data acquisition module, central processing module, safety channel monitoring center and personnel positioning and evacuation module;

[0060] The data acquisition module includes a high-definition camera, a pressure sensor, a displacement sensor, a dust sensor, a methane sensor, an oxygen sensor and a carbon dioxide sensor; the carbon dioxide sensor collects the carbon dioxide concentration in the safety channel; and transmits the collected real-time video and data to the central processing module;

[0061] The high-definition camera collects real-time video in the safety passage; the pressure sensor collects air pressure data in the safety passage; the displacement sensor collects inclination or displacement data of the walls and ceiling of the safety passage; the dust sensor collects dust concentration in the safety passage; the methane sensor collects methane concentration in the safety passage; the oxygen sensor collects oxygen concentration in the safety passage;

[0062] The central processing module includes a preprocessing module, a YOLOv7 model detection module and a real-time data analysis module; the preprocessing module preprocesses the real-time video and data; the YOLOv7 model detection module inputs the preprocessed real-time video into the YOLOv7 model, detects whether there are obstacles or wall cracks in the video, and transmits the detection results to the safety channel monitoring center; the real-time data analysis module compares the preprocessed real-time data with the set upper and lower limits of the normal range, determines whether there are abnormal conditions, and transmits the abnormal conditions to the safety channel monitoring center;

[0063] The preprocessing includes extracting images frame by frame from real-time video, and performing image unification, image denoising, image enhancement and image normalization on the images, and performing data denoising, missing value processing and normalization processing on real-time data;

[0064] The YOLOv7 model includes training and testing the YOLOv7 model using historical data, and stopping the training when the performance index of the YOLOv7 model is stable and no longer significantly improved; using the trained YOLOv7 model to detect whether there are obstacles or wall cracks in the real-time video;

[0065] The abnormal conditions include abnormal pressure in the safety passage, tilt or displacement of the wall or ceiling in the safety passage, excessive dust concentration in the safety passage, excessive methane concentration in the safety passage, too low or too high oxygen concentration in the safety passage, and too high carbon dioxide concentration in the safety passage;

[0066] The safety channel monitoring center receives the judgment results transmitted by the central processing module, and displays the detection situation and abnormal situation for early warning;

[0067] The personnel positioning and evacuation module uses the underground positioning system to determine the personnel's location, sends instant notifications through the application, provides navigation guidance to personnel through arrow indications and voice prompts, and indicates to personnel a safe passage without problems.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] 1. The present invention monitors the situation in the safety channel in real time through real-time video and data, and detects abnormal situations to trigger early warning;

[0070] 2. The present invention uses the YOLOv7 model and real-time data comparison to provide early warning of abnormal situations in the safety channel, ensuring that the safety channel can be safely passed in an emergency;

[0071] 3. The present invention provides navigation guidance to personnel through an application program, so that personnel can quickly reach a safe passage. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a flow chart of the steps of the underground coal mining safety protection method of the present invention;

[0073] Figure 2 It is a structural schematic diagram of the underground coal mining safety protection system of the present invention. DETAILED DESCRIPTION

[0074] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0075] Example: Figure 1-Figure 2 As shown, the present invention provides a technical solution, a safety protection method for underground coal mining;

[0076] Install high-definition cameras and sensors; install high-definition cameras at the turns and intersections of the safety passage to capture the entire safety passage; install pressure sensors on brackets to monitor air pressure data; install displacement sensors on the walls and ceilings of the safety passage to monitor tilt or displacement data; install dust sensors in the route of the safety passage to monitor dust concentration; install methane sensors on the ceiling and floor of the safety passage to monitor methane concentration; install oxygen sensors in the route of the safety passage to monitor oxygen concentration; install carbon dioxide sensors in the route of the safety passage to monitor carbon dioxide concentration;

[0077] Images are extracted frame by frame from real-time video, and the extracted images are preprocessed, the preprocessing including image unification, image denoising, image enhancement and image normalization. The image unification is to scale the image to a uniform size. The image denoising is to remove noise in the image using non-local mean denoising. The image enhancement is to apply histogram equalization to different areas of the image using adaptive histogram equalization to enhance local contrast. The image normalization is to normalize the image pixel values ​​to between 0 and 1.

[0078] The historical security channel images are randomly divided into training and test sets, with 70% as training set and 30% as test set.

[0079] The training set is input into the input layer, and the image is subjected to data enhancement to increase the diversity of the data and the generalization ability of the YOLOv7 model; the data enhancement includes randomly flipping, cropping, scaling and color transformation of the image to obtain feature images of different directions, different positions and different colors;

[0080] The image processed by the input layer enters the feature extraction layer; the feature extraction layer uses a convolutional neural network; the convolutional neural network gradually extracts low-level to high-level features of the image through a combination of convolutional layers and pooling layers; and through residual connections, information is effectively transmitted in the convolutional neural network, enhancing information flow and helping the YOLOv7 model learn deeper features;

[0081] The feature fusion layer includes a feature pyramid and a path aggregation network;

[0082] The feature map output by the feature extraction layer is input into the feature pyramid; the feature pyramid integrates feature maps from different levels from top to bottom, the high-level feature map has a larger receptive field and is suitable for detecting larger targets; the low-level feature map has a higher resolution and is suitable for detecting smaller targets; and through the integration of the feature pyramid, multi-scale targets are captured;

[0083] The path aggregation network transmits context information through lateral connections, interacts information between feature maps at different levels, and ensures that small targets can also be detected;

[0084] The target detection layer includes anchor frame generation, position regression and classification;

[0085] The anchor frame generation generates a predefined anchor frame for each feature map. The anchor frame is a set of rectangular frames with different sizes and aspect ratios, covering the area where the target appears. The classification classifies the features in the anchor frame, and the model predicts the object category of each anchor frame and its specific position in the image to determine the bounding box of the target. The position regression adjusts the anchor frame position by learning the offset between the anchor frame and the true target bounding box, and outputs the category probability, which indicates the possibility that the target in the anchor frame belongs to each category.

[0086] The result output layer includes non-maximum suppression, threshold screening and visualization result output;

[0087] The non-maximum suppression removes candidate boxes with high overlap in the model output and only retains the boxes with the highest confidence to reduce redundant detection; the threshold screening screens the final detection results according to the set confidence threshold; the visualization result output draws the detected target box, classification information and confidence on the original image;

[0088] In the process of training the YOLOv7 model, an optimization loss function is used to adjust the parameters of the model so that the prediction results of the model are close to the true value; the loss function includes classification loss, position regression loss and confidence loss; the classification loss is expressed as the prediction accuracy of the model for the target category; the position regression loss is expressed as the prediction accuracy of the model for the target bounding box position; the confidence loss is expressed as the prediction accuracy of the model for whether the target exists or not.

[0089] After the YOLOv7 model training is completed, the test set is input into the YOLOv7 model for testing to obtain performance indicators;

[0090] The performance indicators include precision, recall and average precision;

[0091] The accuracy is the ratio of the number of samples that are correctly classified to the total number of samples. The formula is as follows:

[0092]

[0093] Among them, P represents the accuracy; TP represents the number of positive samples correctly predicted by the model as positive samples; FP represents the number of negative samples incorrectly predicted by the model as positive samples;

[0094] The recall rate is the ratio of the number of correctly detected samples among all positive samples to the total number of positive samples. The formula is as follows:

[0095]

[0096] Among them, R represents the recall rate; FN represents the number of positive samples that the model incorrectly predicts as negative samples;

[0097] The average precision is a balance indicator between precision and recall, and the formula is as follows:

[0098]

[0099] Among them, AP represents the average precision; N represents the number of detection results; R k Expressed as the recall rate of the kth threshold, P k It is expressed as the precision of the kth threshold.

[0100] The preprocessed image is input into the YOLOv7 model. The YOLOv7 model draws a rectangular frame of the detected obstacles and wall cracks on the image, indicates that the detected target is an obstacle or a wall crack, and indicates the confidence level of the detection result. The image of the detected obstacle and wall crack is transmitted to the safety channel monitoring center.

[0101] Preprocessing the real-time data includes data denoising, missing value processing and normalization processing; the data denoising uses a Kalman filter to remove noise; the missing value processing is used to fill or delete missing values ​​to ensure data integrity; the normalization processing normalizes the data to the same scale;

[0102] Based on historical data, set upper and lower limits of the normal range for each sensor's measurement value;

[0103] Set the upper and lower limits of the normal range:

[0104] Normal pressure range: 980-1020hPa;

[0105] Normal range of wall or ceiling tilt or displacement: displacement 0mm-1mm or tilt angle 0-1 degree;

[0106] Normal range of dust concentration: 0mg / m 3 -5mg / m 3 ;

[0107] Normal range of methane concentration: 0%-1% volume ratio;

[0108] Normal range of oxygen concentration: 19.5%-23.5% volume ratio;

[0109] Normal range of carbon dioxide concentration: 0%-0.5% volume ratio;

[0110] Compare the real-time data collected by the sensor with the upper and lower limits of the set normal range in real time; when the real-time data collected by a sensor exceeds the upper and lower limits of the set normal range, it is judged as an abnormal situation;

[0111] Transmitting detection conditions and abnormal conditions to the safety passage monitoring center for early warning; the detection conditions include obstacles in the safety passage and cracks in the walls of the safety passage; the abnormal conditions include abnormal pressure in the safety passage, tilt or displacement of the walls or ceilings in the safety passage, excessive dust concentration in the safety passage, excessive methane concentration in the safety passage, too low or too high oxygen concentration in the safety passage, and too high carbon dioxide concentration in the safety passage;

[0112] The warning will display the detection situation, abnormal situation, images where the YOLOv7 model detects obstacles or wall cracks, and abnormal data values ​​in the abnormal situation.

[0113] According to the current emergency situation and the location of the personnel, the location of the safe passage closest to the personnel is analyzed; the safe passage includes no obstacles in the safe passage, no cracks in the wall, normal pressure, no tilt or displacement of the wall or ceiling, normal dust concentration, normal methane concentration, normal oxygen concentration and normal carbon dioxide concentration, otherwise it is an abnormal safe passage;

[0114] The application sends an instant notification including the location of the emergency and safe passage, displays the safe passage on a map, and provides navigation guidance to personnel by means of arrow indications and voice prompts.

[0115] The present invention provides a technical solution, an underground coal mining safety protection system: the system includes a data acquisition module, a central processing module, a safety channel monitoring center and a personnel positioning and evacuation module;

[0116] The data acquisition module includes a high-definition camera, a pressure sensor, a displacement sensor, a dust sensor, a methane sensor, an oxygen sensor and a carbon dioxide sensor; the carbon dioxide sensor collects the carbon dioxide concentration in the safety channel; and transmits the collected real-time video and data to the central processing module;

[0117] The high-definition camera collects real-time video in the safety passage; the pressure sensor collects air pressure data in the safety passage; the displacement sensor collects inclination or displacement data of the walls and ceiling of the safety passage; the dust sensor collects dust concentration in the safety passage; the methane sensor collects methane concentration in the safety passage; the oxygen sensor collects oxygen concentration in the safety passage;

[0118] The central processing module includes a preprocessing module, a YOLOv7 model detection module and a real-time data analysis module; the preprocessing module preprocesses the real-time video and data; the YOLOv7 model detection module inputs the preprocessed real-time video into the YOLOv7 model, detects whether there are obstacles or wall cracks in the video, and transmits the detection results to the safety channel monitoring center; the real-time data analysis module compares the preprocessed real-time data with the set upper and lower limits of the normal range, determines whether there are abnormal conditions, and transmits the abnormal conditions to the safety channel monitoring center;

[0119] The preprocessing includes extracting images frame by frame from real-time video, and performing image unification, image denoising, image enhancement and image normalization on the images, and performing data denoising, missing value processing and normalization processing on real-time data;

[0120] The YOLOv7 model includes training and testing the YOLOv7 model using historical data, and stopping the training when the performance index of the YOLOv7 model is stable and no longer significantly improved; using the trained YOLOv7 model to detect whether there are obstacles or wall cracks in the real-time video;

[0121] The abnormal conditions include abnormal pressure in the safety passage, tilt or displacement of the wall or ceiling in the safety passage, excessive dust concentration in the safety passage, excessive methane concentration in the safety passage, too low or too high oxygen concentration in the safety passage, and too high carbon dioxide concentration in the safety passage;

[0122] The safety channel monitoring center receives the judgment results transmitted by the central processing module, and displays the detection situation and abnormal situation for early warning;

[0123] The personnel positioning and evacuation module uses the underground positioning system to determine the personnel's location, sends instant notifications through the application, provides navigation guidance to personnel through arrow indications and voice prompts, and indicates to personnel a safe passage without problems.

[0124] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A safety protection method for underground coal mining, characterized in that: The method includes: S1. Install high-definition cameras and sensors to collect real-time video and real-time data of the safety channel; S2, extracting images frame by frame from the real-time video and preprocessing the extracted images; S3, input the preprocessed image into the YOLOv7 model to detect obstacles and wall cracks in the image; S4. Analyze real-time data to determine whether there are any abnormal conditions in the safety channel; S5, transmit the detection situation and abnormal situation to the safety channel monitoring center for early warning; S6. When an emergency evacuation situation occurs underground, the underground positioning system is used to determine the location of personnel, and an instant notification is sent through the application to indicate a safe passage without problems.

2. The underground coal mining safety protection method according to claim 1, characterized in that: In step S1: The installation of high-definition cameras includes installing high-definition cameras at the turns and intersections of the safety passage, so that the high-definition cameras can capture the entire safety passage; The sensors include pressure sensors, displacement sensors, dust sensors, methane sensors, oxygen sensors and carbon dioxide sensors; The pressure sensor is installed on the bracket; the displacement sensor is installed on the wall of the safety passage and on the ceiling of the safety passage; the dust sensor, oxygen sensor and carbon dioxide sensor are installed in the travel route of the safety passage; the methane sensor is installed on the ceiling and floor of the safety passage; The real-time data includes air pressure data in the safety passage, inclination or displacement data of the walls and ceiling of the safety passage, and dust concentration, methane concentration, oxygen concentration and carbon dioxide concentration in the safety passage.

3. The underground coal mining safety protection method according to claim 1, characterized in that: In step S2: The preprocessing includes image unification, image denoising, image enhancement and image normalization; The image unification is to scale the image to a uniform size; the image denoising is to remove noise in the image using non-local mean denoising; The image enhancement uses adaptive histogram equalization to apply histogram equalization to different areas of the image to enhance local contrast; The image normalization normalizes the image pixel values.

4. The underground coal mining safety protection method according to claim 1, characterized in that: In step S3, the YOLOv7 model includes: The historical safety passage images were randomly divided into training sets and test sets. The YOLOv7 model was trained with the training set to test its ability to detect obstacles and wall cracks. The detection ability of the trained YOLOv7 model was tested with the test set. The YOLOv7 model includes an input layer, a feature extraction layer, a feature fusion layer, a target detection layer and a result output layer; The training set is input into the input layer, where the image is subjected to data enhancement to increase the diversity of the data and the generalization ability of the YOLOv7 model; the data enhancement includes randomly flipping, cropping, scaling and color transformation of the image to obtain feature images of different directions, different positions and different colors; The image processed by the input layer enters the feature extraction layer; the feature extraction layer uses a convolutional neural network; the convolutional neural network gradually extracts low-level to high-level features of the image through a combination of convolutional layers and pooling layers; and through residual connections, information is effectively transmitted in the convolutional neural network, enhancing information flow and helping the YOLOv7 model learn deeper features; The feature fusion layer includes a feature pyramid and a path aggregation network; The feature map output by the feature extraction layer is input into the feature pyramid; the feature pyramid integrates feature maps from different levels from top to bottom, the high-level feature map has a larger receptive field and is suitable for detecting larger targets; the low-level feature map has a higher resolution and is suitable for detecting smaller targets; and through the integration of the feature pyramid, multi-scale targets are captured; The path aggregation network transmits context information through lateral connections, interacts information between feature maps at different levels, and ensures that small targets can also be detected; The target detection layer includes anchor frame generation, position regression and classification; The anchor frame generation generates a predefined anchor frame for each feature map. The anchor frame is a set of rectangular frames with different sizes and aspect ratios, covering the area where the target appears. The classification classifies the features in the anchor frame, and the model predicts the object category of each anchor frame and its specific position in the image to determine the bounding box of the target. The position regression adjusts the anchor frame position by learning the offset between the anchor frame and the true target bounding box, and outputs the category probability, which indicates the possibility that the target in the anchor frame belongs to each category. The result output layer includes non-maximum suppression, threshold screening and visualization result output; The non-maximum suppression removes candidate boxes with high overlap in the model output and only retains the boxes with the highest confidence to reduce redundant detection; the threshold screening screens the final detection results according to the set confidence threshold; the visualization result output draws the detected target box, classification information and confidence on the original image; In the process of training the YOLOv7 model, an optimization loss function is used to adjust the parameters of the model so that the prediction results of the model are close to the true value; the loss function includes classification loss, position regression loss and confidence loss; the classification loss is expressed as the prediction accuracy of the model for the target category; the position regression loss is expressed as the prediction accuracy of the model for the target bounding box position; the confidence loss is expressed as the prediction accuracy of the model for whether the target exists or not.

5. The underground coal mining safety protection method according to claim 4 is characterized in that: After the YOLOv7 model training is completed, the test set is input into the YOLOv7 model for testing to obtain performance indicators; The performance indicators include precision, recall and average precision; The accuracy is the ratio of the number of samples that are correctly classified to the total number of samples. The formula is as follows: Among them, P represents the accuracy; TP represents the number of positive samples correctly predicted by the model as positive samples; FP represents the number of negative samples incorrectly predicted by the model as positive samples; The recall rate is the ratio of the number of correctly detected samples among all positive samples to the total number of positive samples. The formula is as follows: Among them, R represents the recall rate; FN represents the number of positive samples that the model incorrectly predicts as negative samples; The average precision is a balance indicator between precision and recall, and the formula is as follows: Among them, AP represents the average precision; N represents the number of detection results; R k Expressed as the recall rate of the kth threshold, P k It is expressed as the precision of the kth threshold.

6. The underground coal mining safety protection method according to claim 5, characterized in that: The preprocessed image is input into the YOLOv7 model. The YOLOv7 model draws a rectangular frame of the detected obstacles and wall cracks on the image, indicates that the detected target is an obstacle or a wall crack, and indicates the confidence level of the detection result. The image of the detected obstacle and wall crack is transmitted to the safety channel monitoring center.

7. The underground coal mining safety protection method according to claim 2, characterized in that: In step S4: Preprocessing the real-time data, the preprocessing includes data denoising, missing value processing and normalization processing; the data denoising uses a Kalman filter to remove noise; The missing value processing is used to fill or delete missing values ​​to ensure data integrity; The normalization process normalizes the data to the same scale; Based on historical data, set upper and lower limits of the normal range for each sensor's measurement value; The real-time data collected by the sensor is compared with the upper and lower limits of the set normal range in real time; when the real-time data collected by the sensor exceeds the upper and lower limits of the set normal range, it is judged as an abnormal situation.

8. The underground coal mining safety protection method according to claim 1, characterized in that: In step S5: The detection conditions include the presence of obstacles in the safety passage and cracks in the walls of the safety passage; The abnormal conditions include abnormal pressure in the safety passage, tilt or displacement of the wall or ceiling in the safety passage, excessive dust concentration in the safety passage, excessive methane concentration in the safety passage, too low or too high oxygen concentration in the safety passage, and too high carbon dioxide concentration in the safety passage; The warning will display the detection situation, abnormal situation, images where the YOLOv7 model detects obstacles or wall cracks, and abnormal data values ​​in the abnormal situation.

9. The underground coal mining safety protection method according to claim 1, characterized in that: In step S6: According to the current emergency situation and the location of the personnel, the location of the safe passage closest to the personnel is analyzed; the safe passage includes no obstacles in the safe passage, no cracks in the wall, normal pressure, no tilt or displacement of the wall or ceiling, normal dust concentration, normal methane concentration, normal oxygen concentration and normal carbon dioxide concentration, otherwise it is an abnormal safe passage; The application sends an instant notification including the location of the emergency and safe passage, displays the safe passage on a map, and provides navigation guidance to personnel by means of arrow indications and voice prompts.

10. Underground coal mining safety protection system, characterized by: The system includes a data acquisition module, a central processing module, a safety channel monitoring center, and a personnel positioning and evacuation module; The data acquisition module includes a high-definition camera, a pressure sensor, a displacement sensor, a dust sensor, a methane sensor, an oxygen sensor and a carbon dioxide sensor; the carbon dioxide sensor collects the carbon dioxide concentration in the safety channel; and transmits the collected real-time video and data to the central processing module; The high-definition camera collects real-time video in the safety passage; the pressure sensor collects air pressure data in the safety passage; the displacement sensor collects inclination or displacement data of the walls and ceiling of the safety passage; the dust sensor collects dust concentration in the safety passage; the methane sensor collects methane concentration in the safety passage; the oxygen sensor collects oxygen concentration in the safety passage; The central processing module includes a preprocessing module, a YOLOv7 model detection module and a real-time data analysis module; the preprocessing module preprocesses the real-time video and data; the YOLOv7 model detection module inputs the preprocessed real-time video into the YOLOv7 model, detects whether there are obstacles or wall cracks in the video, and transmits the detection results to the safety channel monitoring center; the real-time data analysis module compares the preprocessed real-time data with the set upper and lower limits of the normal range, determines whether there are abnormal conditions, and transmits the abnormal conditions to the safety channel monitoring center; The preprocessing includes extracting images frame by frame from real-time video, and performing image unification, image denoising, image enhancement and image normalization on the images, and performing data denoising, missing value processing and normalization processing on real-time data; The YOLOv7 model includes training and testing the YOLOv7 model using historical data, and stopping the training when the performance index of the YOLOv7 model is stable and no longer significantly improved; using the trained YOLOv7 model to detect whether there are obstacles or wall cracks in the real-time video; The abnormal conditions include abnormal pressure in the safety passage, tilt or displacement of the wall or ceiling in the safety passage, excessive dust concentration in the safety passage, excessive methane concentration in the safety passage, too low or too high oxygen concentration in the safety passage, and too high carbon dioxide concentration in the safety passage; The safety channel monitoring center receives the judgment results transmitted by the central processing module, and displays the detection situation and abnormal situation for early warning; The personnel positioning and evacuation module uses the underground positioning system to determine the personnel's location, sends instant notifications through the application, provides navigation guidance to personnel through arrow indications and voice prompts, and indicates to personnel a safe passage without problems.

Images