Coal mine safety monitoring system and method based on big data analysis

Abstract

The invention discloses a coal mine safety monitoring system and method based on big data analysis, and belongs to the technical field of mine safety monitoring. A video processing chip integrated in the explosion-proof distribution box is used for segmenting the monitoring video to obtain a monitoring image, the hardware core of the processing module is a server cluster deployed in a ground data center, and a processor in the server cluster is used for carrying out image enhancement processing on the collected monitoring image; and whether the image quality after the image enhancement processing accords with the preset image quality is judged through an image quality detection chip built in the server. The method has the advantages that the peak signal-to-noise ratio of the image is increased by combining weighted guide filtering and FPGA edge detection, dead-corner-free comprehensive supervision of underground coal mine operation is achieved through global comprehensive monitoring, and coal mine monitoring images with good image quality can be obtained under the condition that the actual environment is severe.

Description

Technical Field

[0001] This invention belongs to the field of mine safety monitoring technology, specifically relating to a coal mine safety monitoring system and method based on big data analysis. Background Technology

[0002] Coal mine safety is a core issue that the coal mining industry continues to focus on. With the continuous development of coal mining technology, the complexity of the underground working environment and safety risks coexist. How to achieve efficient and accurate safety monitoring and reduce the occurrence of safety accidents has become the primary task of coal mine production enterprises.

[0003] In current technologies, coal mine safety monitoring largely relies on a combination of traditional monitoring equipment and manual inspections, which has several limitations: First, the unique underground environment, with its uneven lighting, high dust concentration, and frequent equipment vibrations, results in poor image quality from traditional monitoring systems, leading to blurry images and loss of detail, severely impacting the accuracy of subsequent analysis. Second, existing image enhancement technologies mostly employ traditional methods such as histogram equalization and filtering denoising, which are ill-suited to the dynamically changing underground environment. Their effectiveness is limited in low-light, high-noise scenarios, failing to meet the high-quality image requirements of intelligent analysis. Third, most systems separate image acquisition, enhancement, and intelligent analysis, lacking an integrated optimization mechanism, resulting in low data flow efficiency and an inability to utilize actual operational data for iterative model optimization. Fourth, existing solutions primarily focus on static image processing, failing to fully utilize video temporal information. This makes it difficult to adapt to dynamic interference such as sudden changes in lighting and dust accumulation, leading to monitoring delays, false alarms, and missed alarms. Summary of the Invention

[0004] The purpose of this invention is to provide a coal mine safety monitoring system and method based on big data analysis to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a coal mine safety monitoring system based on big data analysis, comprising:

[0006] The acquisition module is installed in the explosion-proof distribution box in the coal mine and the ground monitoring cabinet. It uses high-definition explosion-proof cameras deployed at key nodes in the underground roadways, around the mine entrance, and in the indoor operating room to collect monitoring videos of the coal mine location. The video processing chip integrated in the explosion-proof distribution box is used to segment the monitoring video to obtain monitoring images. The monitoring images include indoor monitoring images and outdoor monitoring images of coal mine exploration. The outdoor monitoring images include environmental image information around the coal mine, and the indoor monitoring images include working condition image information inside the coal mine.

[0007] Processing module: The hardware core is a server cluster deployed in a ground data center. The processor in the server cluster performs image enhancement processing on the acquired monitoring images. The image quality detection chip built into the server determines whether the image quality after image enhancement processing meets the preset image quality. If it does not meet the preset image quality, an alarm message is issued through the audible and visual alarm connected to the server, and the acquisition module is triggered to re-acquire and process the images. If it meets the preset image quality, the processed monitoring image is sent to the backend database via gigabit Ethernet.

[0008] Analysis module: Consists of a backend server and output devices. It uses the Spark big data analysis framework in the backend server to integrate and analyze all information in the database. The analysis results are transmitted via Ethernet to the monitoring center's large screen and mobile terminals to display real-time images and integrated analysis results.

[0009] Preferably, the acquisition module includes:

[0010] Extraction Unit: Utilizes GOLDWAVE and Cooledit software installed in an industrial computer to extract a training set of surveillance video images from the surveillance video;

[0011] Selection of Units: Based on the MATLAB computing environment in industrial computers, the training set of surveillance video images is vectorized, and a set of orthogonal eigenvectors is obtained by singular value decomposition according to the overall complex divergence matrix. The largest eigenvalue and its corresponding eigenvector based on the time series are selected and recorded as the selection result.

[0012] The first identification unit: a deep learning acceleration card mounted on an industrial computer, which identifies spatial transformation projection feature values ​​in the surveillance video based on the selection results through a convolutional neural network model;

[0013] The second identification unit relies on the parallel computing capabilities of the deep learning accelerator card to identify and extract feature information of the monitoring images in the surveillance video based on the spatial transformation projection feature values ​​and the YOLOv8 target detection algorithm. The feature information includes indoor and outdoor monitoring image information of the coal mine area.

[0014] Preferably, the processing module includes:

[0015] First processing unit: Relying on the GPU acceleration card of the server cluster, the monitoring image is enhanced with low frequency components through the OpenCV 4.8.0 computer vision library. This includes weighted guided filtering by combining guided filtering with edge perception weights. The filter window size is set to 5×5 to obtain the first image, which is the low frequency component enhancement image.

[0016] The second processing unit: Based on the CPU parallel computing capability of the server cluster, it uses a bilateral filtering image smoothing method to perform secondary processing on the low-frequency component enhancement image, with the filtering standard deviation set to 1.5, and simultaneously calculates the cost of the 5×5 image window to obtain the second image.

[0017] Construction Unit: The second image is processed by a dedicated image processing chip in the server cluster. The processing includes combining histogram equalization with local contrast enhancement to construct a brightness transformation function, and combining local value histogram equalization algorithm to improve the contrast and brightness of the second image, finally obtaining a reflection image corresponding to the second image.

[0018] Detection Unit: Relying on the high-speed computing power of the FPGA chip, the Canny edge detection algorithm is used to perform image edge detection on the reflected image to obtain the final enhanced monitoring image. The edge detection accuracy is ≤1 pixel.

[0019] Preferably, the processing module further includes:

[0020] The third processing unit: uses the digital signal processing module of the server cluster to perform peak signal-to-noise ratio noise reduction processing on the monitoring images that meet the preset image quality to obtain the first result;

[0021] Calculation Unit: A memory-based computing module based on a server cluster calculates the root mean square error of the first result to obtain the second result. The server's built-in comparator chip determines whether the second result is below a threshold. If it is below the threshold, the second result is complete; if it is above the threshold, the second result is incomplete, triggering the digital signal processing module to recalculate the error.

[0022] Preferably, when performing peak signal-to-noise ratio (PSNR) denoising, the third processing unit of the processing module will automatically adjust the denoising parameters according to the scene type of the monitored image.

[0023] A preferred method for coal mine safety monitoring based on big data analysis is as follows:

[0024] S1. Data Acquisition Stage: The high-definition explosion-proof camera of the acquisition module is activated to collect monitoring videos of underground roadways, the area around the mine entrance, and the indoor operating room. The monitoring videos are segmented by the video processing chip in the explosion-proof distribution box to generate indoor and outdoor monitoring images. At the same time, the underground gas concentration, temperature, and humidity data are collected synchronously by environmental sensors and embedded into the EXIF ​​information of the monitoring images.

[0025] S2, Feature Extraction Stage: The industrial computer runs GOLDWAVE and Cooledit software to extract 100 consecutive frames of images from the collected monitoring video to build a training set. In the MATLAB computing environment, the training set is converted into a 128-dimensional feature vector. Based on the total complex divergence matrix, the orthogonal feature vector is obtained through singular value decomposition. The maximum feature value of the time series and its corresponding vector are selected. The CNN model of the deep learning accelerator card is used to identify the spatial transformation projection feature value. Then, the YOLOv8 algorithm is used to extract image feature information. After feature extraction is completed, the data is temporarily stored in the local cache of the industrial computer.

[0026] S3, Image Processing Stage: The server cluster of the processing module receives image data transmitted by the acquisition module. The GPU accelerator card performs weighted guided filtering through the OpenCV 4.8.0 library to enhance the low-frequency components and obtain the first image. The CPU uses bilateral filtering to process the first image a second time, calculates the 5×5 window cost to generate the second image. The FPGA chip combines histogram equalization and local contrast enhancement to construct a brightness transformation function. It generates a reflection image through 16×16 block local value histogram equalization. Finally, the enhanced image is obtained through Canny edge detection. The image quality detection chip determines whether its peak signal-to-noise ratio is ≥35dB and its contrast ratio is ≥80:1. If it does not meet the standards, it triggers an audible and visual alarm and re-acquires the image. If it meets the standards, it is transmitted to the background database.

[0027] S4. Noise Reduction and Verification Stage: The digital signal processing module of the server cluster performs peak signal-to-noise ratio (PSNR) noise reduction on the qualified images and dynamically adjusts the parameters according to the scene type. The noise reduction intensity is increased by 30% in outdoor dim scenes and reduced by 20% in indoor bright scenes. The memory computing module performs 256×256 window root mean square error calculation on the denoised images. The comparator chip judges whether the result is ≤5. If it meets the standard, the image is confirmed to be complete. If it does not meet the standard, the noise reduction calculation is recalculated.

[0028] S5. Data Analysis and Display Stage: The backend server uses the Spark big data analysis framework to integrate and analyze images, feature data, and environmental parameters in the database, including personnel statistics in the underground area based on personnel feature tags, equipment contour matching detection, and gas concentration trend prediction. The analysis results are synchronously transmitted to the monitoring center's large screen and mobile terminals via Ethernet.

[0029] Preferably, in S1, the high-definition explosion-proof camera uses infrared night vision supplementary lighting, which automatically turns on when the underground light intensity is <10 lux. The camera is connected to the explosion-proof distribution box through a mining intrinsically safe cable. At the same time, the industrial computer communicates with the camera through an RS485 bus to obtain the camera's working status in real time, and triggers a local buzzer alarm when there is an abnormality.

[0030] Preferably, in S3, the server cluster adopts a redundant architecture design, and the master-slave switching is realized through the VRRP protocol. The backend database adopts a RAID5 redundant architecture. When a single hard drive fails, data reconstruction is used to ensure that the image is not lost. The database read and write speed is ≥1000MB / s, which meets the storage requirements of 100 frames per second.

[0031] Preferably, in S5, the Spark big data analytics framework is deployed on the Hadoop distributed platform, supporting the processing of more than 1,000 frames of image data per second. When an anomaly is detected in a device, the system will automatically retrieve the device's historical operating images from the past 24 hours, compare the current and historical outlines, generate an anomaly analysis report, and push it to administrators via a mobile terminal APP.

[0032] Preferably, the system also includes a self-inspection and maintenance process: during the daily downhole production slump, the camera lens cleanliness of the acquisition module is automatically checked, the memory redundancy of the server cluster of the processing module is checked, and the display function of the output device of the analysis module is tested. The self-inspection results generate a PDF report and are sent to the operation and maintenance management platform. At the same time, the self-inspection records of the past year are retained for traceability and query.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] 1. This invention, by combining weighted guided filtering and FPGA edge detection, improves the peak signal-to-noise ratio of images to ≥35dB, and the edge detection accuracy to ≤1 pixel. Simultaneously, the camera's infrared night vision function and the environmental sensor data embedding design avoid the limitations of single-image monitoring. For example, when gas concentration exceeds the limit, the corresponding area image can be quickly correlated, reducing the risk of misjudgment. Through comprehensive global monitoring, it achieves comprehensive, blind-spot-free supervision of underground coal mine operations, greatly improving the safety and efficiency of underground work. Furthermore, it achieves the technical effect of obtaining high-quality coal mine monitoring images even in harsh actual environments.

[0035] 2. The third processing unit of the processing module of this invention can automatically adjust the noise reduction parameters according to indoor / outdoor conditions and light intensity. Compared with the fixed parameter processing method of the prior art, this application is more flexible and can dynamically adapt to scene changes. In addition, the RS485 bus of the acquisition module monitors the camera status in real time and promptly alarms when the lens is blocked or the power supply is abnormal, avoiding the problem of no notification when the connection is lost in traditional systems.

[0036] 3. This invention integrates image acquisition, enhancement, analysis, and storage functions, supports multi-terminal interaction and multi-level early warning, reduces manual intervention, and realizes intelligent monitoring of the entire coal mine safety process. Attached Figure Description

[0037] Figure 1 This is a system flowchart of the present invention. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Example 1,

[0040] Please see Figure 1 This embodiment provides a coal mine safety monitoring method based on big data analysis. The monitoring method is as follows: S1, Data Acquisition Stage: The high-definition explosion-proof camera of the acquisition module is activated to collect monitoring videos of underground roadways, the area around the mine entrance, and the indoor operating room. The monitoring videos are segmented by the video processing chip in the explosion-proof distribution box to generate indoor and outdoor monitoring images. At the same time, underground gas concentration, temperature, and humidity data are collected synchronously by environmental sensors and embedded into the EXIF ​​information of the monitoring images. In S1, the high-definition explosion-proof camera uses infrared night vision supplementary lighting. It automatically turns on when the underground light intensity is <10 lux. The camera is connected to the explosion-proof distribution box through a mining intrinsically safe cable. At the same time, the industrial computer communicates with the camera through an RS485 bus to obtain the camera's working status in real time. In case of abnormality, a local buzzer alarm is triggered.

[0041] It should be noted that the underground environment of coal mines is characterized by high dust and high humidity. Intrinsically safe cables for mining must possess tensile strength and corrosion resistance. The recommended model is MYPTJ-3.6 / 6, whose signal attenuation rate can be controlled within 5% at a transmission distance ≤1000 meters, ensuring packet loss-free transmission of image and sensor data. Furthermore, the local buzzer alarm uses an intermittent sound design, which avoids continuous noise interference with underground operations while effectively alerting on-site personnel to troubleshoot camera malfunctions, such as lens obstruction by coal dust or poor power contact. When embedding image EXIF ​​information into the data collected by environmental sensors, JSON format is used for encapsulation, facilitating rapid parsing of related data by subsequent analysis modules.

[0042] S2. Feature Extraction Stage: The industrial computer runs GOLDWAVE and Cooledit software to extract 100 consecutive frames of images from the acquired monitoring video to construct a training set. In the MATLAB computing environment, the training set is converted into a 128-dimensional feature vector. Based on the total complex divergence matrix, orthogonal feature vectors are obtained through singular value decomposition. The largest feature value of the time series and its corresponding vector are selected, and the CNN model of the deep learning accelerator card is used to identify the spatial transformation projection feature value. Then, the YOLOv8 algorithm is used to extract image feature information. After feature extraction is completed, the data is temporarily stored in the local cache of the industrial computer.

[0043] It should be noted that GOLDWAVE software is mainly used to extract audio signals from videos, such as abnormal noises from equipment operation and shouts from people. Cooledit software assists in filtering video frames with synchronized audio. The two together can eliminate invalid frames caused by equipment vibration, light flicker, or blurry single frames without feature information, ensuring that the validity of the 100 frames in the training set is ≥98%. In the MATLAB computing environment, feature vector vectorization adopts a fusion method of grayscale value and texture feature. In the 128-dimensional vector, the first 64 dimensions are pixel grayscale statistics, and the last 64 dimensions are LBP texture features. The differences in image features of different scenes can be highlighted by calculating the overall complex divergence matrix. The feature vector corresponding to the largest feature value after singular value decomposition can improve the accuracy of video temporal correlation recognition. The underground personnel count uses a combination of YOLOv8 target detection and Kalman filter tracking, with double marking based on the color of personnel's safety helmets and work clothes. The overload threshold for personnel at the tunneling face can be customized according to the actual production capacity of the coal mine. When overload occurs, the system will simultaneously trigger a large-screen pop-up alarm and a push notification on the mobile terminal APP, with the push information including the overloaded area, the current number of people, and the overload duration. Historical images for equipment contour matching detection are retrieved from the InfluxDB time-series database. The SSD algorithm is used to compare the intersection-union ratio (IU / U) of the current contour with that of the historical contours. An IU / U is considered abnormal when it is less than 80%. The anomaly analysis report includes three core contents: the abnormal equipment number, a screenshot of the contour difference area, and the trend curve of the equipment operating parameters for the past 24 hours, to help managers quickly locate the cause of the fault.

[0044] S3, Image Processing Stage: The server cluster of the processing module receives image data transmitted from the acquisition module. The GPU accelerator card performs weighted guided filtering using the OpenCV 4.8.0 library to enhance low-frequency components and obtain the first image. The CPU uses bilateral filtering to process the first image a second time, calculating the 5×5 window cost to generate the second image. The FPGA chip constructs a brightness transformation function by combining histogram equalization and local contrast enhancement, generating a reflection image through 16×16 block local value histogram equalization. Finally, the enhanced image is obtained through Canny edge detection. The image quality detection chip determines whether its peak signal-to-noise ratio is ≥35dB and its contrast ratio is ≥80:1. If the standards are not met, an audible and visual alarm is triggered and re-acquisition is performed. If the standards are met, the image is transmitted to the background database. In S3, the server cluster adopts a redundant architecture design, realizing master-slave switching through the VRRP protocol. The background database adopts a RAID5 redundant architecture. In the event of a single hard drive failure, data reconstruction ensures that the image is not lost, and the database read / write speed is ≥1000MB / s, meeting the storage requirements of 100 frames per second.

[0045] It should be noted that the edge perception weights of the weighted guided filter are dynamically adjusted by calculating pixel gradient values. A larger gradient value results in a higher weight, avoiding the edge blurring problem caused by traditional guided filtering. The 5×5 filter window size balances processing efficiency and detail preservation, with a single frame image processing time of ≤0.1 seconds. When the standard deviation of the bilateral filter is set to 1.5, it can filter out high-frequency noise caused by underground dust without excessively smoothing image textures. The cost calculation for the 5×5 image window is based on the Euclidean distance of pixel grayscale, with a cost threshold set to 20. When the average pixel cost within the window exceeds the threshold, it is determined to be a local image anomaly, such as coal blocks obscuring the lens, requiring re-acquisition. The 16×16 block size can accurately improve the brightness of dark areas in the image, making the brightness of reflected images more uniform.

[0046] S4. Noise Reduction and Verification Stage: The digital signal processing module of the server cluster performs peak signal-to-noise ratio (PSNR) noise reduction on the qualified images and dynamically adjusts the parameters according to the scene type. The noise reduction intensity is increased by 30% in outdoor dim scenes and reduced by 20% in indoor bright scenes. The memory computing module performs 256×256 window root mean square error calculation on the denoised images. The comparator chip judges whether the result is ≤5. If it meets the standard, the image is confirmed to be complete. If it does not meet the standard, the noise reduction calculation is repeated.

[0047] S5, Data Analysis and Display Phase: The backend server uses the Spark big data analytics framework to integrate and analyze images, feature data, and environmental parameters within the database. This includes personnel statistics in underground areas based on personnel feature tags, equipment contour matching detection, and gas concentration trend prediction. The analysis results are synchronously transmitted to the monitoring center's large screen and mobile terminals via Ethernet. In S5, the Spark big data analytics framework is deployed on a Hadoop distributed platform, supporting the processing of over 1000 frames of image data per second. When equipment anomalies are detected, the system automatically retrieves historical operating images of the equipment from the past 24 hours, compares the current contours with historical ones, generates an anomaly analysis report, and pushes it to management personnel via a mobile app.

[0048] It also includes system self-inspection and maintenance processes: during the daily downtime of underground production, the system automatically checks the lens cleanliness of the cameras in the acquisition module, performs memory redundancy verification on the server cluster of the processing module, and tests the display function of the output devices of the analysis module. The self-inspection results generate a PDF report and send it to the operation and maintenance management platform. At the same time, the system retains self-inspection records for nearly one year for traceability and query.

[0049] It should be noted that lens cleanliness testing calculates an image clarity score. A score ≥80 indicates that cleaning is up to standard; a score <80 generates a "cleaning reminder" work order, which includes the camera number, installation location, and suggested cleaning time. Server memory redundancy verification uses ECC technology. When the error rate >0.001%, the system automatically marks the faulty memory slot and pushes a "memory replacement" warning to the operations and maintenance platform to avoid data calculation errors due to memory errors. Display function testing covers large screen bright line detection and mobile terminal APP lag testing. The self-test report PDF contains four columns: "Device Name, Test Items, Results, and Handling Suggestions." Self-test records for the past year are stored in partitions and support quick retrieval by device number and test date.

[0050] Example 2:

[0051] This embodiment provides a coal mine safety monitoring system based on big data analysis. The system includes: a data acquisition module: installed in an explosion-proof distribution box underground and a ground monitoring cabinet in the coal mine, using high-definition explosion-proof cameras deployed at key nodes in underground roadways, around the mine entrance, and in indoor operating rooms to acquire monitoring videos of the coal mine's location. The monitoring videos are segmented into monitoring images by a video processing chip integrated in the explosion-proof distribution box. The monitoring images include indoor and outdoor monitoring images of coal mine exploration, where outdoor monitoring images include environmental image information around the coal mine, and indoor monitoring images include working condition image information inside the coal mine. A processing module: the hardware core is a server cluster deployed in a ground data center. The processor in the server cluster performs image enhancement processing on the acquired monitoring images, and the image quality detection chip built into the server determines whether the image quality after image enhancement processing meets the preset image quality. If it does not meet the preset image quality, an alarm is issued through an audible and visual alarm connected to the server, triggering the data acquisition module to re-acquire and process the images. If it meets the preset image quality, the processed monitoring images are sent to the background database via gigabit Ethernet.

[0052] In this embodiment, the image preprocessing includes, but is not limited to, grayscale transformation, binary transformation, and noise reduction. Then, the Adaboost-based method is used to detect and locate the video. If a valid video is detected, it is saved in the database.

[0053] Analysis module: Consists of a backend server and output devices. It uses the Spark big data analysis framework in the backend server to integrate and analyze all information in the database. The analysis results are transmitted via Ethernet to the monitoring center's large screen and mobile terminals to display real-time images and integrated analysis results.

[0054] The data acquisition module includes:

[0055] Extraction Unit: Extracts a training set of surveillance video images from surveillance videos using GOLDWAVE and Cooledit software installed on an industrial computer.

[0056] Furthermore, the aforementioned training set of surveillance video images needs to be preprocessed to reduce noise and output clearer audio.

[0057] Unit selection: Based on the MATLAB computing environment in industrial computers, the training set of monitoring video images is vectorized, and a set of orthogonal eigenvectors is obtained by singular value decomposition according to the overall complex divergence matrix. The largest eigenvalue and its corresponding eigenvector based on the time series are selected and denoted as the selection result.

[0058] The first recognition unit: a deep learning accelerator card mounted on an industrial computer, which identifies spatial transformation projection feature values ​​in the surveillance video based on the selection results through a convolutional neural network model.

[0059] The second identification unit relies on the parallel computing capabilities of the deep learning accelerator card to identify and extract feature information of the monitoring images in the surveillance video based on the spatial transformation projection feature values ​​and the YOLOv8 target detection algorithm. The feature information includes indoor and outdoor monitoring image information of the coal mine area.

[0060] It should be noted that a video image background removal module can be used to acquire real-time video images from a video acquisition device, and an image background dehazing method can be used to process irrelevant backgrounds in the video images. The processed image retains the video image that needs to be identified, which is recorded as the first image to be identified, and the first image to be identified is sent to the video image detection module.

[0061] The processing module includes:

[0062] The first processing unit: relying on the GPU acceleration card of the server cluster, the monitoring image is enhanced with low-frequency components through the OpenCV 4.8.0 computer vision library. This includes weighted guided filtering by combining guided filtering with edge perception weights. The filter window size is set to 5×5 to obtain the first image, which is the low-frequency component enhancement image.

[0063] It should be noted that for harsh environments, such as those with insufficient or excessive lighting conditions, the image enhancement processing effect is not obvious. By combining computer image enhancement processing methods with actual lighting condition adjustment methods, it is possible to obtain coal mine monitoring images with better image quality even in harsh environments.

[0064] The second processing unit: Based on the parallel computing capabilities of the server cluster's CPU, it uses a bilateral filtering image smoothing method to perform secondary processing on the low-frequency component enhancement image, with the filtering standard deviation set to 1.5. At the same time, it calculates the cost of the 5×5 image window to obtain the second image.

[0065] Construction Unit: The second image is processed by a dedicated image processing chip in the server cluster. The processing includes combining histogram equalization with local contrast enhancement to construct a brightness transformation function, and combining local value histogram equalization algorithm to improve the contrast and brightness of the second image, ultimately obtaining a reflection image corresponding to the second image.

[0066] Detection Unit: Relying on the high-speed computing power of the FPGA chip, the Canny edge detection algorithm is used to perform image edge detection on the reflected image to obtain the final enhanced monitoring image. The edge detection accuracy is ≤1 pixel.

[0067] It should be noted that, in response to the problem of blurred or false edges in images, this invention uses guided filtering combined with edge-aware weights to form a weighted guided filter, thereby enhancing the low-frequency components of the image and avoiding the appearance of false or blurred edges.

[0068] The processing module also includes:

[0069] The third processing unit uses the digital signal processing module of the server cluster to perform peak signal-to-noise ratio (PSNR) denoising on the monitoring images that meet the preset image quality, and obtains the first result.

[0070] Calculation Unit: A memory-based computing module based on a server cluster calculates the root mean square error of the first result to obtain the second result. The server's built-in comparator chip determines whether the second result is below a threshold. If it is below the threshold, the second result is complete; if it is above the threshold, the second result is incomplete, triggering the digital signal processing module to recalculate the error.

[0071] It should be noted that image denoising includes both subjective and objective evaluation. Subjective evaluation refers to assessment through human visual observation, while objective evaluation uses established evaluation metrics to determine the effectiveness of image denoising. A higher mean square signal-to-noise ratio (MSNR) between the denoised image and the original image indicates a better denoising effect.

[0072] When performing peak signal-to-noise ratio (PSNR) denoising, the third processing unit of the processing module automatically adjusts the denoising parameters according to the scene type of the monitored image.

[0073] In summary, comprehensive monitoring across the entire network enables complete and seamless supervision of underground coal mine operations, significantly improving the safety and efficiency of underground work. Furthermore, it provides high-quality coal mine monitoring images even in harsh environments, achieving the goal of intelligent monitoring.

[0074] Working principle and usage process of this invention: Usage method: System initialization is required upon first use. The specific operation process is as follows:

[0075] Hardware connection: Deploy cameras, sensors, servers and other equipment as required by the implementation plan, connect mining cables / fiber optic cables, and ensure that the explosion-proof power distribution box and server cluster are properly grounded.

[0076] Software configuration: Install GOLDWAVE, Cooledit, and MATLAB software on the industrial computer, configure the video segmentation frequency and feature vector dimension, deploy OpenCV 4.8.0 and Spark framework on the server cluster, set the image quality threshold and root mean square error threshold, and install a dedicated APP on the mobile terminal to complete the IP binding with the server.

[0077] Parameter calibration: Activate the environmental sensor, introduce standard gas, set the standard temperature, calibrate the sensor data accuracy, adjust the camera angle to ensure no blind spots in monitoring coverage, and test the infrared night vision function.

[0078] The daily operating procedures are as follows:

[0079] Real-time monitoring: Monitoring center personnel can view real-time images of various areas on a large screen. Clicking on the image can zoom in on details. The mobile terminal APP can switch between partition views such as "coal mining face" and "wellhead" and receive abnormal alarm push notifications.

[0080] Data Query: In the server management interface, historical data can be retrieved by "time, region, type" etc. It supports exporting PDF analysis reports, including people statistics and equipment status curves. Historical data can be traced back up to 3 years of records.

[0081] Manual intervention: When the system alarms, maintenance personnel can remotely restart the equipment via an industrial computer. If the image quality continues to be substandard, the cleanliness of the lens and the status of the cable connection should be checked on-site, and the camera or sensor should be replaced if necessary.

[0082] Daily self-check: Log in to the operation and maintenance platform every day to check the self-check report at midnight, focusing on indicators such as "lens cleanliness score" and "server memory error rate", and arrange manual maintenance when necessary.

[0083] Monthly calibration: Environmental sensors are calibrated once a month, and memory redundancy tests are performed on the server cluster to simulate primary server failure and verify whether standby switching is normal.

[0084] Quarterly maintenance: Inspect the explosion-proof housing of the camera and the insulation of the cable every quarter, replace aging parts, clean up redundant data in the database, and ensure that the remaining capacity of the storage array is ≥30%.

[0085] Workflow: 30 minutes before each day's downhole production, maintenance personnel log in to the system management platform, click the "System Start" button, and the acquisition module, processing module, and analysis module are powered on and started in sequence. The industrial computer automatically detects the online status of the cameras and sensors, and enters standby mode after all are online.

[0086] Operational Phase: Cameras capture video at 25 frames per second. The video processing chip segments each image into two frames. Environmental sensors collect data once per second and embed EXIF ​​information into the images. Data is transmitted to a ground-based industrial computer via cable / fiber optic cable. The industrial computer extracts 100 frames from the video every 10 minutes to build a training set, performing feature vector transformation, singular value decomposition, CNN recognition, and YOLOv8 feature extraction. Data is temporarily cached locally. The server cluster receives data from the industrial computer every 5 minutes, sequentially performing weighted guided filtering, bilateral filtering, FPGA detail processing, and Canny edge detection. An image quality detection chip determines whether the image meets the standards; if it does, it is uploaded to the database; otherwise, an alarm is triggered and re-acquisition is initiated. The digital signal processing module adjusts noise reduction parameters according to the scene, performing noise reduction on compliant images. The in-memory computing module calculates the root mean square error (RMSE); if the result is ≤5, the image is considered complete; otherwise, noise reduction is repeated. The Spark framework analyzes the database data once per second, performing personnel statistics, equipment anomaly detection, and gas trend prediction. Results are synchronized to a large screen and mobile terminals, triggering alarm push notifications in case of anomalies.

[0087] When maintenance personnel click "System Shutdown", the system first stops data collection, then shuts down the analysis module and processing module in sequence, and finally generates a daily operation report, including the number of alarms, device online rate, data storage volume, etc., which is automatically uploaded to the maintenance platform before power is cut off.

[0088] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A coal mine safety monitoring system based on big data analysis, characterized in that, include: The acquisition module is installed in the explosion-proof distribution box in the coal mine and the ground monitoring cabinet. It uses high-definition explosion-proof cameras deployed at key nodes in the underground roadways, around the mine entrance, and in the indoor operating room to collect monitoring videos of the coal mine location. The video processing chip integrated in the explosion-proof distribution box is used to segment the monitoring video to obtain monitoring images. The monitoring images include indoor monitoring images and outdoor monitoring images of coal mine exploration. The outdoor monitoring images include environmental image information around the coal mine, and the indoor monitoring images include working condition image information inside the coal mine. Processing module: The hardware core is a server cluster deployed in a ground data center. The processor in the server cluster performs image enhancement processing on the acquired monitoring images. The image quality detection chip built into the server determines whether the image quality after image enhancement processing meets the preset image quality. If it does not meet the preset image quality, an alarm message is issued through the audible and visual alarm connected to the server, and the acquisition module is triggered to re-acquire and process the images. If it meets the preset image quality, the processed monitoring image is sent to the backend database via gigabit Ethernet. Analysis module: Consists of a backend server and output devices. It uses the Spark big data analysis framework in the backend server to integrate and analyze all information in the database. The analysis results are transmitted via Ethernet to the monitoring center's large screen and mobile terminals to display real-time images and integrated analysis results.

2. The coal mine safety monitoring system based on big data analysis according to claim 1, characterized in that, The acquisition module includes: Extraction Unit: Utilizes GOLDWAVE and Cooledit software installed in an industrial computer to extract a training set of surveillance video images from the surveillance video; Selection of Units: Based on the MATLAB computing environment in industrial computers, the training set of surveillance video images is vectorized, and a set of orthogonal eigenvectors is obtained by singular value decomposition according to the overall complex divergence matrix. The largest eigenvalue and its corresponding eigenvector based on the time series are selected and recorded as the selection result. The first identification unit: a deep learning acceleration card mounted on an industrial computer, which identifies spatial transformation projection feature values ​​in the surveillance video based on the selection results through a convolutional neural network model; The second identification unit relies on the parallel computing capabilities of the deep learning accelerator card to identify and extract feature information of the monitoring images in the surveillance video based on the spatial transformation projection feature values ​​and the YOLOv8 target detection algorithm. The feature information includes indoor and outdoor monitoring image information of the coal mine area.

3. The coal mine safety monitoring system based on big data analysis according to claim 1, characterized in that, The processing module includes: First processing unit: Relying on the GPU acceleration card of the server cluster, the monitoring image is enhanced with low frequency components through the OpenCV 4.8.0 computer vision library. This includes weighted guided filtering by combining guided filtering with edge perception weights. The filter window size is set to 5×5 to obtain the first image, which is the low frequency component enhancement image. The second processing unit: Based on the CPU parallel computing capability of the server cluster, it uses a bilateral filtering image smoothing method to perform secondary processing on the low-frequency component enhancement image, with the filtering standard deviation set to 1.5, and simultaneously calculates the cost of the 5×5 image window to obtain the second image. Construction Unit: The second image is processed by a dedicated image processing chip in the server cluster. The processing includes combining histogram equalization with local contrast enhancement to construct a brightness transformation function, and combining local value histogram equalization algorithm to improve the contrast and brightness of the second image, finally obtaining a reflection image corresponding to the second image. Detection Unit: Relying on the high-speed computing power of the FPGA chip, the Canny edge detection algorithm is used to perform image edge detection on the reflected image to obtain the final enhanced monitoring image. The edge detection accuracy is ≤1 pixel.

4. The coal mine safety monitoring system based on big data analysis according to claim 1, characterized in that, The processing module further includes: The third processing unit: uses the digital signal processing module of the server cluster to perform peak signal-to-noise ratio noise reduction processing on the monitoring images that meet the preset image quality to obtain the first result; Calculation Unit: A memory-based computing module based on a server cluster calculates the root mean square error of the first result to obtain the second result. The server's built-in comparator chip determines whether the second result is below a threshold. If it is below the threshold, the second result is complete; if it is above the threshold, the second result is incomplete, triggering the digital signal processing module to recalculate the error.

5. A coal mine safety monitoring system based on big data analysis according to claim 4, characterized in that, When performing peak signal-to-noise ratio (PSNR) denoising, the third processing unit of the processing module automatically adjusts the denoising parameters according to the scene type of the monitored image.

6. A coal mine safety monitoring method based on big data analysis, applicable to the coal mine safety monitoring system based on big data analysis as described in any one of claims 1-5, characterized in that, The monitoring methods are as follows: S1. Data Acquisition Stage: The high-definition explosion-proof camera of the acquisition module is activated to collect monitoring videos of underground roadways, the area around the mine entrance, and the indoor operating room. The monitoring videos are segmented by the video processing chip in the explosion-proof distribution box to generate indoor and outdoor monitoring images. At the same time, the underground gas concentration, temperature, and humidity data are collected synchronously by environmental sensors and embedded into the EXIF ​​information of the monitoring images. S2, Feature Extraction Stage: The industrial computer runs GOLDWAVE and Cooledit software to extract 100 consecutive frames of images from the collected monitoring video to build a training set. In the MATLAB computing environment, the training set is converted into a 128-dimensional feature vector. Based on the total complex divergence matrix, the orthogonal feature vector is obtained through singular value decomposition. The maximum feature value of the time series and its corresponding vector are selected. The CNN model of the deep learning accelerator card is used to identify the spatial transformation projection feature value. Then, the YOLOv8 algorithm is used to extract image feature information. After feature extraction is completed, the data is temporarily stored in the local cache of the industrial computer. S3, Image Processing Stage: The server cluster of the processing module receives image data transmitted by the acquisition module. The GPU accelerator card performs weighted guided filtering through the OpenCV 4.8.0 library to enhance the low-frequency components and obtain the first image. The CPU uses bilateral filtering to process the first image a second time, calculates the 5×5 window cost to generate the second image. The FPGA chip combines histogram equalization and local contrast enhancement to construct a brightness transformation function. It generates a reflection image through 16×16 block local value histogram equalization. Finally, the enhanced image is obtained through Canny edge detection. The image quality detection chip determines whether its peak signal-to-noise ratio is ≥35dB and its contrast ratio is ≥80:

1. If it does not meet the standards, it triggers an audible and visual alarm and re-acquires the image. If it meets the standards, it is transmitted to the background database. S4. Noise Reduction and Verification Stage: The digital signal processing module of the server cluster performs peak signal-to-noise ratio (PSNR) noise reduction on the qualified images and dynamically adjusts the parameters according to the scene type. The noise reduction intensity is increased by 30% in outdoor dim scenes and reduced by 20% in indoor bright scenes. The memory computing module performs 256×256 window root mean square error calculation on the denoised images. The comparator chip judges whether the result is ≤5. If it meets the standard, the image is confirmed to be complete. If it does not meet the standard, the noise reduction calculation is recalculated. S5. Data Analysis and Display Stage: The backend server uses the Spark big data analysis framework to integrate and analyze images, feature data, and environmental parameters in the database, including personnel statistics in the underground area based on personnel feature tags, equipment contour matching detection, and gas concentration trend prediction. The analysis results are synchronously transmitted to the monitoring center's large screen and mobile terminals via Ethernet.

7. A coal mine safety monitoring method based on big data analysis according to claim 6, characterized in that, In S1, the high-definition explosion-proof camera uses infrared night vision supplementary lighting. It automatically turns on when the underground light intensity is <10 lux. The camera is connected to the explosion-proof distribution box through a mining intrinsically safe cable. At the same time, the industrial computer communicates with the camera through an RS485 bus to obtain the camera's working status in real time. In case of abnormality, a local buzzer alarm is triggered.

8. A coal mine safety monitoring method based on big data analysis according to claim 6, characterized in that, In S3, the server cluster adopts a redundant architecture design, and the master-slave switch is realized through the VRRP protocol. The backend database adopts a RAID5 redundant architecture. In the event of a single hard drive failure, data reconstruction is used to ensure that the images are not lost. The database read and write speed is ≥1000MB / s, which meets the storage requirements of 100 frames of images per second.

9. A coal mine safety monitoring method based on big data analysis according to claim 6, characterized in that, In S5, the Spark big data analytics framework is deployed on the Hadoop distributed platform, supporting the processing of more than 1,000 frames of image data per second. When an anomaly is detected in a device, the system will automatically retrieve the device's historical operating images from the past 24 hours, compare the current and historical outlines, generate an anomaly analysis report, and push it to administrators via a mobile app.

10. A coal mine safety monitoring method based on big data analysis according to claim 6, characterized in that, It also includes system self-inspection and maintenance processes: during the daily downtime of underground production, the system automatically checks the lens cleanliness of the cameras in the acquisition module, performs memory redundancy verification on the server cluster of the processing module, and tests the display function of the output devices of the analysis module. The self-inspection results generate a PDF report and send it to the operation and maintenance management platform. At the same time, the system retains self-inspection records for nearly one year for traceability and query.

Images