Methane screen data identification and detection method and system

Abstract

According to the technical scheme, a YOLOV8 algorithm is adopted to detect a single number on a screen, multiple filtering is carried out through NMS non-maximum suppression and IQR quartile distance methods, complete sensor data are combined according to the abscissa sequence of detection frames, and the methane screen data are identified and detected. And a time sequence verification mechanism is established to ensure the reliability of a detection result. The system compares recognized screen data with data uploaded by a sensor in real time, and when a data difference value exceeds a set threshold value or data is abnormal before and after adjustment, images or video evidences are automatically saved, and an alarm is triggered. The problems of low efficiency and insufficient real-time performance of traditional manual inspection are solved, full automation and intelligence of methane sensor monitoring are achieved, 24-hour uninterrupted monitoring service can be provided, illegal operation and equipment abnormity can be found in time, and the safety guarantee level of coal mine underground operation is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coal mine safety monitoring, and in particular to a methane screen data identification and detection method and system. BACKGROUND

[0002] Methane sensors play a crucial role in the safety production of coal mines, directly related to the safety of underground workers and production efficiency. Methane, as one of the most harmful gases in coal mines, its concentration level is the core indicator of the safety of underground working environment. When the concentration of methane exceeds the safety threshold, it is easy to cause an explosion, resulting in significant casualties and property losses. Therefore, accurate monitoring and timely warning of methane sensor data have become an indispensable part of the modern coal mine safety management system, directly affecting the safety production level and economic benefits of the entire coal mine enterprise.

[0003] Traditional methane sensor regulatory inspections mainly rely on manual inspection methods, which have many drawbacks and safety hazards. Manual inspection methods not only have poor reliability and low efficiency, but are also limited by the number of personnel and working hours, making it impossible to achieve real-time monitoring around the clock. In complex underground working environments, inspectors are prone to omissions and misjudgments, especially at night or in poor working conditions, making it difficult to ensure the accuracy and timeliness of manual inspections. In addition, relying solely on manual inspections to check for violations is a huge workload, not only increasing labor costs, but also making it difficult to respond quickly in emergency situations, which may delay the best opportunity for disposal.

[0004] With the rapid development of artificial intelligence technology and computer vision technology, intelligent monitoring methods provide a new technical path to solve the problem of traditional methane sensor monitoring. By introducing an intelligent detection system based on visual analysis, the comprehensiveness, real-time and accuracy of the monitoring work can be significantly improved, effectively making up for the shortcomings of manual inspection. Intelligent monitoring systems not only enable 24-hour uninterrupted monitoring, but also use advanced image recognition algorithms and data analysis techniques to accurately identify methane sensor screen data and compare it with sensor uploaded data in real time, allowing for timely detection of data anomalies and violations. This technological trend represents an important direction for the transformation of coal mine safety monitoring from traditional manual methods to intelligent and digital methods, laying a solid technical foundation for building a modern intelligent mine safety protection system. SUMMARY

[0005] The technical problem to be solved by the present application is the key problems of strong artificial dependence, low supervision efficiency and insufficient real-time in the prior art methane sensor monitoring. The traditional manual inspection method not only increases the labor cost, but also is easy to miss inspection and misjudge in the complex downhole operation environment, which cannot meet the urgent needs of modern coal mine safety production for all-weather real-time monitoring. Especially, the existing supervision means is difficult to find and dispose in time the potential risks of irregular operation and abnormal state of equipment in the methane sensor data uploading process. Therefore, it is urgent to develop a methane sensor monitoring method based on intelligent technology to realize automatic identification of sensor screen data and real-time early warning of abnormal state.

[0006] To solve the above technical problems, the present application provides a methane screen data identification and detection method and system based on visual analysis. The core idea of the technical solution is to deploy an industrial camera in front of the downhole methane sensor screen, collect real-time image data of the sensor screen, use advanced target detection algorithm to accurately identify the numbers displayed on the screen, and compare and analyze the data in real time. The technical solution uses YOLOV8 target detection algorithm to detect single numbers on the methane sensor screen, and uses multiple filtering mechanisms such as NMS non-maximum suppression and IQR interquartile range to ensure the accuracy and reliability of the detection results. The system establishes a data verification mechanism based on time series analysis, which can effectively avoid the influence of single image quality problems on detection accuracy through comprehensive judgment of continuous multiple frame image detection results. When the sensor screen data and the uploaded data have abnormal difference or the data changes abnormally before and after adjustment, the system automatically triggers the alarm mechanism, saves the relevant image or video evidence and uploads it to the monitoring platform, providing reliable basis for subsequent abnormal analysis and disposal.

[0007] A methane screen data identification and detection method, comprising the following steps: Collecting methane sensor screen images through a camera, and detecting single numbers on the methane sensor screen using a target detection algorithm; Using NMS non-maximum suppression method to filter the detection frame for the first time, wherein the NMS non-maximum function is: , wherein, represents the confidence of the detection frame, represents the set of all detection frame confidences, represents the candidate frame set the detection target with the maximum confidence, represents the threshold value of NMS; Using IQR interquartile range method to filter the detection frame for the second time, and calculating the upper and lower bounds of the abnormal value: ,in , and These represent the 25th percentile and the 75th percentile, respectively. Indicates constraint value; The individual numbers are combined into complete sensor screen data according to the horizontal coordinate order of the upper left corner of the detection frame; Create a list of length 7 to store the detection results of consecutive frames. When there are 5 values ​​in the list that are the same as the latest detection result, update the detection result to the methane sensor screen data. The detected screen data is compared with the data uploaded by the sensor. When the screen recognition data exceeds the sensor upload data and the difference exceeds the first threshold, the image is saved and uploaded to the platform to trigger an alarm.

[0008] Furthermore, the target detection algorithm adopts the YOLOv8 algorithm.

[0009] Furthermore, the threshold of NMS Set to 0.2, constraint value Set it to 1.5.

[0010] Furthermore, the above method also includes a calibration state anomaly detection step: setting a calibration state for the detection result of each frame image; determining to enter the calibration state when the calibration state of the current frame is True and the calibration state of the previous frame is False; determining to end the calibration state when the calibration state of the current frame is False and the calibration state of the previous frame is True; comparing the difference between the values ​​before and after calibration; and saving the video and uploading it to the platform to trigger an alarm when the difference exceeds the second threshold.

[0011] Furthermore, the first threshold is set to 0.2, and the second threshold is set to 0.1.

[0012] Furthermore, the above method achieves real-time image acquisition and detection by reading the camera's RTSP address.

[0013] A methane screen data recognition and detection system, comprising: An intrinsically safe camera for underground coal mines is installed directly over a methane sensor screen and is dustproof. A hard disk recorder, connected to a camera, is used to store historical playback videos from the camera; An edge processing device, equipped with an AI chip, is connected to a camera and a hard disk recorder for performing the methane screen data recognition and detection method according to any one of claims 1 to 6.

[0014] Furthermore, the edge processing device is an edge server, which has model inference capabilities.

[0015] Further, the system further comprises an AI application platform, configured to receive the alarm information and the picture or video evidence.

[0016] The beneficial effects of the present application are as follows: first, the scheme realizes the full automation and intelligentization of methane sensor monitoring, completely changes the traditional supervision mode relying on manual inspection, and significantly improves the efficiency and accuracy of monitoring work. The system can provide 24-hour uninterrupted real-time monitoring service, timely discovers and warns various abnormal conditions, and effectively safeguards the safety of underground operation personnel. Secondly, through multiple algorithm filtering and time series verification mechanism, the system has high detection accuracy and stability, can adapt to complex underground operation environment, and effectively cope with the challenges such as camera position change, light condition change and other practical applications. In addition, the scheme has good scalability and adaptability, which is not only suitable for methane sensor monitoring, but also can be popularized and applied to other types of underground equipment monitoring scenes, providing important technical support for building a comprehensive intelligent mine safety monitoring system. The successful application of the technical scheme will strongly promote the digitalization and intelligentization transformation of coal mine safety production, and create favorable conditions for the coordinated development of safety production and economic benefits of coal mine enterprises. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only a part of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0018] Fig. 1 It is a detection method flowchart of the present application. Fig. 2 It is a schematic diagram of the sensor at different positions of the screen. Fig. 3 It is a complete reasoning image schematic diagram of the present application. DETAILED DESCRIPTION

[0019] The present application will be described in detail below in combination with the drawings and specific embodiments. It should be noted here that in order to make the embodiments more detailed, the following embodiments are the best, preferred embodiments, and other alternative ways can also be used by those skilled in the art to implement; and the drawings are only used to describe the embodiments more specifically, and are not intended to specifically limit the present application.

[0020] It is noted that the use of "one embodiment," "an embodiment," "certain embodiments," "some embodiments," etc., in the specification, indicates that the feature, structure, or characteristic being discussed can be included in one or more embodiments of the disclosure, but is not required to be included in every embodiment. Further, the appearance of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily referring to a single, special embodiment.

[0021] In general, the terminology or phraseology employed herein, and not specifically defined herein above, is for the purpose of description only and the level of generality is not intended to limit the scope of the present disclosure. As used herein, the terms "one or more" can be used to describe any feature, structure, or characteristic in the singular or can be used to describe combinations of features, structures, or characteristics, in the plural, depending on the context in which the term is used. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, and instead can allow for existence of additional factors not necessarily expressly described, again, depending on the context in which the term is used.

[0022] Referring to Figs. 1 to 3 shown The present application provides a kind of methane screen data identification detection method and system, by deploying industrial camera in the key position before downhole methane sensor screen, the screen data of methane sensor is collected, establishes methane sensor screen image data set, realizes the identification of methane sensor screen data in image by training target detection model.This method combines advanced computer vision technology and intelligent monitoring system, can realize all-weather uninterrupted methane sensor data monitoring, effectively solve the inefficiency and omission risk problem existing in traditional manual inspection mode.

[0023] The core technical route of the method includes the following key steps: first, read the RTSP address of the camera, detect the methane sensor screen on the image in real time, save the picture upload platform when the sensor screen data exceeds the threshold of sensor upload data and alarm; when the data before and after the adjustment state is inconsistent, save the video upload platform and alarm. The entire detection process ensures the accuracy and reliability of the detection results through multiple filtering mechanisms.

[0024] Step one: detect sensor screen single value The target detection method is used to detect the methane sensor screen data, and YOLOV8 is preferably used as the detection method for the methane sensor screen data. The YOLOV8 algorithm has the characteristics of high detection accuracy and fast processing speed, and can meet the demand of real-time monitoring. It should be noted that no matter which target detection algorithm is used to detect the methane sensor screen data, it will not affect the essential content of the scheme. This step trains a deep learning model to enable the system to automatically recognize and locate each digital character displayed on the methane sensor screen.

[0025] Step two: filter the detection box using NMS non-maximum suppression During the detection of individual numbers on the methane sensor screen, multiple detection results may occur, so NMS non-maximum suppression method is needed for filtering. The NMS non-maximum function expression is: Among them, represents the confidence of the detection box, represents the set of all detection box confidences, represents the candidate box set the detection target with the maximum confidence, represents the threshold of NMS, which is usually set to 0.2. When the IOU exceeds the threshold the confidence of the detection box will be directly set to 0, effectively removing the duplicate detection box and ensuring that only one optimal detection result is retained for each digital character.

[0026] Step three: filter outlier detection boxes After the first filtering of the methane sensor screen single data by NMS, considering that the numbers outside the sensor screen may also be detected, the method of filtering outlier detection boxes is used to further filter non-screen data. Due to the frequent changes in the position of the camera in the downhole working environment, using the method of filtering outlier detection boxes instead of the ROI region method can effectively avoid the need to re-determine the ROI region after the position of the camera changes.

[0027] In order to better filter the abnormal data of outliers, the embodiment adopts the method of IQR. First, sort the detection target data from small to large into 4 equal parts, and the IQR calculation formula is: Among them and represent the 25th and 75th percentiles, respectively, so IQR represents the range of the middle 50% data. According to IQR, the upper and lower bounds of outliers can be calculated: where LB, UB represent lower bound and upper bound respectively, represents the constraint value, which is generally set to 1.5. After filtering the abnormal outlier target value by IQR, the detection target can be accurately screened, and the problem of using ROI region to limit the detection target affected by the change of camera angle can be effectively solved, improving the adaptability and stability of the system.

[0028] Step four: merge single numbers into sensor screen data After multiple filtering of the single numbers on the methane sensor screen by the above steps, it is necessary to determine the order of each detection target to combine into the final detection result. In this embodiment, the horizontal coordinate of the top-left corner point of each detection frame is used to determine the order, and the recognized single number characters are combined in order from left to right, forming a complete methane sensor numerical reading. This sorting method is simple and effective, and can accurately reflect the actual arrangement order of the numbers on the screen.

[0029] Step five: historical data to determine whether the detection result is reliable This embodiment introduces time series analysis, not just detecting a single frame of image. By comprehensively judging the detection results of consecutive multiple frames of images, it avoids the influence of a few fluctuations in the detection process on the actual detection effect. The specific implementation method is to establish a list of length 7 to store the detection results of each frame of image. When there are 5 values in the list that are the same as the latest detection result, update the latest detection result as the methane sensor screen data. This multi-frame fusion method can effectively improve the stability and reliability of the detection result, and reduce the false detection caused by the quality problem of a single frame of image.

[0030] Step six: real-time inference image The system can realize the complete real-time image inference process on site, and has been deployed online in Tangkou Coal Mine to monitor the methane sensor screen data in real time. The real-time inference process includes image acquisition, preprocessing, target detection, post-processing and result output, etc. The whole process is completed on the edge processing device to ensure the response speed and processing efficiency. The system can trigger the corresponding alarm mechanism immediately when an abnormal situation is detected.

[0031] Step seven: abnormal processing The abnormality processing mechanism includes two main cases. The first case is that the sensor screen recognition data is higher than the sensor uploaded data, and exceeds the threshold value (generally set to 0.2), it is considered that the sensor uploaded data is inaccurate or is adjusted abnormally low, so the picture evidence is saved and uploaded to the application platform for alarm. The second case is that the numerical difference before and after the sensor adjustment exceeds the threshold value (generally set to 0.1), by setting an adjustment state for the detection result of each frame of picture, when the current frame picture adjustment state is True and the last frame picture adjustment state is False, it represents that it just enters the adjustment state; when the current frame picture adjustment state is False and the last frame picture adjustment state is True, it represents that it ends the adjustment state; compare the numerical difference before and after the adjustment, if it exceeds the threshold value, save the video evidence and upload it to the application platform.

[0032] System hardware configuration The hardware equipment system of the present application comprises three core components. The first part is a coal mine underground intrinsically safe camera, which needs to be installed opposite the methane sensor screen. At the same time, in order to prevent dust interference from affecting the imaging quality of the picture, the camera needs to be dustproof to ensure that clear and stable images can be obtained in the harsh underground environment. The second part is a hard disk recorder, which is used to store historical playback videos of the camera, providing convenience for data collection and post-analysis, and at the same time serving as an important device for evidence preservation. The third part is an edge processing device, which is an edge server with AI chips, used for model inference function, capable of real-time processing of video streams and executing methane sensor data recognition algorithms.

[0033] Example 1: Tangkou Coal Deployment Case, refer to Fig. 2 and Fig. 3 In the actual deployment of Tangkou Coal, the system has been successfully applied to the underground methane sensor monitoring scene. By installing intrinsically safe cameras in key work areas and aiming at the methane sensor screen, 24-hour uninterrupted monitoring is realized. The system successfully detects multiple sensor data abnormality cases during operation, including screen display value and uploaded data inconsistency, abnormal data change before and after adjustment, etc., timely warning and saving relevant evidence, effectively ensuring underground work safety.

[0034] Example 2: Abnormality detection and processing flow In a monitoring process, the system detected that the methane sensor screen displayed a concentration of 0.8%, while the sensor uploaded data to the monitoring system was 0.5%, with a difference of 0.3%, exceeding the set threshold value of 0.2%. The system immediately triggered the first type of abnormality alarm, automatically saved the current image as evidence, and uploaded the alarm information and image evidence to the AI application platform. After receiving the alarm, the operation and maintenance personnel timely rushed to the scene for inspection, and found that the sensor indeed had data transmission abnormality problem, avoiding potential safety risks.

[0035] The application realizes the automatic monitoring and abnormal early warning of the methane sensor data by integrating advanced computer vision technology and intelligent monitoring system, significantly improves the intelligent level of coal mine safety production, and provides reliable guarantee for the safety of underground personnel.

[0036] The present application encompasses any substitutions, modifications, equivalent methods and solutions made to the essence and scope of the present application. In order for the public to have a thorough understanding of the present application, specific details are described in the following preferred embodiments of the present application, and the present application can also be fully understood without the description of these details to those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, procedures, elements and circuits are not described in detail.

[0037] The above is only the preferred embodiment of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can also be made, which should be considered as the protection scope of the present application.

Claims

1. A method for identifying and detecting methane screen data, characterized in that, Includes the following steps: The image of the methane sensor screen is captured by a camera, and a target detection algorithm is used to detect individual numbers on the methane sensor screen. The detection box is first filtered using the Non-Maximum Suppression (NMS) method, where the NMS function is: ,in, This indicates the confidence level of the detection box. This represents the set of confidence scores for all bounding boxes. Represents the set of candidate boxes The detection target with the highest confidence level. The threshold representing NMS; The detection box is filtered a second time using the IQR interquartile range method, and the upper and lower bounds of outliers are calculated: in , and These represent the 25th percentile and the 75th percentile, respectively. Indicates constraint value; The individual numbers are combined into complete sensor screen data according to the horizontal coordinate order of the upper left corner of the detection frame; Create a list of length 7 to store the detection results of consecutive frames. When there are 5 values ​​in the list that are the same as the latest detection result, update the detection result to the methane sensor screen data. The detected screen data is compared with the data uploaded by the sensor. When the screen recognition data exceeds the sensor upload data and the difference exceeds the first threshold, the image is saved and uploaded to the platform to trigger an alarm.

2. The methane screen data recognition and detection method according to claim 1, characterized in that, The target detection algorithm used is the YOLOv8 algorithm.

3. The methane screen data recognition and detection method according to claim 1, characterized in that, The threshold of the NMS The constraint value is set to 0.

2. Set it to 1.

5.

4. The methane screen data recognition and detection method according to claim 1, characterized in that, It also includes an abnormal calibration state detection step: set the calibration state for the detection result of each frame image. When the calibration state of the current frame is True and the calibration state of the previous frame is False, it is determined to enter the calibration state. When the calibration state of the current frame is False and the calibration state of the previous frame is True, it is determined to end the calibration state. Compare the difference between the values ​​before and after calibration. When the difference exceeds the second threshold, save the video and upload it to the platform to alarm.

5. The methane screen data recognition and detection method according to claim 1 or 4, characterized in that, The first threshold is set to 0.2, and the second threshold is set to 0.

1.

6. The methane screen data recognition and detection method according to claim 1, characterized in that, Real-time image acquisition and detection are achieved by reading the camera's RTSP address.

7. A methane screen data recognition and detection system, characterized in that, include: An intrinsically safe camera for underground coal mines is installed directly over a methane sensor screen and is dustproof. A hard disk recorder, connected to the camera, is used to store historical playback videos of the camera; An edge processing device, equipped with an AI chip, is connected to the camera and hard disk recorder to perform the methane screen data recognition and detection method according to any one of claims 1 to 6.

8. The methane screen data recognition and detection system according to claim 7, characterized in that, The edge processing device is an edge server with model inference capabilities.

9. The methane screen data recognition and detection system according to claim 7, characterized in that, The system also includes an AI application platform for receiving alarm information and image or video evidence.

Images