Method and system for identifying periodic roof pressure at working face based on YOLO algorithm

Abstract

The present invention belongs to the technical field of coal mine mining, and discloses a method and system for identifying periodic roof pressure of a working face based on the YOLO algorithm. The present invention maps data into three-dimensional space to generate a 3D stress cloud map of the working face of the mining field, and generates a uniformly distributed and visualized three-dimensional grid stress distribution cloud map of the mining field; identifies high stress, low stress and pressure areas with corresponding labels; uses the YOLO algorithm to perform real-time image recognition of high and low stress areas, and accurately locates the roof pressure area; through image edge detection technology, extracts the time range of the roof pressure area and its corresponding hydraulic support distribution range, and realizes intelligent analysis of regional pressure changes. By combining image recognition algorithms such as YOLO, the present invention enables coal mines to no longer rely on traditional data processing methods, which not only reduces potential errors caused by manual intervention, but also greatly improves the reliability and real-time performance of data processing, thereby improving the efficiency and accuracy of safety monitoring.

Description

Technical Field

[0001] The present invention belongs to the technical field of coal mine mining, and in particular to a method and system for identifying periodic roof pressure on a working face based on a YOLO algorithm. Background Art

[0002] Periodic pressure is a phenomenon in which the basic top rock layer above the goaf periodically collapses or breaks due to loss of support during the mining process of the coal face, thereby exerting periodic pressure on the support system. Periodic pressure is an important feature of mine pressure and has an important impact on the safe production and support design of coal mining.

[0003] The direct roof collapse during the advancement of the coal mine fully-mechanized mining face causes the basic roof to be suspended. As the working face is further advanced, the length of the basic roof suspension reaches the limit value, and collapse will occur, causing periodic pressure. Figure 2 shown. are the loads on rock blocks M and N respectively, is the pressure exerted by the basic top on the rock block M, is the support force on rock block N, is the rotation angle of rock block M, are the shear forces exerted by rock block M at points A and C, respectively. is the cut length of rock block M, are the basic top thickness and the direct top thickness, To increase the height of the working surface, is the sinking amount of rock block N, see attached Figure 3 .

[0004] Factors affecting the compressive strength: The sinking amount and mass of the rock block M jointly determine the magnitude of the impact force generated during the rock collapse: as the rock sinks, the process of converting the potential energy of the rock into kinetic energy becomes more and more intense, resulting in greater impact force; the larger the mass of the rock, the more kinetic energy it will accumulate during the sliding process, causing a greater impact on the structure below, and the higher the compressive strength that the support needs to withstand. When the rock mass increases, the amount of its subsidence The corresponding decrease leads to a decrease in the pressure strength; when the working face is high When the value increases, the amount of rock sinking increases accordingly, resulting in an increase in the compressive strength. In addition, since the quality of the rock block mainly depends on the lithology and basic roof thickness, the main influencing factors of the compressive strength are the direct roof thickness, basic roof thickness, mining height, and roof lithology.

[0005] Factors affecting the pressure step distance: When the rock block M breaks, the cut-off length determines the stress of the rock block in the suspended state. According to the cantilever beam calculation model, the cut-off length of the rock block M is:

[0006] ;

[0007] When the tensile strength of the basic top rock remains unchanged, the basic top thickness The larger it is, the larger the force-bearing area of ​​the basic top is, and it can withstand greater vertical stress. As a result, the length of the basic top cut-off is longer, the deadweight of the rock block is greater, the ability of the rock block to resist instability is enhanced, and the pressure step distance is correspondingly increased.

[0008] Through the above analysis, it can be concluded that the main factors affecting the pressure include basic roof thickness, direct roof thickness, mining height, coal seam inclination and roof lithology. The thickness of the coal seam and the inclined length of the working face will affect the spatial structure after mining, and the advancement speed determines to a certain extent whether the self-stable structure can be quickly achieved after the basic roof is broken. Therefore, the thickness of the coal seam, the inclined length of the working face and the advancement speed are also factors affecting the intensity of the pressure.

[0009] In recent years, the technology of mine pressure monitoring and prediction in fully mechanized mining working faces has made significant progress, realizing the real-time display, transmission and analysis of underground monitoring data. However, the basic principle of mine pressure observation and prediction still relies on traditional station observation, which can only monitor the characteristics of partial mine pressure manifestation in the working face, and cannot obtain information such as the support pressure resistance of the entire working face. Secondly, due to the long interval of data acquisition frequency, it is impossible to accurately obtain the pressure changes in a short time such as support frame shifting and safety valve opening. In addition, various mine pressure monitoring systems can only perform a relatively simple analysis of the data monitored on site, and in-depth analysis and mining are still manual methods, lacking effective work cycle identification technology, and work cycle identification is the basis of all subsequent work; in addition, the mine pressure analysis indicators are single, and the support-surrounding rock related information contained in a large amount of mine pressure data cannot be effectively mined, and it is even more impossible to effectively realize the prediction of incoming pressure, roof fall warning, and monitoring and maintenance of the operation status of hydraulic supports.

[0010] At present, the following problems still exist in the research of mine pressure monitoring and analysis, roof pressure prediction and roof fall warning in fully mechanized mining working faces:

[0011] 1. Mine pressure monitoring and analysis cannot cover the entire working face. Mine pressure monitoring and analysis are still based on the traditional station / line concept, and do not fully consider the distribution characteristics and transfer rules of the load between the support groups; that is, a small number of stations and lines are arranged at different positions of the working face. Generally, the working face is divided into three areas: upper, middle and lower. The mine pressure prediction model is constructed based on the area, and the change in the working cycle resistance of the center position support is used to characterize the mine pressure in the area. This method is a point monitoring method and cannot monitor the impact of changes in local production geological conditions on the mine pressure manifestation and roof stability. All-round monitoring is the core idea of ​​modern sensor network technology. Therefore, it is impossible to accurately and comprehensively realize roof pressure prediction and roof fall warning based on traditional station-based mine pressure monitoring.

[0012] 2. The existing hydraulic support working resistance curve identification algorithm is too simple. The working cycle refers to the analysis of the original pressure data of the support through the algorithm to determine the actual initial support and unloading and moving time of the hydraulic support. The working cycle identification is the basis and key to all subsequent mine pressure analysis and prediction work, and the accuracy of the identification also determines the accuracy of mine pressure analysis and prediction. However, the current identification algorithm cannot accurately identify it. Because all the algorithm cores are concentrated on the "slope method" to identify the end-of-cycle resistance.

[0013] 3. Data analysis and processing capabilities are still insufficient. In recent years, the mine pressure monitoring and forecasting technology of fully mechanized mining working faces has made significant progress, realizing the real-time display and transmission of underground monitoring data. However, the integration between mine pressure monitoring and data analysis still has certain deficiencies. See attached Figure 4 Working resistance curve of hydraulic support.

[0014] (1) Identification of hydraulic support working resistance cycle. First, based on the working resistance curve data of the hydraulic support, identify the end-of-cycle resistance and initial support force, and distinguish each coal mining cycle.

[0015] (2) Calculate the weighted working resistance. For each coal mining cycle, multiply its working resistance value by the time of the cycle to obtain the weighted working resistance of the cycle.

[0016] (3) Calculate the mean square error. Calculate the average value of the weighted working resistance of all coal mining cycles to obtain the average working resistance of the working face. At the same time, calculate the mean square error of the weighted working resistance to evaluate the degree of dispersion of the data.

[0017] (4) Result analysis. The average working resistance and mean square error are used as the basis for judging the roof pressure. If the average working resistance exceeds the expected range or the mean square error is large, it may indicate that the roof pressure is abnormal and corresponding measures need to be taken.

[0018] At present, the analysis and processing capabilities of mine pressure monitoring data are still insufficient, and real-time monitoring and effective analysis cannot be achieved. This leads to a lack of sufficient prediction and response strategies when dealing with changes in mine pressure. Existing methods mainly rely on the analysis of the working resistance curve of hydraulic supports. Although some basic indicators such as initial support force and final resistance can be identified, they fail to meet the requirements in terms of dynamic monitoring and data processing. This limitation makes the monitoring of mine pressure changes lack of timeliness and cannot provide substantial support for on-site decision-making, thus affecting the effectiveness of safety management and risk control. Summary of the invention

[0019] In order to overcome the problems existing in the related technology, the disclosed embodiments of the present invention provide a method and system for identifying periodic roof pressure on a working face based on the YOLO algorithm, and specifically relate to a technology for identifying periodic roof pressure on a coal mining working face using image recognition.

[0020] The technical solution is as follows: a method for identifying periodic pressure on the roof of a working face based on a YOLO algorithm, comprising the following steps:

[0021] S1, data fusion and spatial modeling; based on the real-time collected hydraulic support pressure monitoring data, the discrete monitoring data of different supports or sensors are mapped into a spatial coordinate system to form a three-dimensional grid; the longitudinal and transverse stress differences are calculated by Kriging interpolation, and the discrete monitoring data are mapped into three-dimensional space to generate a uniformly distributed and visualized three-dimensional grid stress distribution cloud map of the mining area;

[0022] S2, image preprocessing: using computer vision recognition technology to analyze and identify different stress areas, using the annotation tool Roboflow to manually annotate the three-dimensional grid pressure distribution cloud map of the mining area, and marking the high stress area, low stress area and pressure area of ​​the roof with corresponding labels;

[0023] S3, target detection and area recognition: According to the currently generated 3D stress cloud map, the high stress area is analyzed through image recognition and area division, and the high and low stress areas in the 3D stress cloud map are automatically identified by the target detection algorithm, and the top plate pressure area is accurately located;

[0024] S4, determination and prediction of periodic pressure; after the image recognition algorithm identifies the high stress area, the high stress area is deeply analyzed to determine whether it has the characteristics of periodic pressure; through the image edge detection technology, the time range of the roof pressure area and its corresponding hydraulic support distribution are extracted, and according to the set stress gradient or absolute stress threshold, the high stress area is extracted and the area, volume and boundary characteristics of the area are calculated; the difference in ground stress between the high stress area and the surrounding area is analyzed to determine whether the criterion for roof pressure is met; combined with the periodic prediction model and historical data, the pressure changes in the roof pressure area are intelligently analyzed.

[0025] In step S1, based on the real-time collected hydraulic support pressure monitoring data, it includes:

[0026] Through pressure sensors or electro-hydraulic control systems, all hydraulic supports on the working face are monitored after data collection and preprocessing; data collection and preprocessing include: missing value processing, outlier detection, duplicate value and redundant data processing, verification of cleaning and correction, and correction of pressure sensor errors.

[0027] In step S1, the longitudinal and transverse stress differences are calculated by Kriging interpolation, including the following steps:

[0028] (2.1) Construction of variation function: Divide the distance into several distance intervals and count the number of point pairs in each distance interval , calculate the square of the pressure difference between the points , take the average of all point pairs and divide by 2 to get ; Variogram The expression is:

[0029] ;

[0030] In the formula, Indicates the distance The value of the variation function at time Indicates the distance The point logarithm of For spatial location The variable value at For spatial location The variable value at is the hydraulic support pressure, is the distance between two points;

[0031] (2.2) Vertical and horizontal interpolation;

[0032] In the longitudinal direction, data interpolation between the time dimension and stress value is performed for a single stent to fill in the missing data between time points with large sampling intervals;

[0033] In the lateral direction, interpolation is performed between adjacent supports to estimate the spatial distribution of stress;

[0034] Through longitudinal and transverse interpolation, the data is mapped into three-dimensional space to generate a uniformly distributed and visualized three-dimensional grid stress distribution cloud map.

[0035] In step S2, the high stress area, low stress area and pressure area of ​​the top plate are marked with corresponding labels, including: one-click export to COCO, VOC and YOLO formats, and automatic conversion to txt format.

[0036] In step S3, high stress areas are analyzed by image recognition and area segmentation, including:

[0037] The YOLO algorithm performs real-time image recognition of high and low stress areas, extracts significant features in the stress cloud image through low-level characteristics of color, texture and shape, captures local high stress areas, and forms a set of candidate boxes; the extracted candidate box features are input into the classifier to classify the local features and identify high stress and periodic stress areas.

[0038] Furthermore, the YOLO algorithm performs real-time image recognition of high and low stress areas, including:

[0039] S301, image feature extraction; use SIFT, Haar and HOG algorithms to extract color, texture and shape feature areas in stress cloud images; SIFT algorithm detects key points with scale and rotation invariance in the image, extracts local features of extreme points and texture changes, and initially forms candidate frames of significant areas; Haar features capture the geometric structure in the image based on the brightness contrast of rectangular areas and screen the sliding window area; HOG algorithm captures complex texture and edge features by calculating the distribution characteristics of local gradient directions, optimizes boundaries and refines the range of candidate frames; by merging the candidate frames generated by SIFT, Haar and HOG, redundancy is eliminated by using feature overlap and significance scores, and only the optimal frame covering the high stress area is retained;

[0040] S302, classifier identification: input the extracted candidate frame features into the classifier, classify the color, extreme point, texture, geometric structure, and edge features, and identify the high stress and periodic pressure areas;

[0041] S303, redundant candidate box optimization;

[0042] S304, manual evaluation.

[0043] In step S302, redundant candidate frames are optimized, including: each candidate frame is assigned a confidence score during the detection process, and the candidate frames are sorted from high to low according to their confidence scores, and the candidate frames with the highest confidence are used as priority processing targets; for the sorted candidate frames, the overlap between the candidate frames used as priority processing targets and other candidate frames is calculated one by one, and the intersection and union ratio is used to determine whether the candidate frames are too similar; redundant candidate frames are deleted until all candidate frames are processed and only frames with the highest confidence and non-overlapping are left; specifically, the non-maximum suppression (NMS) algorithm is used to process the detection results, redundant candidate frames are removed, and only the frames with the highest confidence are retained to locate the optimal position of the top plate pressure area; for the screened candidate frames, the position is optimized according to the centroid coordinates, boundary points and size of the frames to improve the boundary accuracy, and the candidate frame positions and filling effects are superimposed on the original stress cloud map to form a complete distribution map containing the optimized high stress area, and the range and characteristics of the pressure area are clarified;

[0044] In step S204, manual evaluation includes:

[0045] For the automatically identified area, the characteristics of the identified area are manually analyzed, including area, volume and boundary characteristics. By comparing the ground stress values ​​inside and outside the high stress area, the stress gradient distribution is calculated. The expression is:

[0046] ;

[0047] In the formula, are the stress values ​​in the high stress area and the surrounding area, respectively. for distance;

[0048] According to the set stress difference threshold or gradient threshold, determine whether the high stress area reaches a dangerous level. If the conditions are met, mark the area as a potential pressure area, otherwise it will be eliminated. According to the calculation results, evaluate whether the stress difference between the high stress area and the surrounding area reaches the set top plate pressure criterion. The areas that meet the criteria are retained, and those that do not meet the criteria are eliminated, and further check whether there are unidentified pressure areas in other areas.

[0049] In step S4, the pressure change in the roof pressure area is intelligently analyzed by combining the periodic prediction model and historical data, including:

[0050] Link the MySQL database MySQL Connector package in the PyCharm installation to connect data monitoring with image recognition;

[0051] Create a monitoring_data database and table in MySQL to store monitoring data, use the OpenCV visual library for image recognition, and insert the results into the database; write the collected pressure value and timestamp data of each sensor to the defined sensor data table in a timely manner; perform stress field modeling based on the stored sensor data and generate a 3D stress cloud map; store the generated 3D stress cloud map data in the corresponding database table, and record the generation time and version information; use the YOLO model to detect the converted image and identify potential problem areas or abnormal stress points; record each detection result, including the identified area, confidence score, and detection time;

[0052] For the collection of mine pressure detection data, the electro-hydraulic control hardware and its system trigger the collection function. When the hydraulic support moves, the pressure value increases or decreases; when the stress value change exceeds a threshold set inside the system, the electro-hydraulic control system of the hydraulic support begins to continuously record the stress changes in this time period until the stress returns to stability and then returns to the normal collection frequency.

[0053] Another object of the present invention is to provide a system for identifying periodic pressure coming from the roof of a working face based on the YOLO algorithm, the system implementing the method for identifying periodic pressure coming from the roof of a working face based on the YOLO algorithm, the system comprising:

[0054] The mining field three-dimensional grid stress distribution cloud map generation module maps the discrete monitoring data of different supports or sensors into a spatial coordinate system based on the real-time collected hydraulic support pressure monitoring data to form a three-dimensional grid; the longitudinal and transverse stress differences are calculated by Kriging interpolation, and the discrete monitoring data are mapped into three-dimensional space to generate a uniformly distributed and visualized mining field three-dimensional grid stress distribution cloud map;

[0055] The identification module is used to analyze and identify different stress areas using computer vision recognition technology, and use the annotation tool Roboflow to manually annotate the three-dimensional grid pressure distribution cloud map of the stope, and identify the high stress area, low stress area and pressure area of ​​the roof with corresponding labels;

[0056] The pressure area positioning module is used to analyze the high stress area through image recognition and area division based on the currently generated 3D stress cloud map, and automatically identify the high and low stress areas in the 3D stress cloud map using the target detection algorithm to accurately locate the top plate pressure area;

[0057] The image edge detection module is used to identify the high stress area through the image recognition algorithm, and then conduct an in-depth analysis of the high stress area to determine whether it has the characteristics of periodic pressure; through the image edge detection technology, the time range of the roof pressure area and its corresponding hydraulic support distribution are extracted, and according to the set stress gradient or absolute stress threshold, the high stress area is extracted and the area, volume and boundary characteristics of the area are calculated; the difference in ground stress between the high stress area and the surrounding area is analyzed to determine whether the criterion for roof pressure is met; combined with the periodic prediction model and historical data, the pressure changes in the roof pressure area are intelligently analyzed.

[0058] Combined with all the above technical solutions, the beneficial effects of the present invention are as follows: In order to solve the problems of low efficiency, poor accuracy and inaccurate positioning of periodic pressure identification on the current working face, the present invention firstly uses the Kriging interpolation method to calculate the longitudinal and transverse stress differences based on the pressure monitoring data of the hydraulic support, maps the data into three-dimensional space to generate a 3D stress cloud map of the working face of the mining area, and generates a uniformly distributed and visualized three-dimensional grid stress distribution cloud map of the mining area. Secondly, using computer vision recognition technology, the annotation tool Roboflow is used to manually annotate the pressure distribution map, and the high stress, low stress and pressure areas are marked with corresponding labels. Furthermore, the YOLO algorithm is used to perform real-time image recognition of high and low stress areas, and accurately locate the roof pressure area; finally, through the image edge detection technology, the time range of the roof pressure area and its corresponding hydraulic support distribution range are further extracted to realize the intelligent analysis of regional pressure changes. By combining image recognition algorithms such as YOLO, coal mines no longer rely on traditional data processing methods, such as manual analysis, calculation and judgment after exporting data sets, thereby improving the efficiency and accuracy of safety monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings herein are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description, serve to explain the principles of the present disclosure;

[0060] Figure 1 It is a flow chart of a method for identifying periodic pressure coming from the roof of a working face based on the YOLO algorithm provided by an embodiment of the present invention;

[0061] Figure 2 It is the principle diagram of the "slope method" of the end-of-cycle resistance in the prior art;

[0062] Figure 3 It is a schematic diagram of the movement of overburden in a stope of the prior art;

[0063] Figure 4 It is a working resistance curve diagram of the hydraulic support in the prior art;

[0064] Figure 5 It is a three-dimensional grid stress distribution cloud map of a stope provided by an embodiment of the present invention;

[0065] Figure 6 A diagram of a periodic pressure region identified by a YOLO algorithm provided in an embodiment of the present invention;

[0066] Figure 7 is a schematic diagram of a system for identifying periodic roof pressure on a working face based on a YOLO algorithm provided by an embodiment of the present invention;

[0067] In the figure: 1. Module for generating three-dimensional grid stress distribution cloud map of mining area; 2. Identification module; 3. Module for locating pressure area; 4. Module for detecting image edge. DETAILED DESCRIPTION

[0068] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific implementation disclosed below.

[0069] In recent years, my country's comprehensive mining face mine pressure monitoring and prediction technology has made significant progress, realizing the real-time display, transmission and analysis of underground monitoring data. However, the basic principle of mine pressure observation and prediction still relies on traditional station observation, which can only monitor the partial mine pressure characteristics of the working face, and cannot obtain information such as the support pressure resistance of the entire working face. Secondly, due to the long interval between data acquisition frequencies, it is impossible to accurately obtain the pressure changes in a short time such as support frame shifting and safety valve opening. In addition, various mine pressure monitoring systems can only perform a relatively single analysis of the data monitored on site, and in-depth analysis and mining are still manual methods, lacking effective work cycle identification technology, and work cycle identification is the basis of all subsequent work; in addition, the mine pressure analysis index is single, and the support-surrounding rock related information contained in a large amount of mine pressure data cannot be effectively mined, and it is even more impossible to effectively realize the prediction of incoming pressure and the early warning of roof collapse. The innovation of the present invention is to break the limitation of the traditional mine pressure movement law analysis relying on manual methods, and overcome the problem of insufficient effective work cycle identification technology. The existing mine pressure analysis method was established in response to the backward mine pressure observation methods such as manual or mechanical methods in the last century. The mine pressure analysis only includes simple indicators such as initial support force, final resistance, time-weighted working resistance, etc., and cannot deeply explore the support and surrounding rock related information contained in a large amount of mine pressure data, resulting in the inability to effectively determine and predict the incoming pressure. At present, by adopting real-time image recognition, supplemented by manual participation, real-time analysis of monitoring data to determine the stress area, it provides technical support for the professional analysis of quickly grasping the law of roof movement.

[0070] Embodiment 1, as Figure 1 As shown, the method for identifying periodic roof pressure on a working face based on the YOLO algorithm provided by an embodiment of the present invention includes:

[0071] S1, data fusion and spatial modeling; based on the real-time collected hydraulic support pressure monitoring data, the discrete monitoring data of different supports or sensors are mapped into a spatial coordinate system to form a three-dimensional grid; the longitudinal and transverse stress differences are calculated by Kriging interpolation, and the discrete monitoring data are mapped into three-dimensional space to generate a uniformly distributed and visualized three-dimensional grid stress distribution cloud map of the mining area;

[0072] S2, image preprocessing: using computer vision recognition technology to analyze and identify different stress areas, using the annotation tool Roboflow to manually annotate the three-dimensional grid pressure distribution cloud map of the mining area, and marking the high stress area, low stress area and pressure area of ​​the roof with corresponding labels;

[0073] S3, target detection and area recognition: According to the currently generated 3D stress cloud map, the high stress area is analyzed through image recognition and area division, and the high and low stress areas in the 3D stress cloud map are automatically identified by the target detection algorithm, and the top plate pressure area is accurately located;

[0074] S4, determination and prediction of periodic pressure; after the image recognition algorithm identifies the high stress area, the high stress area is deeply analyzed to determine whether it has the characteristics of periodic pressure; through the image edge detection technology, the time range of the roof pressure area and its corresponding hydraulic support distribution are extracted, and according to the set stress gradient or absolute stress threshold, the high stress area is extracted and the area, volume and boundary characteristics of the area are calculated; the difference in ground stress between the high stress area and the surrounding area is analyzed to determine whether the criterion for roof pressure is met; combined with the periodic prediction model and historical data, the pressure changes in the roof pressure area are intelligently analyzed.

[0075] After the image recognition algorithm identifies the high stress areas, these areas can be analyzed in depth to determine whether they have the characteristics of periodic pressure. Through image edge detection technology, the time range of the roof pressure area and its corresponding hydraulic support distribution are extracted. According to the set threshold (such as stress gradient or absolute stress value), the high stress area is extracted, and the area, volume and boundary characteristics of the area are calculated. At the same time, the difference in ground stress between the high stress area and the surrounding area is analyzed to determine whether the criterion for roof pressure is met. Combined with the periodic prediction model and historical data, the pressure changes in the roof pressure area are analyzed intelligently.

[0076] In the mine pressure analysis of the present invention, the data of the hydraulic support is usually discrete, that is, the data of the support number and the time point are limited measurement values. Through the Kriging Interpolation method, using spatial correlation and optimal linear unbiased estimation, interpolation can be performed in the two dimensions of support number and time to generate a continuous stress distribution cloud map, which can not only make up for the limitations of data collection, but also provide an important basis for analysis and decision-making.

[0077] For the collection of mine pressure detection data, the electro-hydraulic control hardware and its system should realize the trigger collection function. When the hydraulic support is moved, the entire operation process time is directly recorded, including the action of lowering or raising the frame, without the need to collect cycle node data separately. The key time point and all relevant data before and after it can be directly captured during the operation, so as to monitor the mine pressure changes more accurately and comprehensively.

[0078] The YOLO algorithm performs real-time image recognition of high and low stress areas, extracts significant features in the stress cloud image through low-level features such as color, texture, and shape, captures local high stress areas, and forms a set of candidate frames. The extracted candidate frame features are input into the classifier to classify local features and identify high stress and periodic stress areas.

[0079] The Canny edge detection algorithm is used to further analyze the high stress area, extract its boundary shape and time range, identify the regional characteristics of sudden pressure rise, and lock the potential pressure area. The monitoring results are compared with the actual mine pressure manifestation, and the Kriging interpolation and YOLO algorithm model parameters are adjusted.

[0080] Exemplarily, in step S1, all hydraulic supports on the working surface are monitored by pressure sensors or electro-hydraulic control systems, and the Kriging interpolation method is used to process discrete and unevenly distributed pressure data. In the longitudinal direction, data interpolation between the time dimension and the stress value is performed for a single support to fill in the missing data between time points with larger sampling intervals to ensure the continuity of the time data. In the transverse direction, interpolation is performed between adjacent supports to estimate the spatial distribution of stress. Through longitudinal and transverse interpolation, the data is mapped into three-dimensional space, and finally a uniformly distributed and visualized three-dimensional grid stress distribution cloud map is generated, such as Figure 5 shown.

[0081] The YOLO algorithm performs real-time image recognition of high and low stress areas, including:

[0082] S301, image feature extraction. Using the SIFT, Haar and HOG algorithms, significant feature areas such as color, texture and shape in the stress cloud image can be efficiently extracted. The SIFT algorithm detects key points with scale and rotation invariance in the image, extracts local features such as extreme points and texture changes, and preliminarily forms candidate frames for significant regions; the Haar feature quickly captures the geometric structure in the image based on the brightness contrast of the rectangular area, further screens the sliding window area, and eliminates low-probability areas; the HOG algorithm captures complex textures and edge features by calculating the distribution characteristics of the local gradient direction, further optimizes the boundary and refines the candidate frame range. Finally, by merging the candidate frames generated by SIFT, Haar and HOG, and eliminating redundancy by using feature overlap and significance scoring, only the optimal frame covering the high stress area is retained, providing an accurate basis for subsequent analysis. Through the above process, local features in the stress cloud image can be efficiently captured, and a preliminary set of candidate frames can be formed to provide basic data for subsequent classification and optimization steps.

[0083] S302, classifier identification. Collect a large amount of stress cloud image data, and manually mark the specific locations of high stress areas and periodic pressure areas. Input the extracted candidate box features into the classifier (such as support vector machine SVM, adaptive enhancement AdaBoost, random forest Random Forest, etc.), classify the local features: color, extreme points, texture, geometric structure, edge features, etc., and identify high stress and periodic pressure areas. The classifier extracts the image features in the candidate box, such as extracting local key point features through SIFT, capturing gradient direction characteristics through HOG, and extracting geometric contrast information through Haar.

[0084] S303, redundant candidate box optimization. Each candidate box is assigned a confidence score (usually determined by the predicted probability of the classifier) ​​during the detection process, indicating the possibility that it is a high stress area. According to the confidence scores of the candidate boxes, they are sorted from high to low, and the candidate boxes with the highest confidence are prioritized. For the sorted candidate boxes, their overlap with other candidate boxes is calculated one by one, and the intersection and union ratio is used to determine whether the candidate boxes are too similar. Redundant candidate boxes are deleted until all candidate boxes are processed, leaving only the boxes with the highest confidence and non-overlapping.

[0085] The non-maximum suppression (NMS) algorithm is used to process the detection results, remove redundant candidate frames, and retain only the frame with the highest confidence to locate the optimal position of the top plate pressure area. The position of the selected candidate frame is optimized according to the coordinates of the center of gravity, boundary points and size of the frame to improve the boundary accuracy. The position and filling effect of the candidate frame are superimposed on the original stress cloud map to form a complete distribution map containing the optimized high stress area, and the scope and characteristics of the pressure area are clarified.

[0086] S304, manual evaluation. For the automatically identified area, the characteristics of the identified area are further analyzed manually, including area, volume and boundary characteristics, to accurately describe the spatial distribution of the high stress area. By comparing the ground stress values ​​inside and outside the high stress area, the stress gradient distribution is calculated:

[0087] ;

[0088] In the formula, are the stress values ​​in the high stress area and the surrounding area, respectively. for distance;

[0089] According to the set stress difference threshold or gradient threshold, determine whether the high stress area reaches a dangerous level. If the conditions are met, mark the area as a potential pressure area, otherwise remove it. According to the calculation results, evaluate whether the stress difference between the high stress area and the surrounding area reaches the set roof pressure criterion. Areas that meet the criteria are retained, and areas that do not meet the criteria are removed, and further check whether there are unidentified pressure areas in other areas to ensure the comprehensiveness and accuracy of the identification.

[0090] Link the MySQL database MySQL Connector package in the PyCharm installation to connect data monitoring with image recognition.

[0091] Create a database and table named monitoring_data in MySQL to store monitoring data. Next, use the OpenCV visual library to perform image recognition and insert the results into the database. Write the collected pressure value and timestamp of each sensor to the defined sensor data table in a timely manner. Model the stress field based on the stored sensor data and generate a 3D stress cloud map. Store the generated stress cloud map data in the corresponding database table and record the generation time and version information. Then, use the YOLO model to detect the converted image and identify potential problem areas or abnormal stress points. Record each detection result, including the identified area, confidence score, and detection time.

[0092] For the collection of mine pressure detection data, the electro-hydraulic control hardware and its system should realize the trigger collection function. When the hydraulic support moves, the pressure value will increase or decrease. When the stress value changes by more than 1MPa (a threshold value set inside the system), the electro-hydraulic control system of the hydraulic support should start to continuously record the stress changes in this time period until the stress returns to stability and then returns to the normal collection frequency. In this way, the key time points and all relevant data before and after them can be directly captured during the operation, so as to monitor the mine pressure changes more accurately and comprehensively.

[0093] It can be seen from the above embodiments that the present invention converts linear data into more representative spatial distribution data by integrating and analyzing the 2D working resistance curve of the support and generating a 3D mining field stress model, thereby overcoming the limitations of traditional reliance on a few hydraulic support data to infer the entire working face situation and significantly improving the upper limit of traditional evaluation methods.

[0094] The YOLO recognition algorithm's real-time monitoring features enable it to meet the requirements of dynamic monitoring and data processing. This advantage ensures that the monitoring of mine pressure changes can be carried out in a timely manner, thereby providing effective support for on-site decision-making and further improving the effectiveness of safety management and risk control.

[0095] By combining the YOLO image recognition algorithm, coal mines no longer rely on traditional data processing methods, such as exporting data sets for manual analysis, calculation and judgment, thereby improving the efficiency and accuracy of safety monitoring.

[0096] In summary, the technology of identifying periodic pressure on coal mining working faces through image recognition avoids the interference of human factors and can solve the current problems of low efficiency, poor accuracy and inaccurate positioning of periodic pressure identification on coal mining working faces.

[0097] Example 2, as another implementation of the present invention, the method for identifying periodic pressure on the roof of a working face based on the YOLO algorithm provided in the embodiment of the present invention includes:

[0098] The first step is data collection and preprocessing. Including:

[0099] (1) Missing value processing. Check the NULL (empty value) or unrecorded points in the monitoring data. If the missing value ratio is small, use interpolation methods (such as linear interpolation) to fill it. If the missing value ratio is large, remove the data segment and mark it as an abnormal segment. And confirm whether there are monitoring points with no data in a continuous time period.

[0100] (2) Outlier detection. Based on the design pressure resistance range of the hydraulic support (common values ​​are 5-30 MPa), outliers are screened by determining statistical indicators (such as 2 times the standard deviation). Values ​​that are beyond the physical reasonable range (such as negative values, extremely high values) are eliminated, the pressure change rate at adjacent time points is compared, mutation points are detected, and possible outliers are analyzed using the interquartile range (IQR).

[0101] (3) Duplicate and redundant data processing: Delete identical monitoring data (same stent number, time, and pressure value), and filter monitoring information not related to pressure analysis (such as data of unused stents).

[0102] (4) Verification of cleaning and correction. Draw a distribution graph of pressure over time and bracket number to confirm data continuity and anomaly rejection. Check the mean, median and variance of the data distribution to verify the rationality after cleaning. Compare the cleaned and corrected data with the original data to confirm the error reduction effect.

[0103] (5) Correct sensor errors. Use GPS satellite timing and unify dimensions to ensure consistency of longitudinal (time) and transverse (bracket number) data.

[0104] The second step is Kriging interpolation calculation and 3D stress cloud map generation.

[0105] When exploring unknown stress areas and processing mining site monitoring data, Kriging interpolation is an optimal unbiased estimation method that considers not only the distance between points but also the spatial distribution pattern. The core idea of ​​Kriging is that the attribute value of any point in an area can be estimated by the attribute values ​​of the known points around the point and the spatial position relationship of these points. Kriging interpolation predicts the value of unknown points by establishing an optimal prediction model with the smallest prediction error variance. Figure 5 .

[0106] Specifically include:

[0107] (2.1) Construction of variogram: The variogram is used to describe the correlation of variables (such as hydraulic support pressure) with distance in space. It can reveal the variation law of stress values ​​between adjacent monitoring points and is used to quantify the similarity of variables in space.

[0108] Divide the distance into several distance intervals (such as 0-1m, 1-2m), and count the number of point pairs in each distance interval , calculate the square of the pressure difference between the points , take the average of all point pairs and divide by 2 to get ; Variogram The expression is:

[0109] ;

[0110] In the formula, Indicates the distance The value of the variation function at time Indicates the distance The point logarithm of For spatial location The variable value at For spatial location The variable value at is the hydraulic support pressure, is the distance between two points;

[0111] The variogram ensures that the interpolation results truly reflect the spatial correlation, avoids unreasonable interpolation, reveals the stability and randomness of the pressure distribution in space, and provides a scientific basis for the interpolation process.

[0112] (2.2) Vertical and horizontal interpolation;

[0113] Spatial interpolation (transverse) includes: in the longitudinal direction, data interpolation between the time dimension and stress value is performed for a single bracket to fill the missing data between time points with large sampling intervals to ensure the continuity of time data; time interpolation (longitudinal) includes interpolation between adjacent brackets in the transverse direction to estimate the spatial distribution of stress;

[0114] Through longitudinal and lateral interpolation, the data is mapped into three-dimensional space, and finally a uniformly distributed and visualized three-dimensional grid stress distribution cloud map is generated, which converts linear data into more representative spatial distribution data, overcoming the limitations of traditional reliance on a few hydraulic support data to infer the entire working surface situation.

[0115] In the third step, the YOLO algorithm identifies high and low stress areas, including:

[0116] (3.1) Data annotation and model training;

[0117] Use the annotation tool Roboflow (https: / / universe.roboflow.com) to manually annotate the pressure distribution map. Each high stress, low stress and pressure area is marked with a corresponding label. One-click export to COCO, VOC and YOLO formats, automatically converted to txt format, the yam file has also been configured, add your own training model and your own dataset location in the edited train.py file, and then start training.

[0118] (3.2) Real-time image recognition: Use the generated 3D cloud map as input, use the YOLO algorithm to locate high and low stress areas, and output the specific bracket number range and time of the high stress area. Figure 6 The YOLO algorithm shown identifies cycles to compress the area map.

[0119] YOLO v11 provides a simple API interface, is easy to deploy, and supports PyTorch and OpenCV vision libraries (OpenCV (Open Source Computer Vision Library) is an open source computer vision and machine learning library, mainly used for image processing, video analysis, and computer vision tasks such as object detection, face recognition, edge detection, etc.). Use the YOLO v11 model to recognize the generated images.

[0120] The fourth step is manual evaluation and verification.

[0121] The Canny edge detection algorithm is used to further analyze the high stress area, extract its boundary shape and time range, identify the regional characteristics of sudden pressure rise, and lock the potential pressure area. The monitoring results are compared with the actual mine pressure manifestation, and the Kriging interpolation and YOLO algorithm model parameters are adjusted.

[0122] In the fifth step, all identified areas are saved as templates and stored in the database for further training and optimization of the recognition algorithm. By quantifying the boundary morphology of high stress areas and their location distribution in the stope, key data support is provided for subsequent evaluation work.

[0123] Embodiment 3, as Figure 3 As shown, the working face roof periodic pressure recognition system based on the YOLO algorithm provided by the embodiment of the present invention includes:

[0124] The stope three-dimensional grid stress distribution cloud map generation module 1 maps the discrete monitoring data of different supports or sensors into a spatial coordinate system based on the real-time collected hydraulic support pressure monitoring data to form a three-dimensional grid; the longitudinal and transverse stress differences are calculated by Kriging interpolation, and the discrete monitoring data are mapped into three-dimensional space to generate a uniformly distributed and visualized stope three-dimensional grid stress distribution cloud map;

[0125] Identification module 2 is used to analyze and identify different stress areas using computer vision recognition technology, and use the annotation tool Roboflow to manually annotate the three-dimensional grid pressure distribution cloud map of the stope, and identify the high stress area, low stress area and pressure area of ​​the roof with corresponding labels;

[0126] Pressure area positioning module 3 is used to analyze the high stress area through image recognition and area division according to the currently generated 3D stress cloud map, automatically identify the high and low stress areas in the 3D stress cloud map using the target detection algorithm, and accurately locate the top plate pressure area;

[0127] The image edge detection module 4 is used to identify the high stress area through the image recognition algorithm, and then conduct an in-depth analysis of the high stress area to determine whether it has the characteristics of periodic pressure; through the image edge detection technology, the time range of the roof pressure area and its corresponding hydraulic support distribution are extracted, and according to the set stress gradient or absolute stress threshold, the high stress area is extracted and the area, volume and boundary characteristics of the area are calculated; the difference in ground stress between the high stress area and the surrounding area is analyzed to determine whether the criterion for roof pressure is met; combined with the periodic prediction model and historical data, the pressure change of the roof pressure area is intelligently analyzed.

[0128] The experimental results show that the image recognition algorithm can find the periodic pressure of the working face and show high accuracy in predicting the intensity and step size of the periodic pressure. This means that in practical applications, the model can more accurately predict the periodic pressure of the working face and reduce the possibility of false positives and false negatives.

[0129] 1. Improve production efficiency: By accurately identifying the periodic pressure on the working face, measures can be taken in advance to avoid production interruptions caused by safety accidents such as roof collapse.

[0130] 2. Reduce the risk of safety accidents: Accurate identification of periodic pressure helps prevent safety accidents such as roof collapse and reduce casualties and property losses.

[0131] 3. Accurate identification of periodic pressure helps to take preventive measures in time, reduce the risk of safety accidents, protect the lives of miners, and meet the requirements of modern society for safe production.

[0132] 4. By accurately predicting the cycle pressure, the coal mine working face can optimize the support plan, reduce unnecessary material waste and manpower input, and reduce production costs. Rationally allocate resources, avoid safety hazards caused by excessive or insufficient support, and improve economic benefits.

[0133] At present, various types of mine pressure monitoring systems can only perform a relatively simple analysis of the data monitored on site, and in-depth analysis and mining are still manual methods, lacking effective working cycle and periodic pressure identification technology. A method for identifying periodic pressure on the roof of a working face based on the YOLO algorithm integrates the 2D working resistance curve of the hydraulic support and generates a 3D stope stress model to convert linear data into more representative spatial distribution data. The YOLO algorithm is combined to perform real-time image recognition of high and low stress areas, accurately locating the roof pressure area; fundamentally solving the limitations of traditional reliance on a few hydraulic support data to infer the entire working face situation.

[0134] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for identifying periodic roof pressure on a working face based on the YOLO algorithm, characterized in that: The method comprises the following steps: S1, data fusion and spatial modeling; based on the real-time collected hydraulic support pressure monitoring data, the discrete monitoring data of different supports or sensors are mapped into a spatial coordinate system to form a three-dimensional grid; the longitudinal and transverse stress differences are calculated by Kriging interpolation, and the discrete monitoring data are mapped into three-dimensional space to generate a uniformly distributed and visualized three-dimensional grid stress distribution cloud map of the mining area; S2, image preprocessing: using computer vision recognition technology to analyze and identify different stress areas, using the annotation tool Roboflow to manually annotate the three-dimensional grid pressure distribution cloud map of the mining area, and marking the high stress area, low stress area and pressure area of ​​the roof with corresponding labels; S3, target detection and area recognition: According to the currently generated 3D stress cloud map, the high stress area is analyzed through image recognition and area division, and the high and low stress areas in the 3D stress cloud map are automatically identified by the target detection algorithm, and the top plate pressure area is accurately located; S4, determination and prediction of periodic pressure; after the image recognition algorithm identifies the high stress area, the high stress area is deeply analyzed to determine whether it has the characteristics of periodic pressure; the time range of the roof pressure area and its corresponding hydraulic support distribution are extracted through image edge detection technology, and the high stress area is extracted and the area, volume and boundary characteristics of the area are calculated according to the set stress gradient or absolute stress threshold; the difference in ground stress between the high stress area and the surrounding area is analyzed to determine whether the criterion of roof pressure is met; combined with the periodic prediction model and historical data, the pressure change of the roof pressure area is intelligently analyzed; In step S3, high stress areas are analyzed by image recognition and area segmentation, including: The YOLO algorithm performs real-time image recognition of high and low stress areas, extracts significant features in the stress cloud image through low-level characteristics of color, texture and shape, captures local high stress areas, and forms a set of candidate boxes; the extracted candidate box features are input into the classifier to classify the local features and identify high stress and periodic stress areas.

2. The method for identifying periodic pressure on the roof of a working face based on the YOLO algorithm according to claim 1 is characterized in that: In step S1, based on the real-time collected hydraulic support pressure monitoring data, it includes: Through pressure sensors or electro-hydraulic control systems, all hydraulic supports on the working face are monitored after data collection and preprocessing; data collection and preprocessing include: missing value processing, outlier detection, duplicate value and redundant data processing, verification of cleaning and correction, and correction of pressure sensor errors.

3. The method for identifying periodic pressure on the roof of a working face based on the YOLO algorithm according to claim 1 is characterized in that: In step S1, the longitudinal and transverse stress differences are calculated by Kriging interpolation, including the following steps: (2.1) Construction of the variation function: Divide the distance into several distance intervals and count the number of point pairs in each distance interval , calculate the square of the pressure difference between the points , take the average of all point pairs and divide by 2 to get ; Variogram The expression is: ； In the formula, Indicates the distance The value of the variogram function at time , Indicates the distance The point logarithm of For spatial location The variable value at For spatial location The variable value at is the hydraulic support pressure, is the distance between two points; (2.2) Vertical and horizontal interpolation; In the longitudinal direction, data interpolation between the time dimension and stress value is performed for a single stent to fill in the missing data between time points with large sampling intervals; In the lateral direction, interpolation is performed between adjacent supports to estimate the spatial distribution of stress; Through longitudinal and transverse interpolation, the data is mapped into three-dimensional space to generate a uniformly distributed and visualized three-dimensional grid stress distribution cloud map.

4. The method for identifying periodic pressure on the working face roof based on the YOLO algorithm according to claim 1 is characterized in that: In step S2, the high stress area, low stress area and pressure area of ​​the top plate are marked with corresponding labels, including: one-click export to COCO, VOC and YOLO formats, and automatic conversion to txt format.

5. The method for identifying periodic roof pressure on a working face based on the YOLO algorithm according to claim 1 is characterized in that: The YOLO algorithm performs real-time image recognition of high and low stress areas, including: S301, image feature extraction; use SIFT, Haar and HOG algorithms to extract color, texture and shape feature areas in stress cloud images; SIFT algorithm detects key points with scale and rotation invariance in the image, extracts local features of extreme points and texture changes, and initially forms candidate frames of significant areas; Haar features capture the geometric structure in the image based on the brightness contrast of rectangular areas and screen the sliding window area; HOG algorithm captures complex texture and edge features by calculating the distribution characteristics of local gradient directions, optimizes boundaries and refines the range of candidate frames; by merging the candidate frames generated by SIFT, Haar and HOG, redundancy is eliminated by using feature overlap and significance scores, and only the optimal frame covering the high stress area is retained; S302, classifier identification: input the extracted candidate frame features into the classifier, classify the color, extreme point, texture, geometric structure, and edge features, and identify the high stress and periodic pressure areas; S303, redundant candidate box optimization; S304, manual evaluation.

6. The method for identifying periodic roof pressure of a working face based on the YOLO algorithm according to claim 5 is characterized in that: In step S302, redundant candidate frames are optimized, including: each candidate frame is assigned a confidence score during the detection process, and the candidate frames are sorted from high to low according to their confidence scores, and the candidate frames with the highest confidence are used as priority processing targets; for the sorted candidate frames, the overlap between the candidate frames used as priority processing targets and other candidate frames is calculated one by one, and the intersection and union ratio is used to determine whether the candidate frames are too similar; redundant candidate frames are deleted until all candidate frames are processed and only frames with the highest confidence and non-overlapping are left; specifically, the non-maximum suppression (NMS) algorithm is used to process the detection results, redundant candidate frames are removed, and only the frames with the highest confidence are retained to locate the optimal position of the top plate pressure area; for the screened candidate frames, the position is optimized according to the centroid coordinates, boundary points and size of the frames to improve the boundary accuracy, and the candidate frame positions and filling effects are superimposed on the original stress cloud map to form a complete distribution map containing the optimized high stress area, and to clarify the scope and characteristics of the pressure area.

7. The method for identifying periodic roof pressure on a working face based on the YOLO algorithm according to claim 5 is characterized in that: In step S204, manual evaluation includes: For the automatically identified area, the characteristics of the identified area are manually analyzed, including area, volume and boundary characteristics. By comparing the ground stress values ​​inside and outside the high stress area, the stress gradient distribution is calculated. The expression is: ； In the formula, are the stress values ​​in the high stress area and the surrounding area, respectively. for distance; According to the set stress difference threshold or gradient threshold, determine whether the high stress area reaches a dangerous level. If the conditions are met, mark the area as a potential pressure area, otherwise it will be eliminated. According to the calculation results, evaluate whether the stress difference between the high stress area and the surrounding area reaches the set top plate pressure criterion. The areas that meet the criteria are retained, and those that do not meet the criteria are eliminated, and further check whether there are unidentified pressure areas in other areas.

8. The method for identifying periodic roof pressure on a working face based on the YOLO algorithm according to claim 1 is characterized in that: In step S4, the pressure change in the roof pressure area is intelligently analyzed by combining the periodic prediction model and historical data, including: Link the MySQL database MySQL Connector package in the PyCharm installation to connect data monitoring with image recognition; Create a monitoring_data database and table in MySQL to store monitoring data, use the OpenCV visual library for image recognition, and insert the results into the database; write the collected pressure value and timestamp data of each sensor to the defined sensor data table in a timely manner; perform stress field modeling based on the stored sensor data and generate a 3D stress cloud map; store the generated 3D stress cloud map data in the corresponding database table, and record the generation time and version information; use the YOLO model to detect the converted image and identify potential problem areas or abnormal stress points; record each detection result, including the identified area, confidence score, and detection time; For the collection of mine pressure detection data, the electro-hydraulic control hardware and its system trigger the collection function. When the hydraulic support moves, the pressure value increases or decreases; when the stress value change exceeds a threshold set inside the system, the electro-hydraulic control system of the hydraulic support begins to continuously record the stress changes in this time period until the stress returns to stability and then returns to the normal collection frequency.

9. A working face roof periodic pressure recognition system based on YOLO algorithm, characterized in that: The system implements the method for identifying periodic pressure on the roof of a working face based on the YOLO algorithm as described in any one of claims 1 to 8, and the system includes: A three-dimensional grid stress distribution cloud map generation module (1) for the mining field is used to map the discrete monitoring data of different supports or sensors into a spatial coordinate system based on the real-time collected hydraulic support pressure monitoring data to form a three-dimensional grid; the longitudinal and transverse stress differences are calculated by Kriging interpolation, and the discrete monitoring data are mapped into a three-dimensional space to generate a uniformly distributed and visualized three-dimensional grid stress distribution cloud map for the mining field; The identification module (2) is used to analyze and identify different stress areas using computer vision recognition technology, and to manually annotate the three-dimensional grid pressure distribution cloud map of the stope using the annotation tool Roboflow, and to identify the high stress area, low stress area and pressure area of ​​the roof with corresponding labels; The pressure area positioning module (3) is used to analyze the high stress area through image recognition and area division based on the currently generated 3D stress cloud map, automatically identify the high and low stress areas in the 3D stress cloud map using the target detection algorithm, and accurately locate the top plate pressure area; The image edge detection module (4) is used to identify the high stress area through the image recognition algorithm, and then conduct an in-depth analysis of the high stress area to determine whether it has the characteristics of periodic pressure; extract the time range of the roof pressure area and its corresponding hydraulic support distribution through the image edge detection technology, extract the high stress area and calculate the area, volume and boundary characteristics of the area according to the set stress gradient or absolute stress threshold; analyze the difference in ground stress between the high stress area and the surrounding area to determine whether the criterion of roof pressure is met; and combine the periodic prediction model and historical data to conduct intelligent analysis of the pressure change in the roof pressure area.

Images