A coal mine safety supervision method and system based on AI training and model technology

Abstract

The present invention provides a coal mine safety supervision method and system based on AI training and model technology, including: collecting several groups of on-site scene images under different operating conditions and performing image annotation training respectively, performing intensive training on the obtained annotation results to generate several scene samples, and constructing a corresponding description data set to construct an ONNX model, using AI technology to train the ONNX model for risk, inputting real-time on-site video transmission into the ONNX model for risk management, determining several violation scenario conditions at the coal mine site based on the management and control output information of the ONNX model, capturing corresponding key violation features for violation backup, and generating and displaying corresponding violation handling suggestions. The model trained with deep learning technology utilizes its powerful recognition ability and can accurately identify almost all daily violations in coal mines.

Description

Technical Field

[0001] The present invention relates to the field of coal mine safety technology, and in particular to a coal mine safety supervision method and system based on AI training and modeling technology. Background Art

[0002] In coal mines, due to complex working conditions, large numbers of personnel, and extensive operating areas, traditional manual supervision methods have numerous limitations, such as low efficiency, slow response, and numerous blind spots. Therefore, the use of real-time video automatic supervision technology based on deep learning is an efficient and intelligent solution that can significantly improve coal mine safety management. This technology, based on deep learning real-time video, can significantly improve the efficiency and accuracy of coal mine safety management and reduce human interference and misjudgment.

[0003] Currently, some monitoring technologies for coal mine safety exist, such as electronic fencing. However, these solutions are mostly limited to fixed-scene monitoring and fail to fully cover the complex and ever-changing daily violations in coal mines. This results in poor monitoring effectiveness and limited coverage. Therefore, how to accurately identify almost all daily violations in coal mines has become an urgent problem.

[0004] Therefore, the present invention provides a coal mine safety supervision method and system based on AI training and model technology. Summary of the Invention

[0005] The present invention provides a coal mine safety supervision method and system based on AI training and model technology. The model trained by deep learning technology takes advantage of its powerful recognition ability and can accurately identify almost all daily violations in coal mines.

[0006] The present invention provides a coal mine safety supervision method based on AI training and model technology, comprising:

[0007] Step 1: Collect several groups of scene scene images under different working situations, perform image annotation training on each group of the scene scene images, and obtain a corresponding annotation result set for each group of the scene scene images;

[0008] Step 2: Perform intensive training on the same annotation results to generate a number of scenario samples corresponding to each set of annotation results, and construct a corresponding description data set for each scenario sample;

[0009] Step 3: Use the description data set to build an onnx model, use AI technology to train the onnx model for risk management, and input real-time on-site video transmission into the onnx model for risk management;

[0010] Step 4: Determine several violation scenarios at the coal mine site based on the control output information of the onnx model, capture the corresponding key violation features for violation backup, and generate and display corresponding violation handling suggestions.

[0011] In one practicable manner,

[0012] The step 1 comprises:

[0013] Step 11: Sampling operations at the coal mine site to obtain several operation scenarios at the coal mine site, searching for several groups of scene images corresponding to each operation scenario in the big data, and annotating each group of scene images using labelimg to obtain several image labels corresponding to each scene image;

[0014] Step 12: Rearranging the plurality of image labels corresponding to each of the operation scenarios to obtain label arrangement information corresponding to the operation scenario being in a safe state, and using the label arrangement information to perform feedback recognition on each of the corresponding on-site scene images to obtain safety features and non-safety features of each of the on-site scene images;

[0015] Step 13: Perform safety item training and non-safety item training on the scene scene image according to the safety features and the non-safety features, obtain a number of safety events and non-safety events corresponding to the scene scene image, and mark each of the safety events and non-safety events in the scene scene image.

[0016] In one practicable manner,

[0017] The step 13 includes:

[0018] Step 131: performing rasterization processing on each of the scene scene images, dividing the corresponding scene scene image into scenes using the safety features and the non-safety features, and obtaining an initial safety area and an initial non-safety area of ​​the scene scene image;

[0019] Step 132: Using AI technology to perform operation simulations on the initial safe area and the initial non-safe area, respectively, to obtain a plurality of first operation data of the initial safe area and a plurality of second operation data of the initial non-safe area, and to determine the operation scenario corresponding to each grid in the scene scene image;

[0020] Step 133: placing a preset worker sample in each of the grids to perform a specified operation example, determining a corresponding safety level of the preset worker sample in each of the grids according to the corresponding operation scenario, and adjusting the first area range of the initial safe area and the second area range of the initial unsafe area according to the corresponding safety level of each grid;

[0021] Step 134: Construct several safety events of the preset worker sample in the adjusted safety area based on several first operating scenarios included in the adjusted safety area, and mark them in the on-site scene image; construct several non-safety events of the preset worker sample in the adjusted non-safe area based on several second operating scenarios included in the adjusted non-safe area, and mark them in the on-site scene image.

[0022] In one practicable manner,

[0023] The step 2 comprises:

[0024] Step 21: performing cluster analysis on the annotation results to obtain several result attribute classes of the coal mine site, constructing corresponding coal mine site attribute models based on the current site image of the coal mine site and the result attribute classes, and extracting dynamic action information contained in each of the coal mine site attribute models;

[0025] Step 22: Reward and enhance the dynamic action information in each of the coal mine site attribute models, obtain the action consequence of each dynamic action in the corresponding coal mine site attribute model, and construct a scenario sample of the coal mine site based on the result attribute class, the corresponding dynamic action, and the corresponding action consequence;

[0026] Step 23: Using natural language description technology to describe each of the scenario samples, obtain scenario text sub-data corresponding to each scenario sample; using digital information extraction technology to extract digital information from each of the scenario samples, obtain scenario digital sub-data corresponding to each scenario sample;

[0027] Step 24: Use the scenario digital sub-data to correct the numerical precision of the corresponding scenario text sub-data, use the scenario text sub-data to correct the numerical attributes of the corresponding scenario digital sub-data, and use the corrected scenario digital sub-data and the corrected scenario text sub-data to construct a description data set corresponding to the scenario sample.

[0028] In one practicable manner,

[0029] The step 3 comprises:

[0030] Step 31: Optimize the description data set using model building conditions, build an ONNX model using the optimized data set, determine several risk items at the coal mine site based on the scenario samples, perform risk training on the ONNX model using AI technology, and improve the ONNX model's identification process for each risk item;

[0031] Step 32: transmitting the real-time field data of the coal mine site to the onnx model for key feature extraction to obtain a plurality of field key features, and performing risk identification on the field key features using each of the identification processes to obtain the corresponding risk value of the coal mine site under each of the risk items;

[0032] Step 33: Identify the risk hazards corresponding to the risk project based on the risk value, determine several current hazard characteristics of the coal mine site, conduct a range assessment on the current hazard characteristics, determine the current danger range of the coal mine based on the assessment results, and use AI technology to perform temporary risk management and control on the danger range based on the hazard attributes of the danger range and the corresponding current hazard characteristics.

[0033] In one practicable manner,

[0034] Also includes:

[0035] The risk control information of the coal mine site is obtained, the effectiveness of the risk control information is evaluated, and a temporary risk control effectiveness report of the coal mine site is obtained and backed up.

[0036] In one practicable manner,

[0037] The step 4 comprises:

[0038] Step 41: Obtain risk control information of the coal mine site and transmit it to the control output information of the ONNX model for control supervision. Determine the real-time control progress of the coal mine based on the control output information of the ONNX model. After the control is completed, use the control output information to restore the coal mine site.

[0039] Step 42: Identify several violation scenarios at the coal mine site in the restored information, construct a risk tree diagram for the coal mine site based on the restored sub-information corresponding to each violation scenario, obtain the performance sub-features and associated sub-features corresponding to each violation scenario, and determine the key violation feature corresponding to each violation scenario;

[0040] Step 43: Back up the key violation features, and use the onnx model to perform feature elimination training on the key violation features, generate and display violation handling suggestions corresponding to each violation scenario.

[0041] In one practicable manner,

[0042] Also includes:

[0043] When a risky project occurs on site in the coal mine, the on-site risk range of the risky project is located and a corresponding alarm is issued.

[0044] In one practicable manner,

[0045] Also includes:

[0046] Performing image transformation on each of the on-site scene images to obtain a supplementary image of each of the scene images;

[0047] The supplementary images are regarded as scene scene images, and labelimg is used to perform image annotation on each group of the scene scene images to obtain a plurality of image labels corresponding to each of the scene scene images.

[0048] The present invention provides a coal mine safety supervision system based on AI training and model technology, including:

[0049] An image training module is used to collect a plurality of sets of scene scene images under different operation situations, perform image annotation training on each set of the scene scene images, and obtain an annotation result set corresponding to each set of the scene scene images;

[0050] A sample construction module is used to perform intensive training on the same annotation results, generate a number of scenario samples corresponding to each set of annotation results, and construct a corresponding description data set for each scenario sample;

[0051] A management and control execution module is used to build an onnx model using the description data set, perform risk training on the onnx model using AI technology, and input real-time on-site video transmission into the onnx model for risk management and control;

[0052] The control and management analysis module is used to determine several violation scenarios at the coal mine site based on the control and management output information of the onnx model, capture the corresponding key violation features for violation backup, and generate and display corresponding violation handling suggestions.

[0053] The achievable beneficial effects of the above technical solution are: in order to greatly improve the speed and efficiency of identifying potential risks at coal mine sites, thereby effectively preventing the occurrence of safety accidents, first screen the scene images under different circumstances in the big data, and construct the scene samples of the coal mine site by labeling and strengthening the scene images, and describe each scene sample, so as to construct the onnx model based on the obtained description data set, and further use AI technology to train the onnx model for risks. At this time, the real-time field data of the coal mine site can be input into the onnx model for risk management, and the violation status of the coal mine site is determined, and the status is backed up and reminded to take appropriate violation handling. Through deep learning, a variety of scenarios in coal mines are trained, and the trained algorithm has been successfully applied to the AI ​​video-assisted risk identification system to provide protection for the safe production of coal mines.

[0054] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0055] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0057] Figure 1 Schematic diagram of the workflow of a coal mine safety supervision method based on AI training and model technology in an embodiment of the present invention;

[0058] Figure 2 This is a schematic diagram of the composition of a coal mine safety supervision system based on AI training and model technology in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention and are not used to limit the present invention.

[0060] Example 1

[0061] This embodiment provides a coal mine safety supervision method based on AI training and model technology, such as Figure 1 Shown, including:

[0062] Step 1: Collect several groups of scene scene images under different working situations, perform image annotation training on each group of the scene scene images, and obtain a corresponding annotation result set for each group of the scene scene images;

[0063] Step 2: Perform intensive training on the same annotation results to generate a number of scenario samples corresponding to each set of annotation results, and construct a corresponding description data set for each scenario sample;

[0064] Step 3: Use the description data set to build an onnx model, use AI technology to train the onnx model for risk management, and input real-time on-site video transmission into the onnx model for risk management;

[0065] Step 4: Determine several violation scenarios at the coal mine site based on the control output information of the onnx model, capture the corresponding key violation features for violation backup, and generate and display corresponding violation handling suggestions.

[0066] In this example, the operation scenarios represent different operations performed at a coal mine site;

[0067] In this example, a scene image represents a specific work situation. To effectively monitor violations, we first need to collect high-definition images of actual work scenarios, such as whether workers are wearing hard hats, a key safety indicator. Using a real-time video surveillance system, we carefully selected and accumulated thousands of real images of workers correctly wearing hard hats. This aims to build a solid and rich training dataset for subsequent precision recognition technology, ensuring the accuracy of safety supervision.

[0068] In this example, image annotation training refers to the process of determining the scene information contained in the scene scene image by adding labels;

[0069] In this example, reinforcement training refers to the process of highlighting various types of scene information in the scene scene image;

[0070] In this example, scenario samples represent job samples of different job situations;

[0071] In this example, risk training refers to the process of writing the risk identification process into the onnx model;

[0072] In this example, the violation scenario represents the situation presented by the violation phenomenon occurring in the coal mine site.

[0073] The working principle and beneficial effects of the above technical solution: In order to greatly improve the speed and efficiency of identifying potential risks at coal mine sites, thereby effectively preventing the occurrence of safety accidents, first screen the scene images under different circumstances in the big data, and construct the scene samples of the coal mine site by labeling and strengthening the scene images. Then, describe each scene sample, and construct the onnx model based on the obtained description data set. Further use AI technology to train the onnx model for risks. At this time, the real-time field data of the coal mine site can be input into the onnx model for risk management, and the violation status of the coal mine site is determined, and the status is backed up and reminded to take appropriate violation handling. Through deep learning, a variety of scenarios in coal mines are trained. At the same time, the trained algorithm has been successfully applied to the AI ​​video-assisted risk identification system to provide protection for the safe production of coal mines.

[0074] Example 2

[0075] On the basis of Example 1, the coal mine safety supervision method based on AI training and model technology, step 1 includes:

[0076] Step 11: Sampling operations at the coal mine site to obtain several operation scenarios at the coal mine site, searching for several groups of scene images corresponding to each operation scenario in the big data, and annotating each group of scene images using labelimg to obtain several image labels corresponding to each scene image;

[0077] Step 12: Rearranging the plurality of image labels corresponding to each of the operation scenarios to obtain label arrangement information corresponding to the operation scenario being in a safe state, and using the label arrangement information to perform feedback recognition on each of the corresponding on-site scene images to obtain safety features and non-safety features of each of the on-site scene images;

[0078] Step 13: Perform safety item training and non-safety item training on the scene scene image according to the safety features and the non-safety features, obtain a number of safety events and non-safety events corresponding to the scene scene image, and mark each of the safety events and non-safety events in the scene scene image.

[0079] In this instance, the image label represents a label used to distinguish the attributes of different scene information in the image;

[0080] In this example, layout rearrangement refers to the process of adjusting the positions of image labels in the scene image after eliminating redundant information in the operation situation;

[0081] In this example, the label arrangement information represents the label distribution presented when the operation situation is in a safe state;

[0082] In this example, the safety feature represents the feature presented by the safe work in the scene scene image, and the non-safety feature represents the feature presented by the unsafe work in the scene scene image;

[0083] In this example, the safety item training refers to the process of performing safety items that appear in the simulated scene scene image, and the non-safety item training refers to the process of performing unsafe items that appear in the simulated scene scene;

[0084] In this example, a safe event refers to a work event that can be completed safely as presented in its on-site scene image, and a non-safe event refers to a work event that cannot be completed safely as presented in its on-site scene image.

[0085] The working principle and beneficial effects of the above technical solution are as follows: In order to train the corresponding algorithm model according to the specific scenarios of different coal mines, a large amount of data sets need to be labeled to ensure that the algorithm can accurately identify various violations. Although this increases the workload of the initial implementation of the technology, it is also a key step in achieving technical customization and precision. First, the coal mine site is sampled and several types of operating situations are determined. Then, the on-site scene images of each operating situation are identified in the big data. By annotating the on-site scene images and rearranging the generated image labels according to the operating situation, the label arrangement information of the operating situation in a safe state is determined, and then the safety features and non-safety features of each on-site scene image are determined. Further, the safety events and non-safety events in the on-site scene images are determined through training, and the safety events and non-safety events are presented in the on-site scene images through labeling. In this way, all operating situations at the coal mine site can be identified in advance, and all possible safety events and all possible non-safety events that may occur at the coal mine site can be determined, laying the foundation for subsequent safety supervision work.

[0086] Example 3

[0087] Based on Example 2, the coal mine safety supervision method based on AI training and model technology, step 13 includes:

[0088] Step 131: performing rasterization processing on each of the scene scene images, dividing the corresponding scene scene image into scenes using the safety features and the non-safety features, and obtaining an initial safety area and an initial non-safety area of ​​the scene scene image;

[0089] Step 132: Using AI technology to perform operation simulations on the initial safe area and the initial non-safe area, respectively, to obtain a plurality of first operation data of the initial safe area and a plurality of second operation data of the initial non-safe area, and to determine the operation scenario corresponding to each grid in the scene scene image;

[0090] Step 133: placing a preset worker sample in each of the grids to perform a specified operation example, determining a corresponding safety level of the preset worker sample in each of the grids according to the corresponding operation scenario, and adjusting the first area range of the initial safe area and the second area range of the initial unsafe area according to the corresponding safety level of each grid;

[0091] Step 134: Construct several safety events of the preset worker sample in the adjusted safety area based on several first operating scenarios included in the adjusted safety area, and mark them in the on-site scene image; construct several non-safety events of the preset worker sample in the adjusted non-safe area based on several second operating scenarios included in the adjusted non-safe area, and mark them in the on-site scene image.

[0092] In this example, rasterization refers to the process of dividing the scene image into a number of pixel grids;

[0093] In this example, the initial safe area refers to the area where the safe features are located, and the initial unsafe area refers to the area where the unsafe features are located;

[0094] In this example, operation simulation refers to the use of AI technology to speculate on the operation of the initial safe area and the initial unsafe area to determine the means of their operation process;

[0095] In this example, the first operating data represents data presented by the initial safe area during the operation simulation process, and the second operating data represents data presented by the initial non-safe area during the operation simulation process;

[0096] In this example, the preset worker sample represents a virtual sample established based on the actual situation of the workers;

[0097] In this example, the prescribed operation example indicates that the standard operation corresponding to the operation performed by the worker in the grid is determined according to the position of the grid in the scene image. For example, the prescribed operation example performed by the worker in grid A is: placing an object steadily on the belt;

[0098] In this example, the safety level represents the safety level corresponding to when a preset worker sample performs a prescribed work example in the grid;

[0099] In this example, the first area range indicates the range position and range size of the initial safety area, and the second area range indicates the range position and range size of the initial non-safety area.

[0100] The working principle and beneficial effects of the above technical solution: In order to further identify safety incidents and non-safety incidents, the scene scene image is first rasterized, and then the scene scene image is divided into an initial safety area and an initial non-safety area according to safety features and non-safety features. In this way, the scope of safety incidents and the scope of non-safety incidents can be roughly determined. Then, the safety level of each grid is determined by simulation and the scope of the initial area is adjusted. Then, the safety incidents and non-safety incidents in the adjusted area are determined according to the operating scenarios in the adjusted area and marked in the scene scene image. In this way, the fine specifications of the grid can be used to improve the accuracy of the range division, and safety incidents and non-safety incidents in multiple scene scene images can be processed at the same time, thereby improving the quality and efficiency of the preliminary preparation work and ensuring the response speed of subsequent safety supervision.

[0101] Example 4

[0102] On the basis of Example 1, the coal mine safety supervision method based on AI training and model technology, step 2 includes:

[0103] Step 21: performing cluster analysis on the annotation results to obtain several result attribute classes of the coal mine site, constructing corresponding coal mine site attribute models based on the current site image of the coal mine site and the result attribute classes, and extracting dynamic action information contained in each of the coal mine site attribute models;

[0104] Step 22: Reward and enhance the dynamic action information in each of the coal mine site attribute models, obtain the action consequence of each dynamic action in the corresponding coal mine site attribute model, and construct a scenario sample of the coal mine site based on the result attribute class, the corresponding dynamic action, and the corresponding action consequence;

[0105] Step 23: Using natural language description technology to describe each of the scenario samples, obtain scenario text sub-data corresponding to each scenario sample; using digital information extraction technology to extract digital information from each of the scenario samples, obtain scenario digital sub-data corresponding to each scenario sample;

[0106] Step 24: Use the scenario digital sub-data to correct the numerical precision of the corresponding scenario text sub-data, use the scenario text sub-data to correct the numerical attributes of the corresponding scenario digital sub-data, and use the corrected scenario digital sub-data and the corrected scenario text sub-data to construct a description data set corresponding to the scenario sample.

[0107] In this instance, the result attribute class represents the result of classifying the annotation results with the same attributes into one class;

[0108] In this example, the dynamic action information represents the dynamic information presented in the coal mine site attribute model;

[0109] In this example, the coal mine site attribute model represents the model when running under a result attribute class at the coal mine site;

[0110] In this example, the action consequence represents the consequence of dynamic actions in the coal mine site simulated by the coal mine site attribute model;

[0111] In this example, the scenario text sub-data represents the result of describing the scenario sample in natural language, and the scenario number sub-data represents the numbers contained in the scenario sample;

[0112] In this example, the numerical precision represents the result of performing numerical correction on the data in the scenario text sub-data, and the numerical attribute represents the result of performing numerical correction on the numerical description in the scenario numerical sub-data.

[0113] The working principle and beneficial effects of the above technical solution are as follows: the annotation results are classified by clustering to obtain several result attribute classes of the coal mine site, and then the dynamic action information in the coal mine site attribute model is extracted and enhanced by modeling, and the action consequences corresponding to each dynamic action are determined, so as to construct the corresponding scenario samples, and then the sub-data in the scenario samples are obtained through natural language description technology and digital information extraction technology, and then the sub-data are trained to generate a description data set for each scenario sample. In this way, it can be ensured that the model algorithm adopted can achieve high accuracy and improve the efficiency and accuracy of data analysis.

[0114] Example 5

[0115] Based on Example 1, the coal mine safety supervision method based on AI training and model technology, step 3 includes:

[0116] Step 31: Optimize the description data set using model building conditions, build an ONNX model using the optimized data set, determine several risk items at the coal mine site based on the scenario samples, perform risk training on the ONNX model using AI technology, and improve the ONNX model's identification process for each risk item;

[0117] Step 32: transmitting the real-time field data of the coal mine site to the onnx model for key feature extraction to obtain a plurality of field key features, and performing risk identification on the field key features using each of the identification processes to obtain the corresponding risk value of the coal mine site under each of the risk items;

[0118] Step 33: Identify the risk hazards corresponding to the risk project based on the risk value, determine several current hazard characteristics of the coal mine site, conduct a range assessment on the current hazard characteristics, determine the current danger range of the coal mine based on the assessment results, and use AI technology to perform temporary risk management and control on the danger range based on the hazard attributes of the danger range and the corresponding current hazard characteristics.

[0119] In this example, optimization processing refers to the process of converting the description dataset into the dataset required for building the onnx model using the model building conditions;

[0120] In this example, risk training refers to the process of adding risk monitoring procedures to the onnx model;

[0121] In this example, the risk item represents the risk that may occur at the coal mine site, and the risk value represents the data value presented when a risk item occurs at the coal mine site;

[0122] In this example, scope assessment means analyzing the extent of the impact of a current hazard feature on the coal mine site.

[0123] The working principle and beneficial effects of the above technical solution are as follows: the functions of the model need to be utilized to achieve safety supervision. First, the description data set is optimized according to the model construction conditions to construct the onnx model. Then, AI technology is used to train the onnx model based on the risk items in the scenario samples, and the identification process of risk items in the onnx model is improved. Then, by analyzing the key on-site features in the onnx model, the risk items and risk values ​​of the coal mine site are determined, and then several current hazard characteristics of the coal mine site and the danger range of each current hazard characteristic are determined. Finally, AI technology is used to conduct temporary risk management of the coal mine site. The rigorously trained model can automatically and efficiently analyze violation information in the coal mine monitoring video, and then perform risk management based on the violation information. Real-time analysis of violations not only means that problems can be discovered in a timely manner, but also greatly improves the speed and efficiency of responding to potential risks, thereby effectively preventing the occurrence of safety accidents.

[0124] Example 6

[0125] Based on Example 5, the coal mine safety supervision method based on AI training and model technology further includes:

[0126] The risk control information of the coal mine site is obtained, the effectiveness of the risk control information is evaluated, and a temporary risk control effectiveness report of the coal mine site is obtained and backed up.

[0127] The working principle of the above technical solution is to determine the temporary control effectiveness report of the coal mine site by evaluating the effectiveness of risk control information, and provide technical reference for relevant personnel.

[0128] Example 7

[0129] Based on Example 1, the coal mine safety supervision method based on AI training and model technology, step 4 includes:

[0130] Step 41: Obtain risk control information of the coal mine site and transmit it to the control output information of the ONNX model for control supervision. Determine the real-time control progress of the coal mine based on the control output information of the ONNX model. After the control is completed, use the control output information to restore the coal mine site.

[0131] Step 42: Identify several violation scenarios at the coal mine site in the restored information, construct a risk tree diagram for the coal mine site based on the restored sub-information corresponding to each violation scenario, obtain the performance sub-features and associated sub-features corresponding to each violation scenario, and determine the key violation feature corresponding to each violation scenario;

[0132] Step 43: Back up the key violation features, and use the onnx model to perform feature elimination training on the key violation features, generate and display violation handling suggestions corresponding to each violation scenario.

[0133] In this example, the restored sub-information represents information presented by a violation scenario;

[0134] In this example, the manifestation sub-feature represents the external manifestation characteristics of the coal mine site under the influence of the violation scenario state, and the association sub-feature represents the characteristics when the violation scenario state changes due to the influence of the changes in other violation scenario states.

[0135] The working principle and beneficial effects of the above technical solution are as follows: the onnx model is used to control and supervise the risk management information of the coal mine site to determine the real-time management progress of the coal mine, so as to restore the coal mine site after the management is completed to identify the violation scenarios and construct a risk tree diagram of the coal mine site. The diagram can be used to clearly know the manifestation sub-features and related sub-features of each violation scenario, thereby determining the key violation features of each violation scenario. Finally, the obtained violation features are backed up, and the onnx model is used for feature elimination training to obtain violation handling suggestions. In this way, the trained model can be used to handle violations, the on-site violation handling process can be backed up, and the violation handling suggestions for each location are determined, so that more risk response experience can be imparted to workers during subsequent education.

[0136] Example 8

[0137] Based on Example 1, the coal mine safety supervision method based on AI training and model technology further includes:

[0138] When a risky project occurs on site in the coal mine, the on-site risk range of the risky project is located and a corresponding alarm is issued.

[0139] The working principle and beneficial effects of the above technical solution are: when risks occur at the coal mine site, an alarm is issued in time to protect the safety of personnel at the coal mine site and reduce the economic losses of the coal mine.

[0140] Example 9

[0141] Based on Example 2, the coal mine safety supervision method based on AI training and model technology further includes:

[0142] Performing image transformation on each of the on-site scene images to obtain a supplementary image of each of the scene images;

[0143] The supplementary images are regarded as scene scene images, and labelimg is used to perform image annotation on each group of the scene scene images to obtain a plurality of image labels corresponding to each of the scene scene images.

[0144] In this example, image transformations include: ① geometric transformations such as random rotation, cropping, and perspective; ② color space transformations such as brightness, contrast, and saturation adjustment and noise addition; and ③ image mixing (CutMix, MixUp, etc.) sample splicing and mixing transformations.

[0145] The working principle and beneficial effects of the above technical solution are as follows: In order to further expand the coverage of the image, the scene scene image is transformed and the function of the model is refined.

[0146] Example 10

[0147] This embodiment provides a coal mine safety supervision system based on AI training and model technology, such as Figure 2 Shown, including:

[0148] An image training module is used to collect a plurality of sets of scene scene images under different operation situations, perform image annotation training on each set of the scene scene images, and obtain an annotation result set corresponding to each set of the scene scene images;

[0149] A sample construction module is used to perform intensive training on the same annotation results, generate a number of scenario samples corresponding to each set of annotation results, and construct a corresponding description data set for each scenario sample;

[0150] A management and control execution module is used to build an onnx model using the description data set, perform risk training on the onnx model using AI technology, and input real-time on-site video transmission into the onnx model for risk management and control;

[0151] The control and management analysis module is used to determine several violation scenarios at the coal mine site based on the control and management output information of the onnx model, capture the corresponding key violation features for violation backup, and generate and display corresponding violation handling suggestions.

[0152] In this example, the operation scenarios represent different operations performed at a coal mine site;

[0153] In this example, a scene image represents a specific work situation. To effectively monitor violations, we first need to collect high-definition images of actual work scenarios, such as whether workers are wearing hard hats, a key safety indicator. Using a real-time video surveillance system, we carefully selected and accumulated thousands of real images of workers correctly wearing hard hats. This aims to build a solid and rich training dataset for subsequent precision recognition technology, ensuring the accuracy of safety supervision.

[0154] In this example, image annotation training refers to the process of determining the scene information contained in the scene scene image by adding labels;

[0155] In this example, reinforcement training refers to the process of highlighting various types of scene information in the scene scene image;

[0156] In this example, scenario samples represent job samples of different job situations;

[0157] In this example, risk training refers to the process of writing the risk identification process into the onnx model;

[0158] In this example, the violation scenario represents the situation presented by the violation phenomenon occurring in the coal mine site.

[0159] The working principle and beneficial effects of the above technical solution: In order to greatly improve the speed and efficiency of identifying potential risks at coal mine sites, thereby effectively preventing the occurrence of safety accidents, first screen the scene images under different circumstances in the big data, and construct the scene samples of the coal mine site by labeling and strengthening the scene images. Then, describe each scene sample, and construct the onnx model based on the obtained description data set. Further use AI technology to train the onnx model for risks. At this time, the real-time field data of the coal mine site can be input into the onnx model for risk management, and the violation status of the coal mine site is determined, and the status is backed up and reminded to take appropriate violation handling. Through deep learning, a variety of scenarios in coal mines are trained. At the same time, the trained algorithm has been successfully applied to the AI ​​video-assisted risk identification system to provide protection for the safe production of coal mines.

[0160] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A coal mine safety supervision method based on AI training and model technology, characterized in that: include: Step 1: Collect several groups of scene scene images under different working situations, perform image annotation training on each group of the scene scene images, and obtain a corresponding annotation result set for each group of the scene scene images; Step 2: Perform intensive training on the same annotation results to generate a number of scenario samples corresponding to each set of annotation results, and construct a corresponding description data set for each scenario sample; Step 3: Use the description data set to build an onnx model, use AI technology to train the onnx model for risk management, and input real-time on-site video transmission into the onnx model for risk management; Step 4: Determine several violation scenarios at the coal mine site based on the control output information of the onnx model, capture the corresponding key violation features for violation backup, and generate and display corresponding violation handling suggestions; The step 2 comprises: Step 21: performing cluster analysis on the annotation results to obtain several result attribute classes of the coal mine site, constructing corresponding coal mine site attribute models based on the current site image of the coal mine site and the result attribute classes, and extracting dynamic action information contained in each of the coal mine site attribute models; Step 22: Reward and enhance the dynamic action information in each of the coal mine site attribute models, obtain the action consequence of each dynamic action in the corresponding coal mine site attribute model, and construct a scenario sample of the coal mine site based on the result attribute class, the corresponding dynamic action, and the corresponding action consequence; Step 23: Using natural language description technology to describe each of the scenario samples, obtain scenario text sub-data corresponding to each scenario sample; using digital information extraction technology to extract digital information from each of the scenario samples, obtain scenario digital sub-data corresponding to each scenario sample; Step 24: Use the scenario digital sub-data to correct the numerical precision of the corresponding scenario text sub-data, use the scenario text sub-data to correct the numerical attributes of the corresponding scenario digital sub-data, and use the corrected scenario digital sub-data and the corrected scenario text sub-data to construct a description data set corresponding to the scenario sample.

2. A coal mine safety supervision method based on AI training and model technology as claimed in claim 1, characterized in that: The step 1 comprises: Step 11: Sampling operations at the coal mine site to obtain several operation scenarios at the coal mine site, searching for several groups of scene images corresponding to each operation scenario in the big data, and annotating each group of scene images using labelimg to obtain several image labels corresponding to each scene image; Step 12: Rearranging the plurality of image labels corresponding to each of the operation scenarios to obtain label arrangement information corresponding to the operation scenario being in a safe state, and using the label arrangement information to perform feedback recognition on each of the corresponding on-site scene images to obtain safety features and non-safety features of each of the on-site scene images; Step 13: Perform safety item training and non-safety item training on the scene scene image according to the safety features and the non-safety features, obtain a number of safety events and non-safety events corresponding to the scene scene image, and mark each of the safety events and non-safety events in the scene scene image.

3. A coal mine safety supervision method based on AI training and model technology as claimed in claim 2, characterized in that: The step 13 includes: Step 131: performing rasterization processing on each of the scene scene images, dividing the corresponding scene scene image into scenes using the safety features and the non-safety features, and obtaining an initial safety area and an initial non-safety area of ​​the scene scene image; Step 132: Using AI technology to perform operation simulations on the initial safe area and the initial non-safe area, respectively, to obtain a plurality of first operation data of the initial safe area and a plurality of second operation data of the initial non-safe area, and to determine the operation scenario corresponding to each grid in the scene scene image; Step 133: placing a preset worker sample in each of the grids to perform a specified operation example, determining a corresponding safety level of the preset worker sample in each of the grids according to the corresponding operation scenario, and adjusting the first area range of the initial safe area and the second area range of the initial unsafe area according to the corresponding safety level of each grid; Step 134: Construct several safety events of the preset worker sample in the adjusted safety area based on several first operating scenarios included in the adjusted safety area, and mark them in the on-site scene image; construct several non-safety events of the preset worker sample in the adjusted non-safe area based on several second operating scenarios included in the adjusted non-safe area, and mark them in the on-site scene image.

4. A coal mine safety supervision method based on AI training and model technology as claimed in claim 1, characterized in that: The step 3 comprises: Step 31: Optimize the description data set using model building conditions, build an ONNX model using the optimized data set, determine several risk items at the coal mine site based on the scenario samples, perform risk training on the ONNX model using AI technology, and improve the ONNX model's identification process for each risk item; Step 32: transmitting the real-time field data of the coal mine site to the onnx model for key feature extraction to obtain a plurality of field key features, and performing risk identification on the field key features using each of the identification processes to obtain the corresponding risk value of the coal mine site under each of the risk items; Step 33: Identify the risk hazards corresponding to the risk project based on the risk value, determine several current hazard characteristics of the coal mine site, conduct a range assessment on the current hazard characteristics, determine the current danger range of the coal mine based on the assessment results, and use AI technology to perform temporary risk management and control on the danger range based on the hazard attributes of the danger range and the corresponding current hazard characteristics.

5. A coal mine safety supervision method based on AI training and model technology as claimed in claim 4, characterized in that: Also includes: The risk control information of the coal mine site is obtained, the effectiveness of the risk control information is evaluated, and a temporary risk control effectiveness report of the coal mine site is obtained and backed up.

6. A coal mine safety supervision method based on AI training and model technology as claimed in claim 1, characterized in that: The step 4 comprises: Step 41: Obtain risk control information of the coal mine site and transmit it to the control output information of the ONNX model for control supervision. Determine the real-time control progress of the coal mine based on the control output information of the ONNX model. After the control is completed, use the control output information to restore the coal mine site. Step 42: Identify several violation scenarios at the coal mine site in the restored information, construct a risk tree diagram for the coal mine site based on the restored sub-information corresponding to each violation scenario, obtain the performance sub-features and associated sub-features corresponding to each violation scenario, and determine the key violation feature corresponding to each violation scenario; Step 43: Back up the key violation features, and use the onnx model to perform feature elimination training on the key violation features, generate and display violation handling suggestions corresponding to each violation scenario.

7. A coal mine safety supervision method based on AI training and model technology as claimed in claim 1, characterized in that: Also includes: When a risky project occurs on site in the coal mine, the on-site risk range of the risky project is located and a corresponding alarm is issued.

8. A coal mine safety supervision method based on AI training and model technology as claimed in claim 2, characterized in that: Also includes: Performing image transformation on each of the on-site scene images to obtain a supplementary image of each of the scene images; The supplementary images are regarded as scene scene images, and labelimg is used to perform image annotation on each group of the scene scene images to obtain a plurality of image labels corresponding to each of the scene scene images.

9. A coal mine safety supervision system based on AI training and model technology, characterized by: include: An image training module is used to collect a plurality of sets of scene scene images under different operation situations, perform image annotation training on each set of the scene scene images, and obtain an annotation result set corresponding to each set of the scene scene images; A sample construction module is used to perform intensive training on the same annotation results, generate a number of scenario samples corresponding to each set of annotation results, and construct a corresponding description data set for each scenario sample; A management and control execution module is used to build an onnx model using the description data set, perform risk training on the onnx model using AI technology, and input real-time on-site video transmission into the onnx model for risk management and control; A control and analysis module is used to determine several violation scenarios at the coal mine site based on the control and management output information of the onnx model, capture the corresponding key violation features for violation backup, and generate and display corresponding violation handling suggestions; The sample construction module performs intensive training on the same annotation results to generate a number of scenario samples corresponding to each set of annotation results, and constructs a corresponding description data set for each scenario sample, including: Performing cluster analysis on the labeled results to obtain several result attribute classes of the coal mine site, constructing corresponding coal mine site attribute models based on the current site image of the coal mine site and the result attribute classes, and extracting dynamic action information contained in each of the coal mine site attribute models; Reward and enhance the dynamic action information in each of the coal mine site attribute models, obtain the action consequence of each dynamic action in the corresponding coal mine site attribute model, and construct a scenario sample of the coal mine site based on the result attribute class, the corresponding dynamic action, and the corresponding action consequence; Using natural language description technology to describe each of the scenario samples, respectively, to obtain scenario text sub-data corresponding to each of the scenario samples; using digital information extraction technology to extract digital information from each of the scenario samples, respectively, to obtain scenario digital sub-data corresponding to each of the scenario samples; The scenario digital sub-data is used to correct the numerical precision of the corresponding scenario text sub-data, the scenario text sub-data is used to correct the numerical attributes of the corresponding scenario digital sub-data, and the corrected scenario digital sub-data and corrected scenario text sub-data are used to construct a description data set corresponding to the scenario sample.

Images