Research On Miner’s Unsafe Behavior Recognition Method For Environmental Sensitivity

Coal resources are the main energy source and important industrial raw material in my country,and play an irreplaceable role in national economic development and social progress.Frequent coal mine safety accidents have brought huge losses to society,and unsafe behavior of miners is one of the direct causes of these accidents.Therefore,studying and identifying unsafe behaviors of miners is of great significance for reducing coal mine accidents.With the development of artificial intelligence technology,certain progress has been made in the study of unsafe behaviors of miners.However,the mining environment is complex,and unsafe behaviors of miners are environmentally sensitive and difficult to identify.Some behaviors require analysis of the interaction between miners and the environment to make judgments.Aiming at the environmentally sensitive issues associated with unsafe behaviors of miners,a comprehensive identification method based on rule inference engines is used to identify complex unsafe behaviors by combining miners,equipment,environmental information,and miner behavior information.The main research contents are as follows:(1)Aiming at complex underground scene problems,a method for extracting personnel,equipment,and environment information based on an improved target detection algorithm is proposed,providing a good representation of underground information for subsequent rule reasoning.First,to address the problem of small and occluded targets for miners,a feature layer containing richer information was added to the YOLOv7-tiny network to improve the detection ability of small targets.Secondly,for the gray underground environment,an attention mechanism was introduced between the trunk and neck networks of the network to improve the feature extraction ability.Finally,a new loss function was introduced to reduce the degree of freedom of loss and enhance robustness,The improved algorithm is validated on a self built mine dataset,and the experimental results show that the accuracy rate is 4.22% higher than the original model.(2)Aiming at the identification of static and dynamic unsafe behaviors,corresponding identification methods are proposed.Divide unsafe behaviors into static,dynamic and interactive unsafe behaviors.Static unsafe behavior recognition uses the MobileNetv2 network for recognition,and dynamic unsafe behavior recognition uses the improved Resnet50 network recognition embedded in the representation stream module,which preserves more motion information during the convolution process,effectively improving the accuracy of behavior recognition.The experimental results show that the accuracy of dynamic unsafe behavior recognition on the self built dataset has been improved by 7.48%,providing a more reliable information source for interactive unsafe behavior recognition.(3)Aiming at complex interactive unsafe behaviors,the rules reasoning engine is used for identification.Combining the miner area,equipment,and environment information identified by target detection and the miner behavior information identified by the behavior identification network to conduct comprehensive identification,set unsafe identification and identification rules on the Drools workbench,and infer complex unsafe behaviors that are sensitive to the environment.At the same time,the unsafe behavior data is saved to the Neo4 j graph database,and then based on the Flask framework and the ECharts visualization component,the visual display and query on the Web side are realized.The thesis has 40 figures,24 tables and 88 references.