Research On Unsafe Behavior Recognition Method Of Miners In Mine Belt Area

Coal is one of the main energy sources currently used in our country.It is of great significance to ensure the safety of coal production for the economic development of our country.In this paper,RGB video surveillance image is used to identify the behavior of miners in the belt conveyor area.Combined with the environment of the area,the main research work is as follows:(1)In the aspect of miner’s behavior representation,the key frame of miner’s behavior video sequence is extracted by equal interval sampling method,which improves the efficiency of calculation and makes the motion contour clearer.In order to solve the problem of self-coverage of motion description,this paper proposes a method of describing the behavior of miners with double-layered optical flow-motion history image;in order to solve the problem of low recognition rate of variable-speed motion in the process of miners’ motion,Gaussian pyramid is introduced,the motion information is obtained by using the spatial gradient information,finally the features of hog and Hu moments are extracted to represent the behavior of miners.(2)Xgboost method is used to identify the behavior data set of miners built by ourselves,and the final recognition rate is 90.5%,which verifies the validity of the behavior representation method of miners.In this paper,a deep transfer learning method is proposed to train the behavior recognition model of miners,which meets the needs of the unsafe behavior recognition of miners.The recognition accuracy on the behavior data set of miners is 95.3%.Through experiments,the robustness of deep transfer learning and miner behavior recognition method is verified.The recognition accuracy is 94.7% on DHA,65.5% on Holywood2,which is about 5% higher than that of traditional xgboost.In order to realize the continuous monitoring of the unsafe behaviors of miners,the template inclusion rate method is used to complete the action sequence segmentation in the long sequence.(3)In order to recognize the unsafe behaviors of the miners in the belt area,this paper uses the deep learning algorithm Mask R-CNN to recognize the environmental targets in the belt area and achieve pixel level segmentation.The recognition accuracy of the targets in this area is 96.2%.Based on the combination of segmented pixel set and miner’s behavior,the model of miner’s unsafe behavior identification is constructed.The accuracy rate of miner’s unsafe behavior identification is 95.1%,and the time of each behavior identification is about 1000 ms.Compared with other related research methods,the time complexity of this paper is lower.In this paper,experiments show that the recognition method of unsafe behavior of miners has a good effect on the identification of unsafe behavior in the belt area,which can meet the needs of real-time operation;the robustness of the recognition algorithm is verified on the public data set.Automatic real-time monitoring of typical unsafe behaviors of miners in the belt area plays an important role in reducing accidents in this area.The paper has 39 pictures,14 tables,and 94 references.