Research On The Detection Of Employee Violation Behavior In Mining Safety Engineering

The issue of coal mine safety production has always been a key topic.Safety production is the prerequisite of enterprise production work,and safety supervision is the guarantee of safety production;in order to effectively supervise the illegal behavior of employees working underground,this paper studies an intelligent identification In the method of violations,three frequently occurring violations were selected as the research objects:not wearing a helmet,not wearing a mask,and violations of violations.By using deep learning methods,an intelligent recognition model for violations of miners was constructed to realize violations of miners.Real-time detection and recognition of behavior.This article is to build a high-quality data set to support model training and follow-up experiments;after conducting video frame-by-frame analysis and image processing on the surveillance video data collected from coal mining enterprises,we can obtain the image data of the three types of violations required.Increase the number of effective detection targets in the image data,and use the CutMix method to enhance the data.This method can increase the number of effective targets in the violation data set constructed in this article,thereby improving the model training effect,and the data enhancement in this article is proved through experiments Effectiveness of the method.In order to improve the accuracy of the detection of the three violations studied in this article,the article proposes an improvement strategy from the perspectives of improving the detection rate and detection accuracy of the YOLO v3 algorithm;for the improvement of the algorithm detection rate,the GhostNet network structure is used to replace Darknet-53.Network structure method;to improve the detection accuracy of the algorithm,the YOLO v3 algorithm is improved from the two perspectives of Anchor's re-clustering and multi-scale feature fusion.For the re-optimization of Anchor size,the K-means algorithm is used to improve the accuracy of this article.Cluster analysis is performed on the target location size in the miner's violation data set to determine the center position of the 9 groups of sizes of the target suggestion box;and for the improvement of multi-scale feature fusion,this paper introduces an enhanced feature pyramid structure Aug-FPN,The method of algorithm extraction feature and multi-scale feature fusion is improved to improve the feature extraction ability of the algorithm and thus the detection accuracy of the algorithm model;and through experimental design and comparison with other mainstream algorithms,the improved algorithm in this paper is better for people without helmets and masks.And the three types of violation detection,violations and cross-border violations,can achieve better results.However,the detection effect of small targets in image data is poor.In order to further improve the detection of small target objects,this paper introduces the YOLOF algorithm,which uses the GhostNet network structure to replace the backbone network of the algorithm.Aiming at the decoder part of the YOLOF algorithm,the algorithm is improved by introducing a deep separable convolution and replacing the activation function.Through the design of comparative experiments in different scenarios,the superiority of the improved YOLOF algorithm is verified.Figure 40 table 11 reference 70...