Research On Object Detection Method Of Calcium Carbide Production Workshop Based On Deep Learning

Safety issues of the chemical industry are related to people’s safety,social stability,and economic development.And local provincial government more focus on the chemical production safety.Chemical companies’ top priority is to protect the personal safety of on-site workers.The workers’ security in calcium carbide production safety can be protected because of the wearing of protective clothing and safety helmets,which has important practical significance.Manual inspection or traditional object detection methods are used to conduct safety inspections for personnel in the calcium carbide production workshop.In view of the problems that the installation position of the network camera in the calcium carbide production workshop causes the target object to become smaller,the network is over-fitting,and the original network feature extraction capability is insufficient,the following work is carried out:1.According to the situation of the calcium carbide production workshop,the data set is preprocessed.Through the data preprocessing operation,the detection performance of the network in complex scenes is improved.The impact of various data enhancement methods are discussed on the indicators of object detection algorithms.The best effective data enhancement method is selected to increase the richness of the data set and improve the generalization and detection effect of the network.By simulating the missing label and expanding the bounding box and Analysising the experimental results,the influence of the quality of the data set on the algorithm’s accuracy is obtained.2.According to the problems of the detection about the protective clothing and safety helmets on the calcium carbide production workshop,a dual attention network based on YOLOv5 algorithm is proposed.The Efficient Channel Attention module and the Pyramid Segmentation Attention module are integrated into the Spatial Pyramid Pool module and the Bottleneck module of the YOLOv5 algorithm,which obtain more global context information,make up for the insufficient of the convolution and enhance the ability of the network to extract features and learn multi-scale information.By using the safety helmets wearing dataset and the data set prepared in this paper,the effectiveness of the improved method is verified.Compared with the original YOLOv5 algorithm,the improved method improves the average accuracy by 2.7% under different thresholds.The effectiveness of the improved method is confirmed by a large number of comparative experiments.3.Safety production camera monitoring system is designed and implemented.The development model bases on the separation of front and back ends.The alarm information detected by the network algorithm is transmitted to the back-end in real time and stored in the database.By using the VUE framework,the front-end and back-end interaction is realized.And then the alarm information is obtained and rendered on the website page in real time,which is helpful for coal chemical plants safety management to reduce accidents.