Helmet Wear Detection Base On Improved YOLOv5

With the continuous development of social economy,power resources have become one of the main energy sources in China.In power operation,safety is the most important and basic requirement of operators.In power construction,often because the staff safety awareness is weak,the failure to wear safety helmets leads to accidents,so timely discovery and supervision of staff to wear safety helmets is a crucial task in power operations.Therefore,it is of great significance to automatically detect whether power plant workers wear safety helmets in a standardized manner.In the past,power construction sites have been a manual way to supervise whether workers wear safety helmets,but this method is not only inefficient,but also difficult to ensure multi-area coverage.Nowadays,video surveillance technology and computer vision technology have been gradually applied in various fields of life,and helmet wearing recognition technology is also a crossfield combination technology advocated by the state in recent years.Therefore,this paper proposes an improved object detection model for hard hats,MCJ-YOLOv5(M: Mobile Netv3;C: CBAM;J:detection layer),as follows:(1)Currently,there are relatively few studies on safety helmets,and no open source data can be used.Therefore,this paper firstly builds the helmet data set by self-collecting data set.According to the research requirements,2171 pieces of target data with and without helmet were obtained through network crawling and video framing.In order to expand the data set,the collected data were enhanced by rotation,reversal,increasing noise,saturation,brightness and other methods.Finally,after screening,a total of 6513 target data sets of two types were obtained as samples of this data set.(2)Configure the experimental environment and learn several current mainstream target recognition algorithms.Through comparative analysis of the principles and results of different algorithms,this paper finally selects YOLOv5 as the basic helmet recognition network model of this experiment,and uses the previously made experimental data set for training.After comparative analysis,the accuracy and detection speed of YOLOv5 model are more suitable for this helmet detection experiment than other mainstream target detection models.(3)In the electric power operation scenario,due to the variable target scale,the target detection method must meet the actual application requirements on the premise of ensuring detection accuracy.In this paper,the Mc J-yolov5 target detection model is proposed.By introducing lightweight network structure Mobile Netv3 to replace the CSPDarknet53 backbone network structure of YOLOv5,the model size is reduced and the detection speed of the model is improved.Secondly,the convolutional attention mechanism module(CBAM)is introduced into the backbone network.Finally,in order to improve the detection performance of small targets in the power plant environment,this paper adds another detection layer on the infrastructure of the original model to further improve the feature extraction ability of the model,so as to strengthen the model’s detection performance of small targets and the ability to deal with complex and dense scenes.The improved YOLOv5 algorithm was trained and tested by the self-made helmet data set,in which the average detection accuracy(m AP)value of the model reached 91.5%,the size of the model reached 23.5MB,and the detection speed of the model was 48 frames /s.Although the detection accuracy of the improved YOLOv5 algorithm was reduced by 0.4% compared with the original model,However,the model size and detection speed are reduced and increased by 151.6MB and 6 frames /s respectively.Through comparative analysis of experiments,the improved Mc Jyolov5 target detection model is more suitable for real-time detection of power plant safety hats than the improved YOLOv5 target detection model.