Research On The Monitoring Algorithm Of Safety Protection Appliances In Construction Site Based On Deep Learning

The frequent occurrence of personnel accidents in the construction field is often due to the weak safety awareness of the construction personnel,who do not wear safety protective equipment according to the specifications.Safety helmets are the most convenient and effective safety protection equipment,so it is necessary to use video monitoring technology to monitor whether the construction personnel wear safety helmets in real time,improve the safety awareness of the construction personnel,and reduce the probability of accidents.This paper studies the helmet wearing monitoring algorithms at home and abroad in recent years,improves the relevant algorithms,and proposes a helmet wearing monitoring algorithm based on deep learning.The main research contents of this paper are as follows:(1)Research the current mainstream target detection algorithm,and consider the detection accuracy,detection speed and lightweight degree of the target detection algorithm.Finally,choose the lightweight algorithm YOLOv5,which performs well in detection accuracy and speed,as the main research object of this paper.(2)Aiming at the problem that YOLOv5 network model can not extract the key feature information in the image sufficiently,this paper proposes to add the convolutional attention mechanism to YOLOv5 network,strengthen the attention of the detection network to the key targets,improve the detection accuracy of the model,and this method will not increase the computational complexity.The experiment shows that the improved algorithm is effective and feasible.(3)YOLOv5 network does not fully use the extracted shallow features,resulting in the failure to recognize small target objects.Therefore,a weighted bi-directional feature pyramid network is proposed to be added to YOLOv5 network,so that the network can make full use of shallow features to solve the problem of multi-scale target detection.Through experiments,it is verified that the improved algorithm can improve the perception of multi-scale targets.(4)Aiming at the problem that the iteration speed of the loss function of YOLOv5 network model is slow,a method using Si OU function to replace the original loss function is proposed.Through experiments,it is verified that the improved method can improve the convergence speed of training,and the location of prediction box is more accurate.