%0 Journal Article
%T 一种检测安全帽佩戴的深度学习算法<br>A Deep Learning Algorithm for Detecting Helmet Wearing
%A 周筱雨
%A 王朝立
%A 孙占全
%J Software Engineering and Applications
%P 60-71
%@ 2325-2278
%D 2022
%I Hans Publishing
%R 10.12677/SEA.2022.111008
%X 在生产流水线过程中，工人因未佩戴安全帽而引发的悲剧不计其数，为了帮助工厂监督人员督促工人们佩戴安全帽以保障自身人身安全，传统的检测方式往往会分离安全帽与人体，导致无法判断人与安全帽之间的佩戴关系，为此本文提出了一种基于深度学习的改进YOLO算法的检测算法，在现有的YOLO算法基础上，对其损失函数进行改进，将原本的GIoU_Loss函数替换为CIoU_Loss函数，使得算法识别准确率得到提升且收敛速度更快。针对YOLO算法在检测时对安全帽的定位易受噪声等多方面的影响，为此本文设计了安全帽检测的多阶段算法，先用YOLO算法对工人所在区域进行定位，获取工人位置信息；然后选取工人的头部区域，区域面积稍加扩大以提高算法容错率；最后，通过一个基础的神经卷积网络对是否佩戴安全帽进行判断。通过实验结果表明，本文的方法相较之前的方法，在安全帽识别的准确率方面达到了92.73%的效果，在识别速度上此方法的检测速度也比之前的方法提升了一倍，证明本文的方法能够满足期望要求。<br />
In the process of production line, there are countless tragedies caused by workers not wearing safety helmets. In order to help factory supervisors urge workers to wear safety helmets to ensure their personal safety, the traditional detection method often separates the safety helmets from the human body, resulting in the inability to judge the wearing relationship between people and safety helmets, Therefore, this paper proposes a detection algorithm of improved Yolo algorithm based on deep learning. Based on the existing Yolo algorithm, its loss function is improved, and replaces the original GIoU_Loss with CIoU_Loss, so that the recognition accuracy of the algorithm is improved and the convergence speed is faster. In view of the fact that the positioning of safety helmet by Yolo algorithm is easily affected by noise and other aspects, a multi-stage algorithm for safety helmet detection is designed in this paper. Firstly, the area where workers are located is located by Yolo algorithm to obtain workers’ location information; then, the head area of the worker is selected, and the area is slightly expanded to improve the fault tolerance of the algorithm; finally, a basic neural convolution network is used to judge whether to wear a helmet. The experimental results show that compared with the previous methods, this method achieves 92.73% in the accuracy of helmet recognition, and the detection speed of this method is twice as fast as the previous methods, which proves that this method can meet the expected requirements.
%K 深度学习，神经网络，安全帽检测，生产安全，YOLO算法<br>Deep Learning
%K Neural Network
%K Helmet Detection
%K Production Safety
%K Yolo Algorithm
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=48532