基于深度学习的安全帽检测算法的实现与优化

Implementation and Optimization of Safety Helmet Detection Algorithm Based on Deep Learning

安全帽作为用来保护头部以免高空重物误伤的一种非常重要的安全用具,对于建筑施工场所的工作人员来说是必不可少的。在有高空重物坠落的高风险施工场所中,不管是对于施工工人还是对于工地管理者而言,是否正确佩戴安全帽都是当前必要且迫切需要解决的问题。本文根据安全帽检测的需求分析,使用Yolov5网络模型对实际建筑施工场所收集的图像数据集进行识别分析,研究Yolov5相比之前版本以及R-CNN系列模型算法之间的优势与缺点,优化完善算法使得模型检测准确率更高,速度更快。实验结果证明,Yolov5模型检测的正确率可以达到96%,可以应用于实时性检测。

As a very important safety appliance to protect the head from the accidental injury of high altitude heavy objects,the safety helmet is essential for the workers in the construction site.In construction sites with high risk of heavy objects falling from high altitude,whether to wear safety helmets correctly is a necessary and urgent problem for both construction workers and site managers.In this paper,according to the analysis of the requirements of hard hat detection,Yolov5 network model is used to identify and analyze the image data set collected from the actual construction site,and the advantages and disadvantages of Yolov5 compared with the previous version and R-CNN series model algorithm are studied.The optimization and improvement of the algorithm makes the model detection accuracy higher and speed faster.The experimental results show that the accuracy of Yolov5 model detection can reach 96%,which can be applied to real-time detection.