摘要
Abstract
Aiming at the problem of low efficiency of traditional methods and insufficient detection accuracy of existing models in road disaster detection,a self attention mechanism based road disaster detection algorithm YOLOv5S-TB is adopted based on machine vi-sion and deep learning methods.Combining the advantages of Transformer structure,such as reducing the number of computational pa-rameters and strong ability to understand local information,into Backbone network;Replace the original FPN+PANet structure with a BIFPN structure that enhances small object detection performance and improves detection accuracy.The results showed that the detec-tion accuracy of YOLOv5S-TB algorithm on the dataset was improved by 2.8%compared to traditional YOLOv5s,and the mAP value was increased by 0.6%,effectively improving the problems of insufficient accuracy and instability in road disaster detection.关键词
YOLOv5/Transformer/BIFPN/自注意力机制/深度学习Key words
YOLOv5/Transformer/BIFPN/Self attention mechanism/Deep learning分类
信息技术与安全科学
