摘要
路面损伤检测是支持基础设施检测的重要任务,及时、准确、自动地检测路面损伤,对于经济有效的道路养护是必要的。针对道路检测中存在漏检问题,提出一种改进的YOLOv5s的道路检测算法。首先,用CBAM注意力机制模块替换主干网络中C3模块,在关注通道特征的同时关注位置信息,加强网络对重要信息的提取能力;其次用EIoU替换GIoU损失函数,解决了GIoU误差大的同时提高了收敛速度和回归精度;最后,为使细微损伤得到有效检测,在原始网络中增加极小目标的检测的输出,使三输出变成四输出,提高模型识别率。从精度和召回率的结果可知,改进后的YOLOv5s算法平均检测精度为96.9%,相较于原YOLOv5s算法提高了7.6%。能够有效检测出道路路面损伤,且其准确率优于其他的道路检测算法。
Pavement damage detection is an important task to support infrastructure testing.Timely,accurate and automatic detection of pavement damage is necessary for economic and effective road maintenance.Aiming at the problem of missing detection in road detection,an improved YOLOv5s road detection algorithm is proposed.Firstly,the CBAM attention mechanism module is used to replace the C3 module in the backbone network,paying attention to the location information while paying attention to the channel characteristics,and strengthening the ability of the network to extract important information.Secondly,EIoU is used to replace GIoU loss function,which solves the large GIoU error and improves the convergence speed and regression accuracy.Finally,in order to effectively detect the slight damage,the output of the minimum target detection is added to the original network,so that the three outputs become four out-puts,and the model recognition rate is improved.According to the results of precision and recall,the average detec-tion accuracy of the improved YOLOv5s algorithm is 96.9%,which is 7.6%higher than that of the original YOLOv5s algorithm.It can effectively detect road pavement damage,and its accuracy is better than other road detection algorithms.
作者
王梦梦
黄德启
刘爽娜
WANG Mengmeng;HUANG Deqi;LIU shuangna(School of Electrical Engineering,Xinjiang University,Urumqi 830047,China)
出处
《激光杂志》
CAS
北大核心
2023年第5期66-71,共6页
Laser Journal
基金
国家自然科学基金项目(No.51468062)。
