基于YOLOv5的岸边集装箱桥式起重机钢丝绳损伤检测方法  

Damage detection method of steel wire rope of quayside container bridge crane based on YOLOv5

作　　者：张俊科 吴敬兵[1] 吴晓晓 Zhang Junke;Wu Jingbing;Wu Xiaoxiao

机构地区：[1]武汉理工大学机电工程学院,武汉430070

出　　处：《起重运输机械》2023年第16期30-36,共7页Hoisting and Conveying Machinery

摘　　要：针对岸边集装箱桥式起重机钢丝绳传统监测方式存在着故障识别率低的问题,提出了一种基于改进YOLOv5的岸边集装箱桥式起重机钢丝绳损伤检测方法。首先,在骨干特征提取网络部分引入注意力机制CBAM,对重要的特征通道进行强化;其次,选用损失函数EIOU对训练模型进行优化;最后替换原YOLOv5算法使用的加权NMS算法,提高边框的位置精度。实验结果表明,改进后的YOLOv5目标检测网络在钢丝绳损伤数据集上对断丝、磨损、畸变3种损伤类型的平均精度均值达90.3%,比原始的YOLOv5算法提高了3%,检测效果更优,实现了对钢丝绳更快速的识别,为今后开发岸边集装箱桥式起重机钢丝绳在线监测系统提供了一定的理论基础。Considering the problem of low fault recognition rate in the traditional monitoring method of steel wire rope of quayside container bridge crane,a method of steel wire rope damage detection of quayside container bridge crane based on improved YOLOv5 is proposed.Firstly,the attention mechanism CBAM was introduced into the backbone feature extraction network to strengthen the important feature channels;secondly,the loss function EIOU was selected to optimize the training model;finally,the weighted NMS algorithm used in the original YOLOv5 algorithm was replaced to improve the position accuracy of the border.The experimental results show that for the data set of wire rope damage,the average precision of broken wire,wear and distortion of the improved YOLOv5 target detection network reaches 90.3%,which is 3%higher than the original YOLOv5 algorithm,and the detection effect is better,which accelerates the identification of wire rope,and provides a theoretical basis for the future development of online monitoring system for wire rope of quayside container bridge crane.

关 键 词：钢丝绳 损伤检测 注意力机制 EIOU YOLOv5 

分 类 号：U653.921[交通运输工程—港口、海岸及近海工程] TD532[交通运输工程—船舶与海洋工程]