基于改进YOLOv5的船体焊缝缺陷自动检测方法  

Automatic detection method of hull weld defects based on improved YOLOv5

作　　者：杜玉红 叶萧然 侯守明[2] DU Yu-hong;YE Xiao-ran;HOU Shou-ming(Hebi Instiute of Engineering and Technology,Henan Polytechnic University,Hebi 458030,China;Henan Polytechnic University,Jiaozuo 454003,China)

机构地区：[1]河南理工大学鹤壁工程技术学院,河南鹤壁458030 [2]河南理工大学,河南焦作454003

出　　处：《舰船科学技术》2023年第19期185-188,共4页Ship Science and Technology

基　　金：河南省高等学校重点科研项目计划(教科技(2021)383号)(22B520018);河南理工大学鹤壁工程技术学院校重点课题(2022-KJZD-006)。

摘　　要：为解决采用人工手段检测船体焊缝速度慢、准确度低的问题,提出基于改进YOLOv5的船体焊缝缺陷自动检测方法。利用相机采集船体焊缝图像,使用正弦灰度变换对焊缝图像进行处理,避免焊缝图像特征消失,提高正常焊缝与存在缺陷焊缝间的对比度,将处理后焊缝图像作为YOLOv5网络的输入样本,经网络Backbone、Neck以及Head部分处理,输出焊缝缺陷自动检测结果,并使用GhostNet替换YOLOv5网络主体部分的一般卷积层(CBS),降低网络进行船体焊缝缺陷检测的计算量和资源消耗量。实验结果表明,采用正弦灰度变换后的图像更加清晰,可突出显示焊缝缺陷特征,提升焊缝缺陷检测结果精准。改进后网络训练损失函数为0.15,平均准确率为98%,可实现不同焊缝位置的缺陷检测。To solve the problems of slow speed and low accuracy in manually detecting ship welds,an improved YOLOv5 based automatic detection method for ship weld defects is proposed.Using a camera to collect images of ship welds,using sine grayscale transformation to process the weld image to avoid the disappearance of weld image features and improve the contrast between normal welds and defective welds.The processed weld image is used as an input sample for the YOLOv5 network,and processed by the Backbone,Neck,and Head sections of the network to output automatic weld defect detection results,and use GhostNet to replace the general convolutional layer(CBS)of the main part of YOLOv5 network,reducing the computational and resource consumption of the network for detecting ship weld defects.The experimental results show that the image obtained by using sine grayscale transformation is clearer,highlighting the characteristics of weld defects,and improving the accuracy of weld defect detection results.The improved network training loss function is 0.15,with an average accuracy of 98%.It can achieve defect detection at different weld positions.

关 键 词：YOLOv5 船体焊缝 缺陷检测 网络优化 灰度变换 图像采集 

分 类 号：TG441.7[金属学及工艺—焊接]