基于改进YOLO V5的缺陷识别与定量分析  

Defect identification and quantitative analysis method based on improved YOLO V5

作　　者：刘洋 李全勇[1,2] 顾健 杨帆[1,2] 王文博 LIU Yang;LI Quanyong;GU Jian;YANG Fan;WANG Wenbo(Key Laboratory of Photoelectric Measurement and Optical Information Transmission Technology of Ministry of Education,College of Optoelectronic Engineering,Changchun University of Science and Technology,Changchun 130022,China;National Demonstration Center for Experimental Opto-Electronic Engineering Education,College of Optoelectronic Engineering,Changchun University of Science and Technology,Changchun 130022,China)

机构地区：[1]长春理工大学光电工程学院光电测控与光信息传输技术教育部重点实验室,长春130022 [2]长春理工大学光电工程学院光电工程国家级教学示范中心,长春130022

出　　处：《无损检测》2023年第1期14-19,22,共7页Nondestructive Testing

基　　金：吉林省教育厅科学技术研究项目(JJKH20210823KJ);吉林省科技资源开放共享服务平台与科研条件保障项目(20191004022TC)。

摘　　要：采用YOLO V5算法对碳纤维复合材料预置夹杂缺陷的识别方法展开研究。为了在提高检测精度的同时保证检测效率,通过添加通道注意力机制、空间注意力机制、使用k-means++重新聚类先验框和优化损失函数等措施改进原算法。利用改进后的网络训练缺陷数据集,每秒处理的图片数量逾12幅,平均精度达到98.8%,召回率为98.1%。与其他算法相比,该算法检测精度和速度都有所提高,可满足实时性和准确性要求。The YOLO V5 algorithm was used to study the identification method of carbon fiber composite material preset defects.In order to improve the detection accuracy while ensuring the detection efficiency,the original algorithm was improved by adding channel attention mechanism and spatial attention module,using k-means++to re-cluster the prior box and optimizing the loss function.Using the improved network training defect data set,the number of images processed per second reached more than 12 frames,the average accuracy reached 98.8%,and the recall rate was 98.1%.Compared with other algorithms,the detection accuracy and speed of this algorithm have been improved to a certain extent which can meet the real-time and accuracy requirements.

关 键 词：无损检测 YOLO V5 通道注意力机制 空间注意力机制 夹杂缺陷 

分 类 号：TG115.28[金属学及工艺—物理冶金] TB332[金属学及工艺—金属学]