摘要
针对瓦楞纸板表面缺陷检测速度慢和识别准确率低等问题,提出基于改进YOLOv5s的轻量化瓦楞纸板表面缺陷检测算法YOLOv5s-GCS。将YOLOv5s骨干网络中原有的Conv模块替换为GhostConv模块,用C2f模块替换C3模块,并集成置换注意力机制(SA)模块;在YOLOv5s颈部网络末端引入SA模块;通过构建瓦楞纸板表面缺陷数据集进行试验验证。试验结果显示,YOLOv5s-GCS算法平均精度均值达到95.0%、召回率达到89.2%、精确率达到92.5%,较原始YOLOv5s分别提高2.3%,1.3%,2.8%;检测速度达到19.9帧/s,较原始YOLOv5s提高5.7帧/s。YOLOv5s-GCS算法更有利于迁移部署与实际应用。研究为表面缺陷领域的实时检测提供参考。
To address the problems of slow detection speed and low recognition accuracy in corrugated cardboard surface defect detection,YOLOv5s-GCS was proposed as a lightweight corrugated cardboard surface defect detection algorithm based on the improved YOLOv5s.The original Conv module in the YOLOv5s backbone network was replaced with the GhostConv module,the C3 module was replaced with the C2f module,and the Replacement Attention Mechanism(SA)module was integrated.SA module was introduced at the terminal of YOLOv5s neck network;and tests were conducted to validate the algorithm by constructing the corrugated cardboard surface defect dataset.The test results show that the average precision mean mAP,recall R,and precision P of YOLOv5s-GCS algorithm are 95.0%,89.2%,and 92.5%,respectively,which are 2.3%,1.3%,and 2.8%higher than that of the original YOLOv5s.The detection speed reaches 19.9 fps,which is 5.7 fps higher than that of the original YOLOv5s.YOLOv5s-GCS algorithm is more conducive to carry out corrugated cardboard surface defect detection migration deployment and practical applications.The study can provide a reference for real-time detection in the field of surface defects.
作者
李西兴
刘涛
周宏娣
吴锐
陈佳豪
LI Xixing;LIU Tao;ZHOU Hongdi;WU Rui;CHEN Jiahao(School of Mechanical Engineering,Hubei University of Technology,Wuhan 430068,China)
出处
《包装与食品机械》
CAS
北大核心
2024年第5期88-95,共8页
Packaging and Food Machinery
基金
国家自然科学基金项目(51805152)
湖北省自然科学基金项目(2024AFB816)
湖北省重点研发计划项目(2023BEB043)。
