| 摘 要: 为了解决织物瑕疵形态的多样性以及瑕疵检测速度慢等问题,提出了基于YOLOv5的增强算法。其中,在C3模块中引入了通用倒瓶颈模,大大减少了模型的参数计算量。此外,还引入了自注意力模块PSA,可以快速锁定目标信息。针对织物瑕疵形态的多样性引入了特征融合模块CCFM。在自建织物瑕疵数据集和 DAGM2007数据集上的实验结果表明,改进YOLOv5的平均精度(mAP)达到94.4%且检测FPS能达到526.31,验证了该增强算法的有效性。在工业织物缺陷监测方面具有较高的应用价值。 |
| 关键词: 深度学习 织物瑕疵检测 YOLOv5 自注意力 特征融合 |
|
中图分类号:
文献标识码: A
|
| 基金项目: 教育部人文社会科学研究一般项目(23YJCZH281);上海市哲学社会科学规划课题(2022ZGL010);信息网络安全公安部重点实验室开放课题 |
|
| Research on Fabric Defect Detection Based on Improved YOLOv5 |
|
WU Jiachen1, ZHU Guoliang2, ZHANG Huashu1
|
(1.School of Mechanical Engineering, Zhejiang Sc-i Tech University, Hangzhou 310018, China;
2.Zhejiang Dafeng Industry Co., Ltd., Yuyao 315400, China)
202220501050@mail.zstu.edu.cn; zgl@chinadafeng.com; zhanghua@mail.zstu.edu.cn
|
| Abstract: To address issues such as the diversity of fabric defect morphology and slow detection speed, an enhanced algorithm based on YOLOv5 is proposed. The C3 module incorporates a universal inverted bottleneck structure, significantly reducing the model’s computational parameters. Additionally, a sel-f attention module (PSA) is introduced to quickly locate target information. A feature fusion module (CCFM) is integrated to accommodate the diversity of fabric defect morphologies. Experimental results on a sel-f built fabric defect dataset and the DAGM2007 dataset demonstrate that the improved YOLOv5 achieves a mean average precision (mAP) of 94.4% and a detection speed of 526.31 FPS, validating the effectiveness of the enhanced algorithm. This approach holds high application value for industrial fabric defect monitoring. |
| Keywords: deep learning fabric defect detection YOLOv5 sel-f attention feature fusion |
