| 摘 要: 织物缺陷是具有极大长径比的缺陷,给检测工作带来了极大的难度。提出一种织物瑕疵检测模型YOLOv8-Fabric:设计自适应特征融合卷积模块(AFFConv),提高特征提取的灵活性和准确性;设计混合空间金字塔池化模块(MSPP),以捕获全局和局部上下文关系;设计多尺度逐步特征金字塔网络(MGFPN),充分交换高级语义信息和低级空间信息。实验结果表明,与 YOLOv8s比较,该算法在工业织物瑕疵检测任务中精确率提升3.63%,mAP50提升3.78%,为智能织物瑕疵检测提供理论和技术支持。 |
| 关键词: 深度学习 织物瑕疵检测 YOLOv8 |
|
中图分类号:
文献标识码: A
|
| 基金项目: 基于有限元分析的交叉韧带生物力学特性与膝骨关节炎致病机理的研究(黔科合基础-ZK[2023]一般052) |
|
| Fabric Defect Detection Algorithm Based on Improved YOLOv8 |
|
CHENG Sheng, XIANG Zhong
|
(School of Mechanical Engineering, Zhejiang Sc-i Tech University, Hangzhou 310018, China)
202230503130@mails.zstu.edu.cn; xz@zstu.edu.cn
|
| Abstract: Fabric defects characterized by extremely large aspect ratios pose significant challenges to detection.This paper proposes YOLOv8-Fabric, a fabric defect detection model, incorporating three key innovations: (1) an Adaptive Feature Fusion Convolution module (AFFConv) enhancing feature extraction flexibility and accuracy; (2) a Mixed Spatial Pyramid Pooling module (MSPP) capturing global and local contextual relationships; (3) a Mult-i scale Gradual Feature Pyramid Network (MGFPN) facilitating comprehensive exchange of high-level semantic and low-level spatial information. Experimental results demonstrate that compared with YOLOv8s, this algorithm achieves a 3.63% improvement in precision and a 3. 78% increase in mAP50 for industrial fabric defect detection tasks, providing
theoretical and technical support for intelligent fabric defect inspection. |
| Keywords: deep learning fabric defect detection YOLOv8 |
