一种基于改进YOLOv5的复杂背景织物疵点检测方法

A defect detection method in complex background fabric based on improved YOLOv5 model

受背景中复杂纹理的影响,有花纹和图案的织物检测疵点一直是该领域的难题。对此提出了一种改进YOLOv5模型的疵点检测方法,该方法将模型中原来的特征金字塔网络模块替换为双向特征金字塔网络,以更好地融合不同尺度下提取的特征。实验中使用该模型对具有花纹和图案的织物样本进行检测,结果表明改进后的模型有较为明显的性能提升。

The defect detection in fabrics with patterns has been a challenge in this field due to the complex textures in the background.In this regard,a defect detection method for improving the YOLOv5 model is proposed.This method replaces the origi⁃nal Feature Pyramid Network in the YOLOv5 model with Bidirectional Feature Pyramid Network to better fuse features extracted at different scales.In the experiment,the model is used to detect the fabric samples with patterns,and the results show that the im⁃proved model has a more obvious performance improvement.