Advanced Textile Technology ›› 2022, Vol. 30 ›› Issue (5): 12-20.DOI: 10.19398/j.att.202107051

• Invited Column: Image Processing and Numerical Simulation • Previous Articles     Next Articles

Fabric defect detection based on improved RefineDet

RUAN Mengyu, LI Min, HE Ruhan, YAO Xun   

  1. School of Computer and Artificial Intelligence, Wuhan Textile University, Wuhan 430200
  • Received:2021-07-30 Online:2022-09-10 Published:2022-09-19

基于改进RefineDet的织物疵点检测

阮梦玉, 李敏, 何儒汉, 姚迅   

  1. 武汉纺织大学计算机与人工智能学院,武汉 430200
  • 作者简介:阮梦玉(1996-),女,湖北黄冈人,硕士研究生,主要从事图像处理方面的研究。
  • 基金资助:
    湖北省教育厅科技项目(D20161605)

Abstract: To realize the automatic detection and classification of fabric defects, a defect detection method based on Refinedet is proposed. Firstly, VGG16 is changed into a full convolution network to extract fabric image features. Secondly, to obtain the important features of defects and suppress unnecessary features, an attention mechanism is added to the anchor refinement module (ARM). To improve the classification performance of the network, SE module (sequence and exception, SE)is added to the transfer connection block (TCB). Finally, the object detection module (ODM)returns the detection results to the accurate target position, predicts the category of defects and locates the defects. The experimental results show that the average accuracy of this algorithm is 79.7%,which is 5.0% higher than that of the traditional algorithm.

Key words: deep learning, defect detection, RefineDet, VGG16, attention mechanism

摘要: 为了实现织物疵点的自动检测与分类,提出了一种基于改进RefineDet的疵点检测方法。首先,将VGG16改为全卷积网络对织物图像特征进行提取;其次,为了获取疵点重要的特征并抑制不必要的特征,在Anchor细化模块(Anchor refinement module,ARM)中加入了注意力机制;为了提高网络的分类性能,在传输连接块(Transfer connection block,TCB)中加入了SE模块(Squeeze and excitation,SE);最后,目标检测模块(Object detection module,ODM)将检测的结果回归到准确的目标位置,并预测疵点的类别,对疵点进行定位。结果表明:本文算法对孔、污渍、纱疵和线状4种类别织物图像的均值平均精度mAP达到了79.7%,比传统RefineDet检测方法均值平均精度提高了5.0%,具有良好的分类和定位效果。

关键词: 深度学习, 疵点检测, RefineDet, VGG16, 注意力机制

CLC Number: