Woven Fabric Defect Detection Based on Convolutional Neural Network for Binary Classification
Fabric defect detection plays an important role in the textile industry. However, this problem is very challenging because of the variability of texture and diversity of defect. In this paper, we investigate the problem of woven fabric defect detection using deep learning. A convolutional neural network with multi-convolution and max-pooling layers is proposed. Moreover, a high-quality database, which covers the common defects in woven fabric with solid color, is built. The experiments conducted on the database indicate that the proposed model could obtain the overall detection accuracy 96.52%, which shows the potential of the model in practical application.
KeywordsWoven fabric defect Deep learning Convolutional neural network Binary classification
This research is supported by the General Research Fund of the Research Grant Council, Hong Kong (Project No.: 15202217), the Guangdong Natural Science Foundation (Project No.:2018A030310451, 2018A030310450).