Semi-Supervised Graph Convolutional Network for Textile Defect Images Recognition

Abstract

Textile defect recognition is a significant technique in the practical textile industry. However, in the practice of industrial processes, large amounts of textile defect samples are difficult to obtain, and textile industrial data with correct label is rare simultaneously. To address these challenges, a semi-supervised graph convolutional network (GCN) model is proposed to few-labeled textile defect recognition. The graph-based semi-supervised learning network can model the correlation of features and learn the discrimination between features. Meanwhile, the textile defect recognition framework can extract the textile image features through the image descriptors, enabling the whole network to be end-to-end trainable. Furthermore, a practical textile defect dataset DHU-Semi1000 is built to evaluate the performance of the proposed model. Experimental results show that the GCN network can achieve best performance compared to other algorithms.