CLAGAN: Contrastive Learning‐based Attention Generative Adversarial Network for Defect Detection of Color‐patterned Fabric

Hongwei Zhang,Xiwei Chen,Shuai Lu,Le Yao,Xia Chen

DOI: https://doi.org/10.1111/cote.12642

2022-09-23

Coloration Technology

Abstract:The pattern style of color‐patterned fabrics is varied. Defective fabric samples are scarce in the production of small batch color‐patterned fabrics. Therefore, the unsupervised defect detection method of color‐patterned fabric has attracted wide attention. Many unsupervised defect detection methods for color‐patterned fabrics based on convolutional neural networks (CNNs) have been proposed. However, CNN methods cannot learn long‐range semantic information interaction well due to the intrinsic locality of convolution operations. Besides, as the number of layers of the convolutional neural network increases, feature maps become more and more complex. Convolutional neural networks are difficult to coordinate many parameters and extract key features from complex feature maps. Both of the above problems will reduce the detection accuracy of the model for defects in color‐patterned fabrics. In this paper, we propose a Contrastive Learning‐based Attention Generative Adversarial Network (CLAGAN) for defect detection of color‐patterned fabrics. CLAGAN possesses two important parts: contrastive learning and channel attention module. Contrastive learning captures long‐range dependencies by calculating cosine similarity between different features. The channel attention module assigns different weights to each channel of feature maps, and it enables the model to extract key features from feature maps. The experimental performance verifies the effectiveness of CLAGAN. It obtained 38.25% IoU, 51.67% F1 on public dataset YDFID‐2.

engineering, chemical,chemistry, applied,materials science, textiles