An Empirical Study on Fabric Defect Classification Using Deep Network Models

Abstract

Fabric defect inspection plays an essential role in the textile manufacturing process. Traditional detection is carried out using defect visualization. This method obviously is inefficient in both accuracy and inspection time. Automatic detection, which is based on image processing and machine learning, has been proven to be a suitable approach for this problem. However, due to the variety of defect kinds in a broad range of deferent fabrics, existing methods are actually proposed for typical group of fabric defects. This paper aims to investigate models of deep neural network for the general fabric classification problem. In particular, two models including VGG16 and Darknet are used to classify the defect fabric categories. The models are tested for the TILDA database to evaluate their performances.