An Application of Convolutional Neural Networks on the Classification of Textile Fabric Flaws

Abstract

Quality control of fabrics is a key issue in textile industry because it can reduce operation cost improve product quality and maintain satisfaction level of customers Convolution neural network has been shown to be an effective method for pattern recognition due to the improvement on its structure and the graphic processing unit This study focuses on applying convolutional neural networks to identify three types of fabric defects: oil mark cut weft and knotted yarn The model of convolutional neural networks is the Inception-Resnet-V2 with different layers and the evaluation method is k-fold cross validation When the number of layers is properly chosen the experimental results on six hundred defect images show that the best accuracy of the model can be up to 87 67% better than that for manual inspection In particular the classification accuracies of defects cut weft and oil mark are both larger than 90% These demonstrate that the proposed method can effectively identify the types of fabric defects to meet practical requirements

Date of Award	2019
Original language	English
Supervisor	Tzu-Tsung Wong (Supervisor)