Detection and classification of fabric defects using deep learning algorithms

The production of fabric, which is the most important raw material of the textile industry, consists of many stages. Due to the excess and complexity of these production stages, some defects may occur in fabrics. In the detection of defects; due to the size of the market share of the sector and the very fast production, detection with human control causes both loss of time and error detection rate to drop to 60%. Therefore, in parallel with the development of technology in recent years, more and more intelligent systems have started to be developed in the error detection of fabrics. Today, with the rapid development of artificial intelligence technology, image processing techniques have started to be applied in this sector. In this study, a real-time defect detection system has been developed on fabric using deep learning techniques. The publicly available Tilda dataset and the dataset we created ourselves were used as datasets. First, a network model was created with Convolutional Neural Network (CNN), an open source neural network library, and 89% accuracy was achieved with this method. In order to improve the study, we classified the fabrics into two classes as defective and non-defective and achieved 86% accuracy with the VGGNet16 architecture and 90% accuracy with the IneptionV3 architecture, which are pre-trained CNN models. ResNet50 has proven to have a much better model structure than other models, with an accuracy of about 95%. This means that fabric defects can be found more consistently.