TextileNet: A Deep Learning Approach for Textile Fabric Material Identification from OCT and Macro Images

Textile fabric classification is a critical task in the textile industry, and traditional methods have limitations in terms of accuracy and efficiency. In this study, we propose a novel approach using deep learning and transfer learning techniques for fabric classification. Two different datasets are utilized, namely the Optical Coherence Tomography (OCT) images dataset and a macro image dataset. We compare the performance of these datasets using a pre-trained convolutional neural network (CNN) architecture. Using deep learning, automatically figures out important fabric details from the raw data, making it good at sorting materials accurately and efficiently. The OCT dataset gives us very detailed information about fabric structures, while the macro dataset includes a wide variety of fabric samples. When we compare them, we can see which one works better for classification tasks. By using transfer learning, we can make our model even better at classifying fabrics without needing a lot of training data. Our research unveils MobileNetV2’s exceptional accuracy (99.87% for OCT images and 95% for macro images) in fabric classification. The research shows that MobileNetV2 has a lot of potential for actually classifying fabrics in the real world.