Machine Learning-Based Textile Fabric Defect Detection Network

The Deep Learning technology known as Convolutional Neural Network (CNNs) are relatively new in the machine learning that has been employed for computer vision in the field of identification of defects in textile fabrics which needs to be done accurately and also autonomously. In this paper, we design a new machine learning specialized network to reduce the decision action for LSTM based on CNNs. That is a rather positive result; however, there are several critical concerns which can be addressed further: the differences in the texture of the fabric, the unpredictable nature of the defects' distribution, and the lack of large-labeled datasets. owing to the advancements in feature extraction and segmentation new approaches like transfer learning and deep CNN based models like U-net, Res net etc. have begun to solve the issue. Unfortunately, efforts to improve these systems' effectiveness and reliability are still challenged by issues related to high calculability costs, sample over-learning and the need for vast annotated dataset. Therefore, the aim of this work is to establish a defect detecting model with high reliability and generalization ability for different textile fabrics using CNN technique. Besides, this study helps in gaining insights into other strategies that could be deployed in order to leverage on those strategies while using slightly more complex mixed structures and modifying loss function.