Real-Time Fabric Defect Detection Using Machine Vision and Neural Networks

Early detection of defects plays a very important part in controlling the quality of a manufactured product, particularly in mass production processes. Many imaging-based systems for fabric quality inspection have been focused on still images. In contrast, real fabric defect detection needs a system that works within the moving production line. This technological barrier hinders the adoption of automated visual inspection for small and medium industries for developing countries. This research presents a prototype of a real-time machine vision system that overcomes this problem by using low-cost vision system (webcam) and neural network to learn different types of defect patterns on blurring images (moving fabric in the production line) based on geometrical features. The time constraint of the real-time requirement was modelled based on Q-model. The experimental results on a moving conveyor line indicate that our approach achieves 97.4% accuracy in detecting and recognizing defect type of blurred textile images and only requires 156 ms for one inspection process. The prototype shows that the proposed system is acceptable for small and medium industries.