Fabric Inspection using Computer Vision

The textile industry is one of the basic parts of global manufacturing and is about to undergo a significant metamorphosis with the rise in technologies of artificial intelligence and computer vision. Traditionally, fabric inspection has been labor-intensive, requiring a lot of time and highly critical eyes for spotting defects, making it a process often blemished by human error. The advent of AI-enabled technology is allowing textile producers to implement automated inspection systems that enhance efficiency and raise quality benchmarks. This shift not only increases productivity but also guarantees uniformity in production by minimizing defects, ushering in a new age of intelligent manufacturing. In this project, we developed a fabric inspection system based on computer vision with the YOLOv8 model. Our approach was trained on a dataset containing most frequent fabric defects, including holes, foreign yarn, and slubs (thread errors), as well as surface contamination like dirty marks, dye patches, and oil patches. Several models have been experimented on the dataset, including VGG, ResNet, and YOLOv8, and YOLOv8 performs best in all major metrics: accuracy, precision, recall, F1 score, model size, and prediction time. To make this solution easily accessible and user-friendly, we implemented a FastAPI-based web application that is capable of real-time fabric quality evaluation. The application integrates our YOLOv8 model with a Logitech C270 camera, which captures RGB images for immediate detection of defects and classification of fabric quality using fault rate. This is an example of how AI can elevate the inspection in textiles, making it faster, more accurate, and more reliable than ever.