Defects Classification on Garment Fabrics and Application of Artificial Intelligence to Detect Defects During Fabric Inspection

This report presents the study results of defects classification that often appear on garment fabrics and a defects identification model made by artificial intelligence. The classification system of fabric defects is created for the artificial intelligence recognition model to ensure a comprehensive and general overview in many aspects: the type of garment fabric, the stage and causes of the defects, objects and the degree of defects. The recognition model for identification of fabric defects is built on the experimental method using the YOLO algorithm version 5. The data for model training and testing are selected and made by the algorithm's requirements, including 261 samples of fabric defect on 2D images of fabrics that have different characteristics such as weaves, textures, thicknesses, weights, yarn densities, and compositions. The testing results show that the model achieves an average of over 65% accuracy depending on the difficulty of the detected defect types. This model is the initial basis for designing an automatic detection system of the defects for many different types of fabrics.