Pages 68–71
Abstract
Research summary supplied by the authors
Accurate detection and grading of defects remain critical challenges in the leather industry, as ordinary manual inspection need labour, biased, and doubtful to inconsistency. Variations in lighting conditions, surface textures, and inspector expertise mostly lead to misclassification, by that negatively affecting material use and product value. Recent advances in artificial intelligence, mainly deep learning techniques, have shown significant potential for automated and objective defect inspection. Although previous studies have shown defect detection accuracies of about 92% to 97% under controlled imaging conditions, their industrial application is limited by less datasets and a very low range of defect categories, which reduce strength under real-world conditions. This study presents a deep learning-based automated framework for leather defect recognition and quality grading, targeting to fair detection accuracy and textue across different leather textures and finishing processes. The proposed framework covers enhanced data preprocessing, defect-specific data augmentation strategies, and an attention-based feature clarification mechanism to improve defect execution and classification performance. The developed approach is intended to support reliable, scalable, and industrially deployable inspection systems, favouring the understanding of near zero-error quality checking in leather manufacturing.