Optimized Hybrid Features Enhance ML-Driven Defect Detection and Quality Control in Textiles

This work aims to investigate the utilization of ML algorithms to detect defects in textile production processes. Traditional textile inspections rely on manual methods, which are time-consuming, inconsistent, and prone to error. The research seeks to develop and evaluate MLalgorithms like SVM, CNN, Random Forest, and KNN to detect imperfections, including surface texture, color quality, and structural flaws. Special attention is paid to such factors as texture, color, and edge detection for their contribution to improving the accuracy of defect recognition. It is revealed that a higher accuracy rate of 96.7% is achievable when CNNs are applied with feature extraction methods, and they have better precision, recall, and F1-score measures. It also evaluates the enhancement in the quality control by presenting the difference in the production of defect- free fabric before and after the implementation of the ML-based system, where significant enhancements were observed. This study demonstrates a 10.3% increase in defect-free fabric rate after implementing the hybrid CNN-based ML model. These studies suggested that incorporating machine learning to improve the quality of textiles has the prospect of being a better approach than the traditional methods of detecting defects in textiles.