AI-driven fabric classification: real-time implementation for sustainable textile practices in industry 5.0

Abstract The growing demand for high-quality textile products requires efficient and intelligent fabric inspection systems. Convectional manual inspection procedure is not only time-consuming but also vulnerable to human errors. The proposed work in this study focused on optimizing a ResNet50 model with PSO (Particle Swarm Optimization) for real-time fabric classification that aligns well with the principles of Industry 5.0 and contributes meaningfully to several Sustainable Development Goals (SDGs). Using a custom dataset of fabric images (plain, satin, and twill) captured with a digital microscope, the proposed model leverages Particle Swarm Optimization (PSO) to fine-tune hyper parameters such as learning rate and momentum, enhancing the classification accuracy of ResNet50. This study claims the contribution of developing a comprehensive dataset providing a resource for training and assessing deep learning models for textile weave studies. The second contribution claimed is the integration of the PSO with ResNet50 model for hyper parameter adjustments, specifically improving learning rate and momentum in ResNet50. This integration significantly increases categorization precision, resulting in a highly efficient deep learning model tailored for textile applications. Our optimized model demonstrated a very high accuracy of 98.32% in correctly classifying the fabrics according to weave pattern. Thus, the proposed model could be used in real-time industrial applications for automatic fabric inspection and categorization. The proposed model also showed superior performance compared to CNN, ResNet201, Mobilenet, VGG16 and Resnet50. Overall, the proposed inspection mechanism Moreoever, it aligns well with Industry 5.0 principles and contributes to Sustainable Development Goals (SDGs) by offering a scalable, privacy-preserving, and intelligent textile quality control solution.