Computer Vision based Automated Fabric Classification for Textile Industries

Manual inspection remains a major bottleneck in textile industries due to its subjectivity, inconsistency, and high labor dependency. Manual inspection is expensive, time consuming and lacks scalability and productivity. This research presents a deep learning–based automated stain and defect classification system using five convolutional neural network (CNN) architectures: MobileNetV2, ResNet50, GoogLeNet (Inception-v1), EfficientNetB0, and a custom CNN. A balanced dataset of 136 textile images was preprocessed and augmented to enhance feature variability. Experimental evaluation demonstrates that GoogLeNet achieves the highest validation accuracy of 96%, while MobileNetV2 provides a strong efficiency– accuracy balance suitable for real-time deployment. Training curves for accuracy and loss across all models, along with a comparative study, are presented. While maintaining performance on par with advanced existing models, the proposed approach significantly improves prediction speed, supporting its application in real-time fabric inspection. The system shows promising potential for integration into industrial inspection pipelines to reduce human error and improve fabric quality control.