Exploration of CNN Architectures for Enhanced Fabric Defect Detection

Detecting fabric defects is crucial to ensure quality control in the textile industry. Convolutional neural networks (CNNs), are inspired by recent advancements in deep learning. Deep learning offers notable upgrades over traditional methods. It has been proven that they are highly effective in automating and enhancing the accuracy of error detection processes. This study provides a comparative overview of four well-known CNN architectures used specifically for fabric defect detection: AlexNet, VGG16, Inception V3, and ResNet50. The analysis examined each model's architecture, performance, and suitability for detecting fabric defects, highlighting their strengths and limitations. AlexNet provides a solid foundation with moderate accuracy, whereas VGG16 offers a deeper feature extraction at the expense of computational power. Inception V3 features an optimal balance between accuracy and speed, making it highly effective for real-time applications. ResNet50 with the remaining connections achieves the highest accuracy, especially when combined with advanced techniques such as Faster R-CNN. Despite progress, challenges remain, such as detecting different fabric patterns and small defects and suggesting future research approaches in the areas of hybrid models, data augmentation, and transfer learning. This report highlights the significant impact of advances in CNNs on fabric defect detection and provides insights into potential improvements to increase accuracy and efficiency further.