Fabric Defect Detection Using Deep Learning

The identification of fabric imperfections presents a considerable obstacle in the fabric manufacturing sector, owing to the complex configurations and diverse array of defects that may exist. Traditionally, visual inspections by human operators have been employed to identify such flaws, but this approach is both slow and prone to inaccuracies. In order to tackle these problems, this study investigates the benefits of employing Faster R-CNN and a range of YOLOv5 algorithms to accurately identify the fabric flaws. The deep learning object detection algorithms like YOLOv5s and Faster R-CNN have undergone rapid development and have been successfully applied in various industries, demonstrating strong performance. This paper proposes the use of the YOLOv5s algorithm, which exhibits superior accuracy in detecting fabric defects. Experimental results indicate that, compared to both YOLOv5s and Faster R-CNN, the YOLOv5s algorithm achieves higher accuracy in terms of mean Average Precision (mAP) and faster performance in terms of Frames Per Second (FPS). The FPS of YOLOv5s and faster RCNN is calculated and compared. From the results, it is inferred that YOLOv5s algorithm can accurately and rapidly identify the location of defects, making it suitable for real-time applications in the defect detection industry.