Glass Wool Defect Detection Using an Improved YOLOv5

Yizhou Jin, Yu Lu, Gang Zhou, Qingjie Liu, Yunhong Wang; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023, pp. 4385-4394

Glass wool defect detection is a key part of product quality assessment in the glass wool production process, yet few studies have been reported in this area. We propose a glass wool defect dataset named GWD, and also use the YOLOv5s model embedded in the GSConv and the CBAM modules for both Gap and Glueless defects in this dataset. The experimental results show that the performance of the improved YOLOv5s on the GWD dataset is superior to other compared methods and achieves a relatively good level on other publicly available datasets. Compared to the vanilla YOLOv5s, the mAP50 increased by 3.7% to 84.1%, the recall increased by 4.2% to 84.4%, and the number of parameters decreased by 0.42 MB to 6.27 MB of the improved YOLOv5s model on the GWD dataset. Speed-wisely, the improved YOLOv5s achieves a 97 FPS on RTX 2080Ti, thus making it practical to be applied in the industry of glass wool defect detection. The research on the GWD dataset is likely to contribute to breakthroughs in research on other datasets of the same type as well. The GWD dataset can be obtained by contacting us via email.