A Comparative Study of YOLOv5 and YOLOv8 for Appearance Defect Detection in Polyester Fiber Yarn Packages

Polyester fiber yarn packages often exhibit various appearance defects during production, including six classes: hairiness, hairiness clusters, broken thread, broken pipe, oil stains, and dust impurities. Failure to detect these defects in time during the production process can lead to significant economic losses. This highlights the importance of real-time detection of yarn packages defects during production. Currently, there are some problems with methods that rely mainly on manual testing, such as high false positive rates and poor timeliness. In order to address the increasing complexity of defect detection in the appearance of yarn packages, in this study, 2480 datasets of size 2448*512 were first collected on the production line, and defect detection was explored using the popular YOLOv5 model and the newer YOLOv8 model. The performance of the model at different scales was also evaluated. The results show that YOLOv8 is better than YOLOv5 in this task and is more suitable for industrial applications. Our study provides an efficient and accurate quality control method for the production of Polyester fiber yarn packages.