Comparative Study of YOLOv8 Nano and YOLOv8 Small for Screw Defect Detection

Le résumé fourni par la source

This study aims to develop a model to detect screw defects. The screws detected were single screws, multiple screws, and multiple stacked screws. The technology that can be applied to industries to improve product inspection performance is deep learning. The technology that can be applied to industries to improve product inspection performance is deep learning. The algorithm used is YOLOv8. This study presents an original approach by comparing the performance of YOLOv8n and YOLOv8s for screw defect detection using an Oriented Bounding Box (OBB) configuration. The findings indicate that the YOLOv8s model with Oriented Bounding Box (OBB) configuration outperforms YOLOv8n in detecting screw defects. YOLOv8s achieved a mAP50 of 0.832 and mAP50-95 of 0.609, higher than those of YOLOv8n, which are 0.785 and 0.595, respectively. Moreover, the use of training configurations such as early stopping and larger image sizes proved to enhance training efficiency without compromising model accuracy. The integration of OBB in YOLOv8s for precision industrial object detection is a novel contribution. A comprehensive evaluation was conducted using MAP metrics, confusion matrix, and F1-confidence curves.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.