SCS-YOLOv8: steel surface defect detection algorithm based on improved YOLOv8n

Le résumé fourni par la source

In order to solve the problems of insufficient target feature extraction capability and large number of model parameters faced in steel surface defect detection, this article proposes an innovative steel surface defect detection algorithm SCSYOLOv8, which is an improved version of the YOLOv8n algorithm. First, the lightweight module statblock is introduced into the backbone network to reduce the number of model parameters and improve detection speed; secondly, a multi-scale attention mechanism (EMA) is embedded in the neck network to enhance the feature extraction capability of the model; finally, in the detection stage A small target detection head with a size of 160×160×32 is added to improve the model’s detection performance of small-sized targets. Experimental results show that these improvements increase the mAP of the model on the NEU-DET data set by 2.4%, reduce the number of parameters by 4.7%, reduce the amount of calculation by 4.9%, and significantly improve the detection performance to meet the needs of industrial applications.

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