Efficient detection of multiscale defects on metal surfaces with improved YOLOv5

In the process of metal production and manufacturing, the surface of the metal will produce defects of different scales, which will seriously affect the quality and performance of the metal, so it is very necessary to detect the defects on the metal surface. The traditional target detection method has the problem of high missing rate and low detection accuracy when detecting multiscale defects on metal surface, so it can not realize the efficient identification of different scale defects on metal surface. To solve these problems, a multiscale defect detection model S6SC-YOLOv5 based on YOLOv5 is proposed in this paper. Firstly, the neck structure was modified to an S6 feature fusion structure to improve the recognition ability of multiscale defects on metal surfaces. Secondly, the neck network is replaced by Slim-Neck to improve the fusion ability of multiscale defect features on metal surfaces. Finally, the up-sampling operator CARAFE module is used to increase the receptive field of the network. The experimental results show that S6SC-YOLOv5 is superior to YOLOv5s in overall performance. The mean average precision (mAP) of the S6SC-YOLOv5 model in the aluminum and NEU-DET data sets is 91.2% and 89.3%, respectively, which is 3.7% and 6.5% higher than that of YOLOv5s. It provides a new solution for multiscale defect detection on metal surfaces. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.