Use Anchor-Free Based Object Detectors to Detect Surface Defects

Many anchor-based object detectors have limitations in the detection process because of the presence of anchors. Therefore, in recent years, many researchers have shifted their focus to designing anchor-free object detectors. In this paper, we initially employ several common anchor-free detectors (CenterNet, FCOS, YOLOX-S, and YOLOV8-S) to detect surface defects in the hot-rolled strip datasets from Northeastern University (NEU-DET). Following experimental comparisons, YOLOV8-S demonstrates the best detection performance (mAP is 78.04% and the FPS is 8.45). Secondly, we utilize transfer learning to mitigate the overfitting problem that large models can induce on small datasets. Finally, we conduct a large number of experimental comparisons to verify the effectiveness of the model.