A Screw Surface Defect Detection Model Based on YOLO11- DySample

Abstract

 As critical components in fastening systems, screws play an essential role in structural connection and load transmission, where surface quality directly affects product safety and reliability. To achieve efficient and accurate detection of various surface defects on screws, this paper proposes a detection model based on the YOLO11-DySample algorithm. The proposed method adopts YOLO11 as the backbone detection framework and integrates the lightweight and efficient DySample dynamic upsampling module, which enhances feature reconstruction and improves the perception of small defects. Experimental results on a screw defect dataset demonstrate that the proposed model outperforms other benchmark algorithms in several key performance metrics, achieving a mAP50 of 0.991, mAP50-95 of 0.859, precision of 0.996, and recall of 0.994, indicating excellent accuracy and robustness. Further analysis of loss curves and precision-recall curves confirms the model’s convergence and generalization capability. Visual inspection results show that the model can effectively identify typical defects such as scratches and dents, demonstrating strong potential for practical industrial deployment.