FD-YOLO11: A Feature-Enhanced Deep Learning Model for Steel Surface Defect Detection
Zichen Dang, Xingshuo Wang
Abstract
Steel surface defect detection plays a critical quality control role in industrial manufacturing. However, the existing methods struggle to balance accuracy and efficiency, especially in complex defect environments, where significant challenges persist. To address this challenge, FD-YOLO11, which is a YOLO11-based deep learning model with enhanced feature extraction and fusion mechanisms for attaining improved detection performance, is proposed in this paper. To enhance the multiscale feature extraction process, self-calibrated convolution is integrated into the C3k2 module. Additionally, an FSPPF structure is designed to optimize the process of fusing local and global information, improving the defect recognition ability of the model in complex backgrounds. Furthermore, the DySample mechanism replaces the traditional upsampling method, effectively refining the feature fusion process and minimizing the semantic information loss. The NEU-DET and GC10-DET datasets are used for evaluation purposes, and the experimental results demonstrate that FD-YOLO11 achieves 4.6% and 4.0% [email protected] improvements over YOLO11s while maintaining an inference speed comparable to that of YOLO11s. This ensures a compromise between detection performance and computational efficiency. The model exhibits enhanced recognition capabilities and greater robustness in complex defect detection tasks. This research indicates that FD-YOLO11 provides a high-precision solution for metal surface defect detection, with broad application potential in intelligent manufacturing and industrial inspection scenarios.