The DFD-YOLO Algorithm for Detecting Small-Scale Surface Damage on Mining Steel Ropes

Abstract

To address the limitations of existing visual detection algorithms for mining wire ropes (MWR), including low detection accuracy, limited capability to recognize tiny defects, and high computational cost, we first employ NAFNet to mitigate motion blur in the input images. We then propose DFD-YOLO, an improved YOLOv11-based algorithm for small object detection. The DGEConv module combines depthwise separable convolution and Ghost convolution with an embedded efficient channel attention (ECA) mechanism to replace the standard convolutional layers in the original network. Furthermore, a multi-branch FEM-S module based on dilated convolution is introduced between the backbone and neck networks. A detection-oriented D-SwinIR module is embedded between the intermediate C3k2 layer and the FEM-S module. In addition, a dedicated small object detection head is incorporated into the network. Experimental results show that DFD-YOLO achieves an mAP@0.5 of 0.952 while maintaining low computational complexity. Compared with the original YOLOv11, the proposed model achieves a 12.5% relative improvement in mAP@0.5 and a 10.0% increase in inference speed. It requires only 2.5 million parameters and 6.2 GFLOPs. These results demonstrate that DFD-YOLO provides high detection accuracy and computational efficiency for MWR defect detection.