YOLO-Slender: An Improved Multi-Scale Stacked Method for Automotive Wire Identification

To tackle challenges in automotive wire production, such as varying lighting, wire stacking, and diverse wire types, we propose YOLO-Slender, an improved model based on YOLO l l n-Seg for fast and accurate wire recognition and segmentation. YOLO-Slender enhances detection performance in complex environments by optimizing algorithms and adjusting network structures. The model introduces a C3k2-RDR module in the Backbone to combine deep and shallow feature information, improving large-scale feature response while reducing model size and increasing speed. In the Neck, a Slim-Neck module reduces computational complexity and inference time without sacrificing accuracy. A lightweight FGS segmentation head, using DEConv, improves detail capture and detection accuracy. Experimental results show YOLO-Slender outperforms YOLOlln-Seg by 6.81 % and 10.74% in mAP@0.5:0.95(B) and mAP@O.5:0.95(M), respectively, while reducing parameters and complexity by 28.53% and 14.7%. This model strikes a balance between lightweight deployment and accuracy, making it ideal for industrial devices with limited computational capacity.