Ghost-YOLO: A Lightweight Traffic Sign Detection Framework Via GhostNetV3

In this paper, we propose a lightweight traffic sign detection method based on improved YOLOv11s, aiming to solve the problems of accuracy, robustness and real-time of traffic sign detection in complex environments. Aiming at the deficiencies of existing models in small target detection and lightweighting, we optimize the network structure of YOLOv11 by adopting a lightweight network skeleton, GhostNetV3, as the backbone network, and introducing a depth separable convolution (DWConv) module to reduce the computational complexity. Through experimental validation on the TT100K dataset, the optimized model outperforms the original YOLOv11 model in terms of detection accuracy and speed, reaching 0.848 and 0.631 in the mAP50 and mAP50-95 metrics, respectively, while achieving 61.94 frames/sec in the average detection speed. The experimental results show that the method significantly reduces the computational complexity of the model while maintaining high accuracy, and is suitable for resource-constrained in-vehicle computing platforms.