SlimEMA-YOLOv8: Enhanced Traffic Sign Recognition for Autonomous Driving Using EMA and Slim-Neck in YOLOv8

Traffic sign recognition is a critical component of visual tasks in autonomous driving. Addressing the issues of small traffic sign regions and frequent occlusions in real-world images captured by cameras, this paper proposes integrating the EMA module into the Neck part of YOLOv8 to enhance the model's ability to extract and fuse features. Additionally, we replace the original convolution and C2f modules with GSConv and VoVGSCSP modules from the Slim-neck to reduce model parameters while further improving detection accuracy. Experimental results on the Road Sign dataset show that compared to YOLOv8n, our model achieves a 1.0 percentage point improvement in accuracy, significantly enhancing traffic sign detection performance in complex environments.