Traffic Target Detection Method Based on Improved YOLOv8s

Aiming at the challenges of insufficient accuracy and inefficient deployment in traffic target detection, this paper proposes a lightweight model, YOLOv8s-MGB, based on YOLOv8s. The model incorporates three optimization strategies: introducing a bidirectional feature pyramid network (BiFPN) to enhance multi-scale feature fusion, adopting a mixed local channel attention module (MLCA) to improve focus on key regions, and applying Ghost convolution to efficiently compress model parameters and reduce computational cost. Experiments on the BDD100K dataset demonstrate that YOLOv8s-MGB achieves a 98.2% mAP—representing a 3.8 percentage point improvement over the original YOLOv8s—while reducing parameters by 45.2% and inference time to 3.0 ms. These results highlight the model's high accuracy, efficiency, and strong potential for edge deployment in real-world traffic scenarios.