BGC-YOLO A Feature Fusion-Based Algorithm for Traffic Sign Detection

Abstract

With the development of intelligent transportation systems, the automatic detection of traffic signs has become a key task in assisted driving and unmanned driving perception systems. In view of the problem that traffic signs are small in scale in images and their accuracy is affected by complex environments, this paper constructs a BGC-YOLO target detection algorithm based on YOLOv11. First, by introducing the bidirectional feature fusion structure BiFPN, the interactive expression of multi-scale features is enhanced. Secondly, the global-local spatial attention mechanism GLSA is combined to improve the model's perception of detail information and contextual semantics. Finally, the content-aware upsampling module CARAFE is used to optimize the feature reconstruction process and effectively retain the key information of small targets. The experimental results on the CCTSDB2021 traffic sign dataset show that the improved model achieves a good balance between accuracy and efficiency, with an increase of 1.4% in mAP@0.5 compared to the original model, and maintains a low computational overhead, which is practical.