Abstract:Traffic sign detection is a critical technology in autonomous driving and intelligent transportation systems. Due to the small size of targets and severe occlusions, existing detection algorithms still suffer from missed detections and false positives in complex road scenes. To address these issues, this paper proposes FCD?YOLO, a traffic sign detection algorithm based on an improved YOLOv11n. Firstly, Frequency Dynamic Convolution (FDConv) is introduced, combined with Kernel Spatial Modulation and Frequency Band Modulation, to enhance the model''s ability to extract high?frequency details and complex textures. Secondly, a C2PSA?CAA module that integrates Context Anchor Attention is proposed in the backbone network to improve the contextual perception and discriminability of small?target features. Finally, a Dynamic Head (DyHead) is adopted to achieve dynamic fusion and adaptive adjustment of multi?level features, thereby enhancing the model''s detection capability for multi?scale and deformable objects. Experimental results demonstrate that FCD?YOLO achieves 70.8% mAP@0.5 on the TT100K dataset, outperforming the baseline YOLOv11n by 5.6 percentage points. Further validation on the CCTSDB dataset also confirms the algorithm''s strong generalization ability, showing its effectiveness in tackling traffic sign detection under complex traffic scenarios.