<p>Traffic sign detection and recognition are an important element of autonomous driving technology, particularly in scenes where the scale of signs varies greatly and the environment is complex. To confront these challenges, in this paper, we propose a traffic sign detection network based on multi-level feature fusion and attention, termed YOLO-TCS. Firstly, a triple feature integration module is designed in the neck network, aiming at enhancing the multi-scale feature fusion and small target detection capabilities. Secondly, a dual-view channel-position attention is incorporated to improve the interaction between channel and spatial feature information. Finally, a star aggregation network is designed to strengthen the interaction of spatial feature information. The experimental results show that the YOLO-TCS model achieves superior mean Average Precision (mAP@0.5) on the TT100K dataset compared to other tested models, reaching an mAP@0.5 of 80.9%, which is 2.3% higher than that of the benchmark model, YOLO11.And on the GTSDB dataset, YOLO-TCS outperforms YOLO11n by 0.7% points. Additionally, these results indicate that the improved method significantly enhances the detection capability for multi-scale traffic signs in complex scenes. The evaluation is currently limited to traffic sign detection on two datasets (TT100K, GTSDB), and that results may not generalize directly to other object detection tasks or real-world deployment scenarios without further validation. But, these preliminary improvements suggest the potential of our approach to enhance traffic sign recognition systems.</p>