Scholarlink

<p>Lightweight object detection models are increasingly important for traffic sign recognition, particularly in real-time applications on resource-constrained devices. This paper presents TSS-YOLO (Timely Swift Small-YOLO), a compact detection framework designed to address the challenges of small object detection while maintaining computational efficiency. The proposed model incorporates three core components: a Feature Extraction and Space-to-Depth (FESD) module that improves the representation of fine-grained features with reduced computational cost; a Weighted Feature Fusion (WFF) module that facilitates more balanced multi-scale feature integration; and a detection head based on a Non-Maximum Suppression-Free (NMS-Free) inference strategy with the large-object branch removed, enhancing both efficiency and small-object performance. Experimental results on the TT100K and CCTSDB datasets show that TSS-YOLO achieves better accuracy than YOLOv10s, with reductions of 5.5 million parameters and 10.9 GFLOPs in computation. These results suggest that TSS-YOLO offers a practical balance between accuracy and efficiency for real-time traffic sign detection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

TSS-YOLO: A timely swift small traffic sign detection model

  • 期刊论文

  • 发表日期：

    2025年9月5日

查看全文
  • Shengsheng Tu,
  • Mingming Kong,
  • Chao Zhang

摘要

Lightweight object detection models are increasingly important for traffic sign recognition, particularly in real-time applications on resource-constrained devices. This paper presents TSS-YOLO (Timely Swift Small-YOLO), a compact detection framework designed to address the challenges of small object detection while maintaining computational efficiency. The proposed model incorporates three core components: a Feature Extraction and Space-to-Depth (FESD) module that improves the representation of fine-grained features with reduced computational cost; a Weighted Feature Fusion (WFF) module that facilitates more balanced multi-scale feature integration; and a detection head based on a Non-Maximum Suppression-Free (NMS-Free) inference strategy with the large-object branch removed, enhancing both efficiency and small-object performance. Experimental results on the TT100K and CCTSDB datasets show that TSS-YOLO achieves better accuracy than YOLOv10s, with reductions of 5.5 million parameters and 10.9 GFLOPs in computation. These results suggest that TSS-YOLO offers a practical balance between accuracy and efficiency for real-time traffic sign detection.