Scholarlink

Detecting small targets in remote sensing and UAV imagery remains challenging due to limited resolution and ambiguous features. To address these issues, we propose a multi-modal detection framework based on an enhanced YOLO architecture that fuses infrared and visible light data. We introduce a Multi-Modal Local-Global Fusion Block (MLGF) to extract globally normalized features and design a dedicated detection head to exploit cross-modal complementarity. Extensive experiments show that our method achieves 82.48% average accuracy on the VEDAI dataset and 81.6% on DroneVehicle, outperforming state-of-the-art methods. It also improves mAP@0.5 by 5.9% on the VisDrone dataset. These results demonstrate the effectiveness and practical potential of our approach for autonomous driving and intelligent transportation systems.

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

Intelligent Transportation Target Recognition Based on Multi-Modal Data Fusion in Remote Sensing and UAV Images

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Zhixing Wang,
  • Jikai Zhang,
  • Qiankai Xi,
  • Wentao Zhao,
  • Dongliang Guo

摘要

Detecting small targets in remote sensing and UAV imagery remains challenging due to limited resolution and ambiguous features. To address these issues, we propose a multi-modal detection framework based on an enhanced YOLO architecture that fuses infrared and visible light data. We introduce a Multi-Modal Local-Global Fusion Block (MLGF) to extract globally normalized features and design a dedicated detection head to exploit cross-modal complementarity. Extensive experiments show that our method achieves 82.48% average accuracy on the VEDAI dataset and 81.6% on DroneVehicle, outperforming state-of-the-art methods. It also improves mAP@0.5 by 5.9% on the VisDrone dataset. These results demonstrate the effectiveness and practical potential of our approach for autonomous driving and intelligent transportation systems.