Scholarlink

<p>Object detection in unmanned aerial vehicle (UAV) imagery poses unique challenges due to dense small objects, large scale variations, and complex backgrounds. To address these issues, we propose ER-YOLO, an enhanced variant of YOLO11 tailored for UAV scenarios. ER-YOLO incorporates a Progressive Window Transformer (PWT) module in the backbone to capture both local and global contextual relationships, strengthening feature extraction for small and densely distributed targets. A Bidirectional Efficient Feature Pyramid Network (BEFPN) is further designed to integrate high-resolution and deep semantic features, effectively improving multi-scale representation. Within the BEFPN, a Cross-Spatial Feature Fusion (CSFF) module adaptively calibrates spatial and channel interactions through a dual-path attention mechanism, ensuring efficient and discriminative feature aggregation. Extensive experiments on standard UAV benchmarks demonstrate that ER-YOLO achieves state-of-the-art accuracy while maintaining high inference efficiency. Specifically, ER-YOLOs attains mAP scores of 42.3% on VisDrone and 88.9% on UAVDT, offering a robust solution for UAV object 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

ER-YOLO: enhanced information representation network for UAV-based small-object detection

  • 期刊论文

  • 发表日期：

    2026年4月1日

查看全文
  • Junbao Wu,
  • Hao Meng,
  • Fang Gao,
  • Lipeng Wang,
  • Zhe Lu,
  • Ming Yuan

摘要

Object detection in unmanned aerial vehicle (UAV) imagery poses unique challenges due to dense small objects, large scale variations, and complex backgrounds. To address these issues, we propose ER-YOLO, an enhanced variant of YOLO11 tailored for UAV scenarios. ER-YOLO incorporates a Progressive Window Transformer (PWT) module in the backbone to capture both local and global contextual relationships, strengthening feature extraction for small and densely distributed targets. A Bidirectional Efficient Feature Pyramid Network (BEFPN) is further designed to integrate high-resolution and deep semantic features, effectively improving multi-scale representation. Within the BEFPN, a Cross-Spatial Feature Fusion (CSFF) module adaptively calibrates spatial and channel interactions through a dual-path attention mechanism, ensuring efficient and discriminative feature aggregation. Extensive experiments on standard UAV benchmarks demonstrate that ER-YOLO achieves state-of-the-art accuracy while maintaining high inference efficiency. Specifically, ER-YOLOs attains mAP scores of 42.3% on VisDrone and 88.9% on UAVDT, offering a robust solution for UAV object detection.