Scholarlink

<p>In UAV applications, detecting small targets is challenging due to the presence of dense targets and extreme lighting conditions. This paper proposes PPD-YOLO, an efficient model for robust detection in these challenging scenes. First, two specialized detection heads are designed to capture finer features and enhance performance for detecting small targets. Second, a PP-FPN network structure is proposed to replace the traditional PAFPN, aiming to retain and convey information better about small targets. Finally, the C2f-DW module is designed to replace part of the C2f module further enabling a lightweight model. On the VisDrone2019 dataset, PPD-YOLO achieves an mAP50 of 46.6% with only 5.7 million parameters, surpassing the baseline by a significant margin of 7.5 percentage points while reducing the parameter count by 5.4 million. On the TinyPerson dataset, it also attains an mAP50 of 22.7%. These results demonstrate that PPD-YOLO outperforms existing YOLO series methods for small object detection, excelling in both detection accuracy and model compactness.</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

PPD-YOLO: A lightweight model for small target detection in extreme light and dense scenes for UAVs

  • 期刊论文

  • 发表日期：

    2025年12月19日

查看全文
  • Jun Li,
  • Rongqing Tang,
  • Jianbing Yi,
  • Feng Cao,
  • Miaomiao Liang

摘要

In UAV applications, detecting small targets is challenging due to the presence of dense targets and extreme lighting conditions. This paper proposes PPD-YOLO, an efficient model for robust detection in these challenging scenes. First, two specialized detection heads are designed to capture finer features and enhance performance for detecting small targets. Second, a PP-FPN network structure is proposed to replace the traditional PAFPN, aiming to retain and convey information better about small targets. Finally, the C2f-DW module is designed to replace part of the C2f module further enabling a lightweight model. On the VisDrone2019 dataset, PPD-YOLO achieves an mAP50 of 46.6% with only 5.7 million parameters, surpassing the baseline by a significant margin of 7.5 percentage points while reducing the parameter count by 5.4 million. On the TinyPerson dataset, it also attains an mAP50 of 22.7%. These results demonstrate that PPD-YOLO outperforms existing YOLO series methods for small object detection, excelling in both detection accuracy and model compactness.