Scholarlink

<p>Object detection in drone (UAV) aerial imagery remains challenging due to small target sizes, dense distributions, varied viewpoints, and complex backgrounds, leading to the limited effectiveness of existing algorithms. To address these issues, we propose FPA-YOLOv8s (Feature Patch-Aware YOLOv8-small), an enhanced lightweight detector based on YOLOv8s. FPA-YOLOv8s introduces three key improvements: (i) a C2f_PPA module that strengthens local and global feature extraction through Patch-Aware Attention and improves multi-scale feature fusion; (ii) a MRSF-Neck that incorporates high-resolution feature layers and reinforced feature aggregation to better adapt to complex scenes and multi-scale targets; and (iii) an Adaptive DySample Module (ADM) for efficient upsampling, leveraging lightweight channel attention and range constraints to enhance small-object perception. Extensive experiments on two public UAV datasets demonstrate the effectiveness of the proposed approach. On VisDrone2019, FPA-YOLOv8s achieves a 7% increase in detection accuracy compared to YOLOv8s with negligible parameter growth. On UAVDT, it obtains 3.6% higher mAP@50 and 4.1% higher mAP@50–95, while maintaining competitive model size and computational cost. These results confirm that FPA-YOLOv8s achieves a favorable balance between accuracy and efficiency, offering a practical solution for small-object detection in UAV applications. The source code is available at <a href="https://github.com/lc1768266731/FPA-Yolov8s/tree/main">https://github.com/lc1768266731/FPA-Yolov8s/tree/main</a>.</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

Fpa-yolov8s: an efficient small object detection algorithm for drone aerial imagery

  • 期刊论文

  • 发表日期：

    2025年10月25日

查看全文
  • Hai Nan,
  • Cheng Li

摘要

Object detection in drone (UAV) aerial imagery remains challenging due to small target sizes, dense distributions, varied viewpoints, and complex backgrounds, leading to the limited effectiveness of existing algorithms. To address these issues, we propose FPA-YOLOv8s (Feature Patch-Aware YOLOv8-small), an enhanced lightweight detector based on YOLOv8s. FPA-YOLOv8s introduces three key improvements: (i) a C2f_PPA module that strengthens local and global feature extraction through Patch-Aware Attention and improves multi-scale feature fusion; (ii) a MRSF-Neck that incorporates high-resolution feature layers and reinforced feature aggregation to better adapt to complex scenes and multi-scale targets; and (iii) an Adaptive DySample Module (ADM) for efficient upsampling, leveraging lightweight channel attention and range constraints to enhance small-object perception. Extensive experiments on two public UAV datasets demonstrate the effectiveness of the proposed approach. On VisDrone2019, FPA-YOLOv8s achieves a 7% increase in detection accuracy compared to YOLOv8s with negligible parameter growth. On UAVDT, it obtains 3.6% higher mAP@50 and 4.1% higher mAP@50–95, while maintaining competitive model size and computational cost. These results confirm that FPA-YOLOv8s achieves a favorable balance between accuracy and efficiency, offering a practical solution for small-object detection in UAV applications. The source code is available at https://github.com/lc1768266731/FPA-Yolov8s/tree/main.