Scholarlink

<p>Birds are vital to maintaining ecological balance, yet their populations are increasingly threatened by habitat loss and human activity. An enhanced YOLOv8-based model, termed YOLO-BirdTracker, is proposed to improve monitoring efficiency and detection accuracy. A multi-scale edge information extraction module is incorporated to capture fine-grained bird features, while a dynamic feature pyramid with adaptive upsampling is employed to strengthen multi-scale feature fusion, and an interactive detection head is designed to improve task alignment and stability. Experimental results demonstrate that, compared with YOLOv8n, YOLO-BirdTracker reduces parameters by approximately 42% and computation by 40%. Performance improvements of 0.6% (mAP@50) and 0.8% (mAP@50–95) were observed on the private dataset, with corresponding increases of 0.4% and 1.3% on the public dataset, indicating consistent performance improvements across datasets. These results indicate that YOLO-BirdTracker achieves higher accuracy and efficiency, offering strong potential for practical bird monitoring applications.</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

A lightweight deep architecture with adaptive multi-scale features for efficient avian detection

  • 期刊论文

  • 发表日期：

    2026年3月12日

查看全文
  • Sheng Yu,
  • Qijun Xing,
  • Yanliang Huang,
  • Tingyu Zhao

摘要

Birds are vital to maintaining ecological balance, yet their populations are increasingly threatened by habitat loss and human activity. An enhanced YOLOv8-based model, termed YOLO-BirdTracker, is proposed to improve monitoring efficiency and detection accuracy. A multi-scale edge information extraction module is incorporated to capture fine-grained bird features, while a dynamic feature pyramid with adaptive upsampling is employed to strengthen multi-scale feature fusion, and an interactive detection head is designed to improve task alignment and stability. Experimental results demonstrate that, compared with YOLOv8n, YOLO-BirdTracker reduces parameters by approximately 42% and computation by 40%. Performance improvements of 0.6% (mAP@50) and 0.8% (mAP@50–95) were observed on the private dataset, with corresponding increases of 0.4% and 1.3% on the public dataset, indicating consistent performance improvements across datasets. These results indicate that YOLO-BirdTracker achieves higher accuracy and efficiency, offering strong potential for practical bird monitoring applications.