Scholarlink

<p>To address the challenging issues in bird detection within natural scenes, this study proposed a YOLO-Bird detection model based on an improved YOLO framework. Bird detection in natural environments faces critical limitations including insufficient visual features of small targets, complex background interference, diverse target morphology, and group occlusion, which lead to feature information loss during detection and severely constrain detection performance. To resolve these challenges, the model implemented a three-stage innovative design: First, a C2f-HLB lightweight feature extraction module was constructed, employing novel HybridLRLUBlock units to significantly enhance the capture capability of fine-grained morphological features of birds while reducing parameter volume. Second, a Detect-LQE intelligent detection head was designed, collaboratively optimized through a localization quality-aware scoring mechanism, small-target feature enhancement strategy, and dynamic false detection suppression algorithm, effectively improving detection robustness in complex scenarios. Additionally, an innovative EIEStem front-end processing module was introduced, establishing a multi-dimensional feature representation system through coordinated extraction of edge features and spatial information. Experimental validation demonstrated significant improvements in key performance metrics: Compared to baseline models, the proposed model achieved a 10.51% increase in mean average precision (mAP@0.5) on benchmark datasets, with precision and recall rates improving by 11.77% and 16.87% respectively, while reducing model parameters by 12.5%. This research provides an effective technical solution for small-target bird detection in natural scenes.</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

YOLO-Bird: Small Bird Object Detection in Natural Scenes

  • 期刊论文

  • 发表日期：

    2025年9月9日

查看全文
  • Rui Li,
  • Yunpeng Luo

摘要

To address the challenging issues in bird detection within natural scenes, this study proposed a YOLO-Bird detection model based on an improved YOLO framework. Bird detection in natural environments faces critical limitations including insufficient visual features of small targets, complex background interference, diverse target morphology, and group occlusion, which lead to feature information loss during detection and severely constrain detection performance. To resolve these challenges, the model implemented a three-stage innovative design: First, a C2f-HLB lightweight feature extraction module was constructed, employing novel HybridLRLUBlock units to significantly enhance the capture capability of fine-grained morphological features of birds while reducing parameter volume. Second, a Detect-LQE intelligent detection head was designed, collaboratively optimized through a localization quality-aware scoring mechanism, small-target feature enhancement strategy, and dynamic false detection suppression algorithm, effectively improving detection robustness in complex scenarios. Additionally, an innovative EIEStem front-end processing module was introduced, establishing a multi-dimensional feature representation system through coordinated extraction of edge features and spatial information. Experimental validation demonstrated significant improvements in key performance metrics: Compared to baseline models, the proposed model achieved a 10.51% increase in mean average precision (mAP@0.5) on benchmark datasets, with precision and recall rates improving by 11.77% and 16.87% respectively, while reducing model parameters by 12.5%. This research provides an effective technical solution for small-target bird detection in natural scenes.