Birds-YOLO: A Bird Detection Model for Dongting Lake Based on Modified YOLOv11

Area of Science:

• Computer Vision
• Machine Learning
• Ecological Monitoring

Background:

• Bird detection in complex environments like Dongting Lake faces challenges from background interference, varied target sizes, and high species diversity.
• Existing models struggle with robustness and accurate detection of diverse bird species in cluttered natural habitats.

Purpose of the Study:

• To develop an enhanced bird detection model, Birds-YOLO, for improved accuracy and robustness in challenging ecological settings.
• To address limitations in current bird detection systems concerning environmental complexity and target variability.

Main Methods:

• The proposed Birds-YOLO model is based on the YOLOv11 framework, incorporating an EMA mechanism for global and local feature capture.
• An improved RepNCSPELAN4-ECO module with depthwise separable convolutions and adaptive channel compression is used for enhanced feature extraction and multi-scale fusion.
• The network's neck component is redesigned with lightweight GSConv convolution for computational efficiency without sacrificing accuracy.

Main Results:

• Birds-YOLO achieved a 5.0% improvement in recall and a 3.5% increase in mAP@0.5 on the CUB200-2011 dataset compared to the baseline YOLOv11n.
• On the DTH-Birds dataset, the model showed a 3.7% increase in precision, 3.7% in recall, and 2.6% in mAP@0.5.
• Ablation studies and comparative experiments validated the model's generalization ability and robustness in complex natural environments.

Conclusions:

• Birds-YOLO significantly enhances bird detection performance, demonstrating superior robustness and accuracy in cluttered environments.
• The model's effective feature extraction and fusion capabilities make it suitable for practical deployment in ecological monitoring and conservation efforts.
• The proposed enhancements offer a promising solution for accurate bird identification in challenging real-world scenarios.