An Efficient Method for Monitoring Birds Based on Object Detection and Multi-Object Tracking Networks.

Area of Science:

• Computer Vision
• Ecological Monitoring
• Artificial Intelligence

Background:

• Traditional bird monitoring relies on manual field methods like point counts, which are often inefficient and prone to errors.
• Accurate species identification and population assessment are vital for effective bird conservation.
• Limitations in current methods hinder comprehensive bird population studies.

Purpose of the Study:

• To develop an efficient and accurate automated method for wetland bird monitoring.
• To enhance object detection and multi-object tracking for bird species identification and population counting.
• To improve upon existing bird monitoring techniques for better conservation outcomes.

Main Methods:

• Construction of manually annotated datasets for bird species detection (3,737 images) and multi-object tracking (11,139 images).
• Comparative analysis of state-of-the-art object detection networks, identifying YOLOv7 as most effective.
• Enhancement of YOLOv7 with GAM modules and Alpha-IoU loss for improved accuracy in detection and bounding box regression.
• Integration with DeepSORT for bird tracking, classification counting, and area-based flock distribution analysis.

Main Results:

• The enhanced YOLOv7 network achieved high accuracy, with mAP@0.5 reaching 0.951 and mAP@0.5:0.95 reaching 0.815.
• The proposed method successfully integrated detection, tracking, and classification for comprehensive bird monitoring.
• Accurate species-specific counting and flock distribution information were obtained.

Conclusions:

• The developed AI-based system offers an efficient and accurate solution for wetland bird monitoring.
• This automated approach overcomes the limitations of traditional manual methods in bird conservation.
• The method provides valuable data for understanding bird populations and their distribution, supporting conservation strategies.