YOLO-WildASM: An Object Detection Algorithm for Protected Wildlife

Area of Science:

• Ecology
• Computer Vision
• Artificial Intelligence

Background:

• Accurate wildlife identification is vital for ecological conservation and species monitoring.
• Conventional object detection methods struggle with challenges like small targets and occlusions in natural habitats.
• Developing robust detection frameworks is essential for effective wildlife management.

Purpose of the Study:

• To develop and evaluate an advanced deep learning-based detection framework for wildlife in natural environments.
• To address limitations of existing methods in detecting small and occluded wildlife targets.
• To enhance the accuracy and efficiency of wildlife monitoring systems.

Main Methods:

• Construction of a custom dataset with over 8000 images of 10 protected wildlife species.
• Proposal of the YOLO-WildASM framework, enhancing YOLOv8 with a P2 detection layer, multi-head self-attention (MHSA), and bidirectional feature pyramid network (BiFPN).
• Comparative analysis against YOLOv8 and other state-of-the-art models using mAP50 metric.

Main Results:

• YOLO-WildASM achieved a mAP50 of 94.1% on the custom wildlife dataset, outperforming YOLOv8 by 2.8% and YOLOv12 (92.2%).
• The model demonstrated superior performance compared to baseline and other state-of-the-art detection models.
• Ablation and generalization experiments confirmed enhanced performance and adaptability for multi-scale wildlife detection.

Conclusions:

• The proposed YOLO-WildASM framework offers an efficient and robust solution for wildlife detection in complex ecosystems.
• This deep learning approach significantly improves the accuracy of wildlife monitoring and conservation efforts.
• The study highlights the potential of advanced AI techniques for addressing ecological challenges.