Automated rhinoceros detection in satellite imagery using deep learning

Area of Science:

• Conservation Technology
• Remote Sensing
• Wildlife Monitoring

Background:

• Rhinoceros populations face critical threats from poaching and habitat loss.
• Monitoring rhinoceroses in extensive, remote areas presents significant logistical challenges.

Purpose of the Study:

• To evaluate the effectiveness of very high-resolution satellite imagery and AI for detecting white rhinoceroses.
• To assess the impact of synthetic data augmentation on detection model performance.
• To determine if rhinoceroses can be differentiated from elephants in satellite imagery.

Main Methods:

• Utilized a YOLO-based object detection model (YOLOv12x) with satellite imagery (33-36 cm resolution).
• Tested synthetic imagery to enhance model performance and distinguish rhinoceroses from elephants.
• Evaluated the visual distinguishability of synthetic versus real rhinoceros images by human annotators.

Main Results:

• Achieved an average precision (AP) of 0.65 for rhinoceros detection.
• Synthetic data augmentation provided a marginal improvement in model performance.
• Demonstrated the potential for differentiating rhinoceroses from elephants in satellite data.

Conclusions:

• Satellite imagery combined with AI offers a viable method for monitoring rhinoceros populations.
• The developed open-access dataset supports advancements in wildlife detection technologies.
• This approach can bolster rhino conservation strategies, including anti-poaching and population assessments.