Using Deep Learning to Automate Orangutan Nest Detections on Aerial Images Collected With Drones.

Area of Science:

• Conservation Technology
• Artificial Intelligence in Ecology
• Wildlife Monitoring

Background:

• Traditional orangutan monitoring relies on costly ground-based line transect methods.
• Manual analysis of drone imagery for orangutan nests is time-consuming and expensive.
• Automated nest detection is crucial for enhancing the efficiency of drone-based wildlife surveys.

Purpose of the Study:

• To explore a deep learning method for automated detection of orangutan nests in aerial images.
• To improve the efficiency of orangutan population monitoring using drone technology.
• To evaluate the performance of a deep learning model for nest detection across different drone types.

Main Methods:

• Utilized the YOLO v10 deep learning model for automated nest detection.
• Trained the model on 868 images with 1568 annotated orangutan nests from Malaysia and Indonesia.
• Employed a transfer learning approach and tested the model on independent datasets from multirotor and fixed-wing drones.

Main Results:

• Achieved a mean Average Precision (mAP) of 0.831 during training.
• Demonstrated high precision (0.98) for both drone types in independent tests.
• Reported recall rates of 0.88 for multirotor and 0.71 for fixed-wing drones.

Conclusions:

• Deep learning models integrated with drone data significantly enhance orangutan monitoring efficiency.
• Automated nest detection reduces survey times compared to traditional methods.
• Further research is needed to improve model recall, especially for fixed-wing drone data, for accurate population trend analysis.