Development of a Slow Loris Computer Vision Detection Model.

Area of Science:

• * Primate Conservation Biology
• * Computer Vision and Machine Learning
• * Wildlife Monitoring Technology

Background:

• * Accurate detection of slow lorises (Genus *Nycticebus*) is crucial for conservation, individual identification, and behavioral studies.
• * Traditional detection methods (manual observation, video review) are labor-intensive, time-consuming, and prone to errors.
• * Developing efficient, automated detection systems using computer vision is essential for endangered taxa like the slow loris.

Purpose of the Study:

• * To establish a novel target detection dataset for Bengal slow lorises (*N. bengalensis*) using monitoring videos.
• * To evaluate the effectiveness of two improved YOLOv5 network schemes for slow loris detection.
• * To identify an optimal computer vision model for automated slow loris detection in captive environments.

Main Methods:

• * Creation of a new dataset from surveillance videos of captive Bengal slow lorises in Chinese wildlife rescue centers.
• * Testing two modified YOLOv5 architectures: YOLOv5-CBAM + TC (incorporating attention mechanism and deconvolution) and YOLOv5-SD (adding a small object detection layer).
• * Quantitative evaluation of detection performance metrics including precision, recall, and mean average precision (mAP).

Main Results:

• * The YOLOv5-CBAM + TC model demonstrated significant improvements in detection performance compared to baseline methods.
• * Precision, recall, and mAP increased by 2.9%, 3.7%, and 3.5%, respectively, with a minimal increase in model size (0.6 MB).
• * The YOLOv5-SD model showed less improvement compared to YOLOv5-CBAM + TC.

Conclusions:

• * The YOLOv5-CBAM + TC model offers an effective and efficient solution for detecting individual slow lorises in captive settings.
• * This computer vision approach facilitates automated face and posture recognition, supporting conservation and research.
• * The developed dataset and model contribute to advancing automated monitoring techniques for endangered primates.