Welfare-Oriented Non-Invasive Body Surface Temperature Monitoring of Endangered Pangolins via Infrared Thermography and Deep Learning

Pangolin rescue and captive breeding require reliable, animal-welfare-oriented health monitoring. Body surface temperature is an important health indicator, yet invasive measurement may cause stress and infection risks, whereas conventional non-contact systems often compromise accuracy to meet edge-computing constraints. This study develops a non-contact body surface temperature monitoring system integrating a Jetson Orin NX controller, an IPM384 infrared camera, and an SHT35D temperature-humidity sensor for environmental compensation. A lightweight YOLOv8n-PDTW detector is introduced for animal-region localization in infrared images affected by blurred edges and interference from metal food bowls. The model achieves 99.1% precision, 98.4% recall, and an mAP@0.5:0.95 of 88.1%, while reducing model size and computational cost by 38.7% and 37.8%, respectively, compared with YOLOv8n. Embedded deployment enables real-time inference at 23.77 frames/s. Validation against PT100 sensors yields a mean absolute error of 0.234 °C. During 17 days of continuous monitoring, the system captured consistent individual temperature patterns, establishing stable physiological baselines. This efficient, non-contact approach supports pangolin conservation and provides a technical reference for health monitoring in other endangered solitary mammals, although further cross-species validation is required.