Near real-time monitoring of wildlife animals and management policy evaluation using UAV with thermal sensors and pre-trained AI models

Keywords: AI models, Deep Learning, Wildlife Animal Monitoring, UAV, thermal sensors
Abstract Type: Paper Abstract

Abstract

Wildlife animals play a pivotal role in supporting a healthy ecosystem by creating vegetation clearings, promoting plant diversity, and maintaining prey-predator balance. Traditional field surveys for monitoring wildlife are laborious, intrusive, dangerous, expensive, and time-consuming, often requiring years to complete. Unmanned Aerial Vehicles (UAVs) equipped with thermal sensors offer a cost-effective and time-efficient alternative, capable of covering vast areas in a matter of hours instead of weeks. Thermal sensors can detect animals based on temperature contrast with their ambient environment, even under dense canopies. However, manually reviewing drone images remains time-consuming. To address this challenge, we propose employing pretrained AI models for automatic animal detection. These models were trained on extensive samples collected at different altitudes and with various color palettes to ensure their robustness. Images were transformed to an optimal palette and automatically georeferenced based on the GPS coordinates of image centers and the associated Gimbal attitudes (i.e. roll, yaw, and pitch). We developed and compared one-stage and two-stage deep learning models to assess their effectiveness and efficiency in detecting wildlife. The one-stage model based on Faster R-CNN achieved 95% accuracy but required 400 milliseconds per image for classification. In contrast, the two-stage model based on YOLOv8 obtained 92% accuracy and only 25 milliseconds per image, striking a balance between effectiveness and efficiency. Diverse animal densities, species composition, and habitat characteristics were observed in 10 survey zones distributed across the national park, buffer zones, and community forests. These observations provide valuable insights for management policy evaluation and subsequent amendment.