Using YOLO Object Detection to Identify Hare and Roe Deer in Thermal Aerial Video Footage—Possible Future Applications in Real-Time Automatic Drone Surveillance and Wildlife Monitoring

Abstract

Wildlife monitoring can be time-consuming and expensive, but the fast-developing technologies of uncrewed aerial vehicles, sensors, and machine learning pave the way for automated monitoring. In this study, we trained YOLOv5 neural networks to detect points of interest, hare (Lepus europaeus), and roe deer (Capreolus capreolus) in thermal aerial footage and proposed a method to manually assess the parameter mean average precision (mAP) compared to the number of actual false positive and false negative detections in a subsample. This showed that a mAP close to 1 for a trained model does not necessarily mean perfect detection and provided a method to gain insights into the parameters affecting the trained models’ precision. Furthermore, we provided a basic, conceptual algorithm for implementing real-time object detection in uncrewed aircraft systems equipped with thermal sensors, high zoom capabilities, and a laser rangefinder. Real-time object detection is becoming an invaluable complementary tool for the monitoring of cryptic and nocturnal animals with the use of thermal sensors.