Identification of drone thermal signature by convolutional neural network

This paper presents the work on drone detection and identification using thermal infrared emission, which is primarily aimed towards night operation. Through both indoor and outdoor trials, the characteristics of the thermal signature emitted by a drone when captured by a drone detection system is examined, and their implications on a machine learning problem are studied. Thermal maps are processed through a YOLOv3 based CNN model to detect and generate a bounding box around the thermal signature of the drone. The presented approach also seeks to utilise the characteristics of drone motion for more effective drone detection through machine learning.