Machine Learning for UAV Classification Employing Mechanical Control Information

Abstract

Range-Doppler images are widely used to classify different types of Unmanned Air Vehicles (UAVs) because each UAV has a unique range-Doppler signature. However, a UAV's range-Doppler signature depends on its movement mechanism. This is why a classifier's accuracy would be degraded if the effect of the mechanical control system of UAVs wasn't taken into consideration, which may lead to a non-unique signature of a UAV while in-flight. In this paper, a full-wave electromagnetic CAD tool is used to investigate the effect of the control systems of two quadcopters, a hexacopter, and a helicopter UAVs on their range-Doppler signatures. A Mechanical Control-Based Machine Learning (MCML) algorithm is introduced to classify the four UAVs. Different Machine Learning (ML) algorithms were applied to the generated datasets that considered the mechanical control information of UAVs. The Convolutional Neural Networks (CNN) algorithms provided robust performance reaching an accuracy of higher than 90%.