A Methodology for UAV Classification using Machine Learning and Full-Wave Electromagnetic Simulations

Abstract

Using micro-doppler signatures is an effective way to classify different types of UAVs, as well as other targets like birds. To generate these datasets, researchers used to conduct campaigns for radar drones’ measurements. However, these measurements are limited to the types of available drones, the used radar parameters, the targets’ range, and the environment these measurements are taken in. In this paper, a new method for simulating these types of datasets is introduced, this new method uses full-wave electromagnetic CAD tools. Radar simulations of five different types of real drones are presented. Using this method, researchers can simulate radar drones’ datasets using different types, sizes, and design materials of drones, they also can change the used radar parameters, detected range, targets speed, and rotors RPM for rotary drones. A 77 GHz FMCW simulated radar is used to generate the required dataset for classification purposes. Finally, a CNN algorithm is used to classify the five types of simulated drones, the accuracy of the used algorithm is better than 97%.