A Deep Learning Approach for Multi-copter Detection using mm-Wave Radar Sensors: Application of Deep Learning for Multi-copter detection using radar micro-Doppler signatures

Abstract

The increasing number of affordable drones and their misuse calls for the need of new multi-copter detection technologies. Such new technologies shall enable the detection of drones flying in restricted regions, for instance in military zones. Therefore, multiple detection techniques have been developed, mainly using camera sensors and digital imaging. Since cameras however suffer the problems of inapplicability in bad weather and low light conditions, radar systems have been recently developed for multi-copter detection using various conventional detection algorithms. Radar systems enable collecting rotor-specific data. Due to the fact that all rotor-objects show similar characteristics in the Radar Doppler spectrogram, i.e. the so called Micro-Doppler signatures, Machine Learning classification techniques on radar collected data enable reaching better and more reliable detection results when compared to the conventional algorithms. This paper introduces a Deep-Learning-based technique that can be used to detect multi-copters. The main idea is making use of the micro-Doppler properties of multi-copter and applying Deep Learning approaches for the automation of the classification. Due to their rotating components, rotor-wing aircrafts induce distinct Radar micro-Doppler signatures. In this work, experimental measurements of radar micro-Doppler signatures for both cases: rotating wing-copters and various other non-rotating objects are detected using continuous wave (CW) Radar. Radar micro-Doppler signature images are collected, and then further processed and used for detecting the various multi-rotor objects. Detection is performed using a trained convolutional neural network (CNN).