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

In this work, it has been proven that radar systems in general can be applied for detecting multi-copters using Deep Learning techniques. The radar micro-Doppler signatures induced due to micro movements inside the target (the rotating propellers) can be used to detect flying multi-copters. For higher classification accuracy and better reliability, a new technology applying Deep Learning can therefore be developed to detect multi-copters merely based on their radar micro-Doppler images. For the training purposes, a huge amount of data is needed. Building a new dataset however is really time consuming, which can be considered as one of the main downsides of Deep Learning. Also, powerful hardware (multiple GPUs) and much time are needed during the training process of a neural network. Data acquisition algorithm has been developed for the extraction of micro-Doppler signatures from the collected raw radar data. The micro-Doppler images were then labelled and fed to the neural network model for the purpose of applying supervised learning. After training the neural network model, the model was tested for classification accuracy to show a recognition accuracy of 99.4% dataset can therefore be applied for use in combination with different short-range radar sensor systems. Using the model, multi-copters up to a range of about 1.5m from the radar system will be detected.