Radar-Based Drone Detection Using Complex-Valued Convolutional Neural Network

With an unprecedented growth in the number of commercially available drones, the detection of drones is becoming increasingly essential. Deep learning-based convolutional neural network (CNN) models utilizing micro-Doppler signatures, are being widely used for drone detection applications. Radar returns from a drone and its corresponding micro-Doppler signatures are often complex-valued. However, the CNNs only consider the magnitude component of the micro-Doppler signatures while ignoring the phase component. This phase component contains essential information that can supplement the magnitude for enhanced drone detection. Thus, this paper proposes a novel complex-valued CNN that considers the magnitude and phase component of the radar returns. This paper also investigates the performance of the proposed model with radar returns of different sampling frequency and duration. A comparative analysis of the performance of the proposed model in the presence of noise is also presented. The proposed complex-valued CNN model achieved the highest detection accuracy of 93.80% when the radar returns were sampled at 16000 Hz and for duration of 0.01s. This shows that the proposed model can successfully detect drones that appear in the radar for an extremely short interval of time.