Micro-Doppler spectrums recognition of aerial targets based on convolutional neural networks

Unmanned aerial vehicles (UAVs) have been widely used in many scenes in our lives, such as photography, emergency rescue, express delivery, and so on. Because of its small size, moving mobility and low cost, the UAV has become a major threat to soldiers in modern warfare. Therefore, the detection and recognition of UAVs are important in modern radar technology. This paper establishes the mathematical models of radar echo signals for different aerial targets such as UAVs, airplanes, birds, etc. Then we utilize the short-time Fourier transform (STFT) to process echo signals, and we will obtain the micro-Doppler spectrums of targets. The spectrums reflect their unique micro-Doppler features caused by their physical structures and micro-motion characteristics, so we can use that to recognize different targets more efficiently. In this paper, a convolutional neural network based on micro-Doppler spectrums (DS-CNN) is trained to realize the recognition. The network mainly consists of 5 convolutional sequences and is aimed to classify 6 different micro-Doppler spectrums of targets. The simulation results show that the DS-CNN has a high recognition accuracy even in 10dB noise and an appropriate run time to recognize each spectrum.