Micro-Doppler Effect and Determination of Rotor Blades by Deep Neural Networks

The paper deals with the analysis of simulated data, where thousands of samples of reflections from a radar target, a helicopter, with propellers were simulated. Simulations were performed for helicopters with 3, 4, 6, and 8 propeller blades. Data collection and evaluation were focused on the measurement of the Doppler Effect, specifically the Micro-Doppler effect for the rotating propeller section. The simulations have been divided into several sections for all types of helicopters differing in the number of propellers. The most considered was the change of Radar Cross Section (RCS), but changes in helicopter movement speed, changes in helicopter position relative to the radar, and changes in helicopter rotation speed have been considered as well. Moreover, a simulation of the change in radar carrier frequency across the microwave band was performed and the changes and effects on the Micro-Doppler measurement data were studied. However, the main task of this paper was to determine the number of propeller blades from any simulated signal sample with parameters corresponding to the Micro-Doppler, which was successfully done. Simulated data has been used to train a deep learning network to classify the number of propeller blades on a randomly selected measured/simulated sample. To detect the number of rotors, we have chosen to use Convolutional Neural Networks (CNN), which achieve good results for object recognition from images.