Micro-Doppler Signal Representation for Drone Classification by Deep Learning

Abstract

There are numerous formats which represent the micro-Doppler signature. Our goal is to determine which one is the most adapted to classify small UAV (Unmanned Aerial Vehicules) with Deep Learning. To achieve this goal, we compare drone classification results with the different micro-Doppler signatures for a given neural network. This comparison has been performed on data obtained during a radar measurement campaign. We evaluate the classification performance in function of different use conditions we identified with a given neural network. According to the experiments conducted, the recommended format is a spectrum issued from long observations as its classification results are better for most criteria.

Drone Detection Sensor With Continuous 2.4 GHz ISM Band Coverage Based on Cost-Effective SDR Platform

“…where k is the number of elements in each subset. Next, the first-order difference function is applied to (17) in range m = 1, 2, . .…”

“…However, ever smaller drone sizes, low altitude operation, low flying speed, and the urban environment highly reduce the radar detection accuracy. Considering these limitations, the Authors in [17] investigate the micro-Doppler signatures induced by the micro-motion of small drones. While effectiveness against drones performing autonomous tasks is a valuable asset, high power active radar beam, needed for long-range detection, usually limits its use to licensed users.…”

“…A sequential calculation of between class variance is performed according to (17) in the second stage of operation to obtain the optimum value of the threshold. All of the arithmetical calculations are done using the fixed point format to save resources and better control the system implementation's maximum frequency.…”

Drone Characterization using Helicopter Rotation Modulation (HERM) Line Signatures

“…This is one of the reasons why HERM line analysis is not very well explored. Recently, Gérard et al [28] compared the efficiency of using different micro-Doppler representations, such as the the HERM line spectrum, the cepstrum, CVD, the spectrogram to classify five different drones using a CNN. The same integration time (0.3 s) was used for all these representations and it was found that using the HERM line spectrum performed the best and was the most resistant against noise.…”

“…Drone classification using micro-Doppler is typically done by training classifiers using different micro-Doppler representation like the cepstrum, cadence velocity diagram and the short windowed STFT spectrogram. Recently, it was shown that using the HERM line spectrum to classify drones was superior to other micro-Doppler representations such as the cepstrum, CVD, and the short-window spectrogram given the same integration time[28]. However, using a smaller window of data of 0.03 s rather than 0.3 s, the classification accuracy dropped from 98% to 93%.…”