Learned Micro-Doppler Representations for Targets Classification Based on Spectrogram Images

This paper proposes a new approach for classifying ground moving targets captured by pulsed Doppler radar. Radar echo signals express the Doppler effect that moving targets produce. A learned feature representation extracted from spectrogram images using a transfer learning paradigm is proposed. A discrimination power analysis that derives highly discriminative features used to train a robust classifier was conducted. The extensive experiments performed on the public RadEch dataset show that the proposed method produces a significant boost in performance when compared to other state-of-the-art methods.