Micro-Doppler Spectrogram Denoising Based on Generative Adversarial Network

In applications based on radar sensors, target movement can be analyzed using micro-Doppler spectrogram, which is a time-frequency representation of micro-Doppler signature. However, the noise in spectrogram brings difficulties for applications. Conventional denoising algorithms are not specific to micro- Doppler data, they could only deal with a fixed level of noise and fail to effectively denoise under low Signal-to-Noise Ratio (SNR) circumstances. To overcome the drawbacks, we propose a method based on Generative Adversarial Network (GAN) to remove the noise in micro-Doppler spectrograms. Our method is applicable to a wide range of noise intensity, for which can be called a blind denoiser. We verify the effectiveness of the proposed method on both simulated and measured data, experimental results compared with competing algorithms demonstrate the superiority of our method.