Towards Adversarial Denoising of Radar Micro-Doppler Signatures

Abstract

Generative Adversarial Networks (GANs) are considered the state-of-the-art in the field of image generation. They learn the joint distribution of the training data and attempt to generate new data samples in high dimensional space following the same distribution as the input. Recent improvements in GANs opened the field to many other computer vision applications based on improving and changing the characteristics of the input image to follow some given training requirements. In this paper, we propose a novel technique for the denoising and reconstruction of the micro-Doppler (µ-D) spectra of walking humans based on GANs. Two sets of experiments were collected on 22 subjects walking on a treadmill at an intermediate velocity using a 25 GHz CW radar. In one set, a clean µ-D spectrum is collected for each subject by placing the radar at a close distance to the subject. In the other set, variations are introduced in the experiment setup to introduce different noise and clutter effects on the spectrum by changing the distance and placing reflective objects between the radar and the target. Synthetic paired noisy and noise-free spectra were used for training, while validation was carried out on the real noisy measured data. Finally, qualitative and quantitative comparison with other classical radar denoising approaches in the literature demonstrated the proposed GANs framework is better and more robust to different noise levels.

Automotive Radar — From First Efforts to Future Systems

“…Moving beyond single-radar scenarios, different authors started exploring the use of machine learning algorithms to detect and mitigate interference between various radar types. The scope of work ranges from classifying the interference type [132] to interference mitigation using denoising neural networks such as convolutional neural network [133], autoencoder [134], [135], or recurrent neural networks [136].…”

Machine Learning and Deep Learning Techniques for Colocated MIMO Radars: A Tutorial Overview

“…3) Generative adversarial networks are able to generate synthetic images on the basis of a set of noisy input data. This property can be exploited in radar systems to de-noise images [91] or to detect abnormalities.…”

“…In the technical literature about this application, the following two methods are exploited to extract relevant features from spectrograms: a) manual extraction of handcrafted features; b) automatic extraction of features based on a data-driven approach. Machine learning methods exploiting manual extraction of features have been investigated in [95]- [103], whereas the automatic extraction of features from micro-Doppler signatures or spectrograms through DL methods has been proposed in [91], [92], [94], [104]- [106]. It is important to keep in mind that: c) [10], [112]- [117] d) [15], [118]- [131] Table IV: Specific learning methods investigated in various manuscripts that concern the four application fields considered in Section VI.…”

“…The use of a deep GAN, based on a convolutional encoder-decoder structure, has been proposed in ref. [91] for this application. The performance results obtained in this case evidence that this network does not affect the relevant components of micro-Doppler spectra and is able to outperform other classic de-noising techniques commonly used for the suppression of background noise.…”

Synthesizing Foot and Ankle Kinematic Characteristics for Lateral Collateral Ligament Injuries Detection

“…Doherty et al presented a GANs-trained unsupervised model using micro-Doppler spectrograms for gait detection [33]. Abdulatif et al used GANs to learn the joint distribution of the training data for denoising and reconstructing the micro-Doppler spectrograms in human walking [34]. Erol et al proposed multi-branch GANs integrating the kinematic analysis of the micro-Doppler signature envelope to generate abnormal gait samples [35].…”