Unsupervised Domain Adaptation across FMCW Radar Configurations Using Margin Disparity Discrepancy

TL;DR

This paper adapts Margin Disparity Discrepancy, a computer vision technique, to unsupervised domain adaptation in radar-based human activity classification, reducing the need for labeled data across different radar configurations.

Contribution

It extends Margin Disparity Discrepancy to radar data, enabling effective unsupervised domain adaptation for deep-learning radar applications.

Findings

Achieved comparable accuracy to few-shot supervised methods

Extended the technique successfully to radar data

Demonstrated effectiveness across different radar configurations

Abstract

Commercial radar sensing is gaining relevance and machine learning algorithms constitute one of the key components that are enabling the spread of this radio technology into areas like surveillance or healthcare. However, radar datasets are still scarce and generalization cannot be yet achieved for all radar systems, environment conditions or design parameters. A certain degree of fine tuning is, therefore, usually required to deploy machine-learning-enabled radar applications. In this work, we consider the problem of unsupervised domain adaptation across radar configurations in the context of deep-learning human activity classification using frequency-modulated continuous-wave. For that, we focus on the theory-inspired technique of Margin Disparity Discrepancy, which has already been proved successful in the area of computer vision. Our experiments extend this technique to radar data, achieving a comparable accuracy to fewshot supervised approaches for the same classification problem.