Human Motion Recognition With Limited Radar Micro-Doppler Signatures

The performance of deep learning (DL) algorithms for radar-based human motion recognition (HMR) is hindered by the diversity and volume available training data. In this article, to tackle issue insufficient data HMR, we propose an instance-based transfer (ITL) method with limited radar micro-Doppler (MD) signatures, alleviating burden collecting annotating a large number samples. ITL unique algorithm that consists three interconnected parts, including DL model pretraining, correlated source selection, adaptive collaborative fine-tuning (FT). Any components cannot be excluded; otherwise, entire decreases. experiments set six motions show achieves state-of-the-art HMR samples, outperforming several existing approaches. Especially, when there are only 100 samples per person class, yields F1 score 96.7%. Last but not least, more generalized differences. Though adapted recognize persons’ in small-scale target set, can also classify used achieving up 11.0% enhancement over conventional FT method.