Moving Target Classification Based on micro-Doppler Signatures Via Deep Learning
Yonatan D. Dadon, Shahaf Yamin, Stefan Feintuch, Haim H. Permuter, Igal Bilik, Joseph Taberkian
Abstract
Radar-based classification of ground moving targets relies on Doppler information. Therefore, the classification between humans and animals is a challenging task due to their similar Doppler signatures. This work proposes a Deep Learning-based approach for ground-moving radar targets classification. The proposed algorithm learns the radar targets' micro-Doppler signatures in the 2D fast-time slow-time radar echoes domain. This work shows that the convolutional neural network (CNN) can achieve high classification performance. Also, it shows that efficient data augmentation and regularization significantly improve classification performance and reduce over-fit.