The radar is expected to go beyond the traditional functionality of range and speed estimation to target classification. The complementary use of radar and video is becoming increasingly popular for applications such as autonomous cars, smart home automation etc. Target classification based on radar depends on the characteristic motion patterns of target nonrigidities. The Micro-Doppler (MD) signal captures such motions that have been used to extract reliable distinguishing features for various classes of targets. Popular MD analysis techniques such as Cadence frequency estimation require long captures before reliably identifying the target. Such a latency has an impact on the response times especially in time critical systems such as autonomous cars. Although a finite latency is unavoidable, it is in the interest of the community to keep it as small as possible. In this paper, we use unsupervised learning, specifically auto-encoders for learning Micro-Doppler features. We use a particular fast learning algorithm which learns very quickly with little training data and deliver reliable classification.
Parashar, K, Oveneke, MC, Rykunov, M, Sahli, H & Bourdoux, A 2017, Micro-Doppler feature extraction using convolutional auto-encoders for low latency target classification. in Radar Conference (RadarConf). IEEE, pp. 1739-1744, 2017 IEEE Radar Conference (RadarConf), Seattle, WA, United States, 8/05/17. https://doi.org/10.1109/RADAR.2017.7944488
Parashar, K., Oveneke, M. C., Rykunov, M., Sahli, H., & Bourdoux, A. (2017). Micro-Doppler feature extraction using convolutional auto-encoders for low latency target classification. In Radar Conference (RadarConf) (pp. 1739-1744). IEEE. https://doi.org/10.1109/RADAR.2017.7944488
@inproceedings{c1265f0c606c4d818b3281dae4df0879,
title = "Micro-Doppler feature extraction using convolutional auto-encoders for low latency target classification",
abstract = "The radar is expected to go beyond the traditional functionality of range and speed estimation to target classification. The complementary use of radar and video is becoming increasingly popular for applications such as autonomous cars, smart home automation etc. Target classification based on radar depends on the characteristic motion patterns of target nonrigidities. The Micro-Doppler (MD) signal captures such motions that have been used to extract reliable distinguishing features for various classes of targets. Popular MD analysis techniques such as Cadence frequency estimation require long captures before reliably identifying the target. Such a latency has an impact on the response times especially in time critical systems such as autonomous cars. Although a finite latency is unavoidable, it is in the interest of the community to keep it as small as possible. In this paper, we use unsupervised learning, specifically auto-encoders for learning Micro-Doppler features. We use a particular fast learning algorithm which learns very quickly with little training data and deliver reliable classification.",
keywords = "convolution, Doppler radar, feature extraction, frequency estimation, image capture, image classification, image motion analysis, object recognition, unsupervised learning, video signal processing",
author = "Karthick Parashar and Oveneke, {Meshia C{\'e}dric} and Maxim Rykunov and Hichem Sahli and Andre Bourdoux",
year = "2017",
doi = "10.1109/RADAR.2017.7944488",
language = "English",
pages = "1739--1744",
booktitle = "Radar Conference (RadarConf)",
publisher = "IEEE",
note = "2017 IEEE Radar Conference (RadarConf) ; Conference date: 08-05-2017 Through 12-05-2017",
}