Abstract

Obtaining lots of synthetic aperture radar (SAR) data for automatic target recognition (ATR) can be unrealistic in many situations. Relatively speaking, electromagnetic simulated SAR data is easier to acquire. However, there is an apparent distribution gap between the electromagnetic simulated data and the real data. It is not suitable to use the simulated data for recognition directly, and thus the way to apply the simulated SAR data for recognition is worth considering. In this paper, we focus on a challenging recognition problem of training solely on simulated data and testing on real data. Firstly, a multi-scale feature extraction module is used for robust feature extraction, which aims to make the extracted feature more generalized for both simulated data and real data. Secondly, the idea of domain adaptation is considered to learn a representation trained to jointly optimize for classification and domain invariance. Thirdly, an ensembling technique is applied, and an ensemble multi-scale deep domain adaptation recognition framework is proposed. The Moving and Stationary Target Acquisition and Recognition (MSTAR) benchmark data set and an electromagnetic simulated SAR data set are used for verification, and experimental results illustrate the effectiveness of our methods.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call