Deep Learning for Direct Automatic Target Recognition from SAR Data

摘要

Automatic target recognition for synthetic aperture radar involves three processing steps: imaging, identifying regions of interests and processing of the identified regions for target classification. Motivated by the successes of data driven approach in various fields, we establish a framework for synthetic aperture radar target recognition that does not require image formation. We present a deep neural network architecture that classifies targets directly from the slow-time and fast-time sampled received signal. Classification decisions are made based on how closely optimized sets of vectors corresponding to each class represent a new sample, and these vectors from all classes under consideration collectively forms a dictionary.