Big SAR data processing: Interferometric and polarimetric SAR data structurization based on complex-valued and quaternion neural networks

Abstract

The big synthetic aperture radar (SAR) data era of earth observation depends more and more on deep learning and other data mining technologies. Since SAR data is coherent, differently from incoherent optical data, the use of complex-valued and quaternion neural networks is significantly important for successful applications. This presentation discusses the essence of neural information processing as well as the complex-valued convolutional neural networks and quaternion self-organizing map for feature discovery and SAR data structurization.