From visual to automatic information extraction on spaceborne SAR imagery

Abstract

The Microwaves and Radar Institute of the German Aerospace Center (DLR) has over 40 years of experience in spaceborne radar imaging. In the past, the main challenges included the development of the sensory part of an imaging radar’s space segment itself. Nowadays, synthetic aperture radar imagery is a well established datasource for a multitude of applications. With the forthcoming of an increasing number of public and commercial missions, one of the biggest current challenges is the handling of the exploding amount of data. This strongly calls for fully automatic ways to extract highervalue information from earth observation data in general, because human resources for image analysis are limited. This presentation will show the current state of DLR’s radar image analysis software toolset RADIAN that has been developed specifically for security applications. It will show the way from manual to automatic information extraction by starting with conventional but highly optimized techniques to prepare imagery for visual analysis. Already during RADIAN’s early development stages, operational usage, robustness, efficient implementation and automated workflows were main objectives, which will be illustrated with several applicational examples. Afterwards, new approaches for an automated extraction of information are given for classical reconnaissance tasks like change detection or object recognition. Conventional rule-based methods still play a major part, but also stateof-the art machine learning techniques are important key factors in current research topics.