Architecture Concept for an Information Mining System for Earth Observation Data

Abstract

EOLib, the Earth Observation Image Librarian is an upcoming Image Information Mining (IIM) system for earth observation (EO) products. As integral part of a payload ground segment (PGS) it operates on the original EO products, metadata, and on computed higher level abstractions including basic features and semantic annotations of product tiles. The core goal of EOLib is to introduce mature information mining functions in existing EO payload ground segments. EOLib is integrated with the multi-mission PGS existing at DLR’s premises. Operations including product tiling, feature extraction and automatic annotation are performed within the PGS services infrastructure. Intermediate data including features and quick look images are forwarded to the EOLib core to perform additional data mining operations as: semantic learning, content-based information retrieval and visual data mining. The PGS user services are augmented with query engines for semantic annotations and metadata and with a semantic catalogue browser. We present in this paper the architecture concept of the EOLib system. It starts with a set of functional requirements, the constraints imposed by the existing PGS and the experience gathered from prototype standalone IIM systems. System components are subsequently identified based on logical functionality blocks. Data flows and interfaces are built in order to allow simple, clear system integration and to maximize performance. We conclude with the implementation of a few use cases.