Automatic Semantic Enrichment Of Big Earth Observation Data For SpatioTemporal Querying In Image Databases

Big earth observation (EO) data is a challenge for efficient and intelligent analysis, storage and distribution. While the main challenge for other big data domains is the sheer amount of data, satellite data requires conversion into information to unfold their potential as a source of relevant, multi-temporal, geo-information. Current EO image retrieval is based solely on simple, text-based metadata (e.g. acquisition time, target geographic area, cloud cover estimates) without the possibility of higher semantic content-based image querying or spatiotemporal image content extraction. To date, no operational semantic content-based image retrieval (SCBIR) systems for EO-data exist based on fully-automated, semantic enrichment of satellite imagery up to (human understandable) basic land cover classes. Such semantic enrichment facilitates complex and context-sensitive queries and exploitation of long timeseries of remotely sensed data with an acceptable performance for big data applications. Within the research project SemEO (Semantic enrichment of optical EO data to enhance spatio-temporal querying capabilities), conceptual strategies and technical framework conditions are investigated to develop and improve a novel, semantic querying system for content-based image retrieval from multi-source big earth data. The project aims to: (1) semantically enrich optical EO data to a level of basic land cover classes based on a convergence of evidence approach; (2) investigate spatio-temporal modelling and querying techniques in semantically enriched EO databases using encoded ontologies; and (3) review and test the usability and scalability of array databases and specific implemented data models (data cubes) for spatio-temporal queries in big image databases. SemEO’s current research status will be reported, showcasing the first results of how human users are able to query big EO data on a higher semantic level for different use cases implemented in an integrated EO data online processing environment, e.g.: - cloud-free semantic content based image retrieval of user defined AOIs - near-real time flood risk analysis, based on time-series analysis of all available Landsat 8 data for an area in Somalia - land cover change detection through time, based on a dense Sentinel-2 time-series