Automatic Semantic Enrichment for Intelligent Big Earth Observation Databases

Dirk Tiede / September 5, 2017

Share

EO Level 2 information layers, are automatically generated and linked with the EO data to be employed as input for spatiotemporal semantic querying (Tiede et al., 2017)

EO Level 2 information layers, are automatically generated and linked with the EO data to be employed as input for spatiotemporal semantic querying (Tiede et al., 2017)

Big Earth observation (EO) Data has unique challenges for efficient and intelligent storage, data analysis, and data distribution. While for other Big Data domains the main challenge is in the sheer volume of data, remote sensing data also requires a transformation of pixel values into usable information to realize their potential as a source of relevant multi-temporal, globally available geoinformation, used for example in the global monitoring of land cover dynamics including deforestation, or for efficient disaster management.  

State-of-the-art EO image retrieval is based on simple metadata text information (i.e. acquisition time, target geographic area, cloud cover estimate), without the possibility of higher semantic content-based querying of images or spatio-temporal analysis directly in the database (i.e. an initial classification of pixel values is missing). Existing EO content-based image retrieval systems do not support semantic spatially explicit content-based querying because they lack EO image understanding (EO-IU) capabilities of multi-source EO big data, which is a pre-condition for semantic content-based image retrieval.

In the approach developed at the universities of Salzburg, Austria and Naples, Italy, as a proof-of-concept, an innovative EO semantic querying (EO-SQ) subsystem was designed and prototypically implemented together with an EO-IU subsystem. The EO-IU subsystem automatically generates Level 2 geospatial products  (including a scene classification map, up to basic land cover units) from optical satellite images. The EO-SQ subsystem comprises a graphical user interface (GUI) and an array database system embedded in a client server architecture. In the array database, all EO images are stored as query-optimized space-time data cubes, together with their products.

The GUI facilitates database querying through a graphical world model, i.e. an ontology of real-world (geo-)spatio-temporal objects and events derived from the underlying data types. Furthermore, it provides access to a shared knowledge base, fostering the online collaboration in spatio-temporal EO image analytics between domain experts and non-experts.

(upper left) Extracted flood mask using a long time series of optical EO images (Landsat 8, 01/2016-09/2016). (right) Analysis of a single flood event in Oct./Nov. 2015. In addition to an extraction of the flood extent, a cloud mask was extracted from the semantically enriched database to support flood mask interpretation (from Sudmanns et al., 2017, modified)

(upper left) Extracted flood mask using a long time series of optical EO images (Landsat 8, 01/2016-09/2016). (right) Analysis of a single flood event in Oct./Nov. 2015. In addition to an extraction of the flood extent, a cloud mask was extracted from the semantically enriched database to support flood mask interpretation (from Sudmanns et al., 2017, modified)

Use cases of semantic spatio-temporal queries within the Big EO data bases through the GUI are limitless, due to the generic semantic enrichment of the underlying EO data. Examples already accomplished using Sentinel-2 and Landsat data include:

• Land cover change detection through time
• Cloud-free semantic content based image retrieval for user defined AOIs
• Extraction of higher-level land cover classes based on shape descriptors (e.g. Lake and River) and/or multi-temporal analysis (e.g. agricultural area)
• Flood extent mapping and aggregation through time (flood risk analysis)

This system contributes to the big data paradigm to ‘bring the user to the data and not the data to the user’, and to the ability of users to retrieve valuable information otherwise hidden in big EO data archives.

Read more:

Tiede, D., Baraldi, A., Sudmanns, M., Belgiu, M., Lang, S., 2017. Architecture and Prototypical Implementation of a Semantic Querying System for Big Earth Observation Image Bases. Eur. J. Remote Sens. 50, 452–463. http://www.tandfonline.com/doi/full/10.1080/22797254.2017.1357432

Sudmanns, M., Tiede, D., Wendt, L., Baraldi, A., 2017. Automatic Ex-post Flood Assessment Using Long Time Series of Optical Earth Observation Images. GI-Forum J. Geogr. Inf. Sci. 1, 217–227. doi:10.1553/giscience2017

About the author:

Dirk Tiede, PhD, is Assistant Professor at the Department of Geoinformatics - Z_GIS, University of Salzburg, Austria, where he is co-leading the research group on object-based image analysis (OBIA). He is involved in a range of national and EU-funded research projects, specialized in image analysis, process automation for Big EO data, object-based information extraction and multi-source data integration. www.zgis.at .