Management of metadata from remote sensing big data

Abstract

A free service provided by the Brazil's National Institute for Space Research (INPE) is the remote sensing images catalog. This catalog contains approximately 120 TB of data (images in GeoTIFF format), and keeps growing, because satellites keep observing the Earth's surface. For example, the CBERS-2B satellite, when it was working, produced 120 megapixels per minute. If we select the region of São Paulo city, Brazil, a search in the catalog will return about 640 available images, starting from 1980. Considering that each image has approximately 7000 lines x 7000 columns, occupying at least 50 MB in disk, the full result is equal to 32 GB of potential data for remote sensing applications. In terms of data volume, this is equal to 8.000 MP3 music files, or 40.000 text documents. However, this information is not fully used, because in general researchers look for certain images in the catalog, download them and do not use the others. The search keys in image catalogs are usually location and sensor, although modern search tools (like in http://images.google.com/) are not available in the context of remote sensing. Such modern tools include the content of the images, or relation between possible targets. In this case, a more sophisticated search could be: find remote sensing images with vegetation, lakes and without clouds. A second example could be: find remote sensing images in Amazonia with more deforestation than forest. Summarizing, this project aims to produce a methodology to compute metadata from remote sensing images to allow this kind of searching tools, based on image content. Such content could be explored using basic image processing techniques, coupled with data mining and database management tools and algorithms. (AU)