How to effectively obtain metadata from remote sensing big data?

Abstract

What can be considered big data when dealing with remote sensing imagery? In general terms, big data is defined as data requiring high management capabilities characterized by 3 V’s: Volume, Velocity and Variety. In the past, (e.g. 1975), considering the computational and databases resources available, a series of Landsat-1 imagery from the same region could be considered big data. Nowadays, several satellites are available, and they produce massive amounts of data. Certainly, an image data set obtained by a single satellite, for a specific region and along time, fills the 3 V’s requirements to be considered big data as well. In order to deal with remote sensing big data, we propose to explore the generation of metadata based on the detection of simple features. Besides the intrinsic geographic information on every remote sensing scene, no additional metadata is usually considered. We propose basic image processing algorithms to detect basic well-known patterns, and include them as tags, such as cloud, shadow, stadium, vegetation, and water, according to what is detectable at each spatial resolution. In this work we show preliminary results using imagery from RapidEye sensor, with 5 meter spatial resolution, composed by two full coverages of Brazil with RapidEye multispectral imagery (around 40k scenes).