A Hadoop-Based Framework for Large-Scale Landmine Detection Using Ubiquitous Big Satellite Imaging Data

Abstract

This paper proposes constructing world-wide landmine maps using the free USGS satellite multispectral image archive. Although the available resolution is not suitable for detecting mines (in excess of 100m), we seek to exploit the archive's 40-years worth of earth scans, with same locations appearing hundreds of times, to significantly improve the resolution to a useful scale. This paper proposes a framework, based in iterative map-reduce programming model, for dealing with such 'big image' data. The paper presents an initial study for reconstructing well-known (large) landmarks from the USGS archive, and estimates the computation and space complexities.