SC13 Denver, CO

The International Conference for High Performance Computing, Networking, Storage and Analysis

Mapping and Characterizing Global-Scale Human Settlements Using HPC.


Authors: Dilip Patlolla (Oak Ridge National Laboratory), Anil Cheriyadat (Oak Ridge National Laboratory), Harini Sridharan (Oak Ridge National Laboratory), Vincent Paquit (Oak Ridge National Laboratory), Jeanette Weaver (Oak Ridge National Laboratory), Mark Tuttle (Oak Ridge National Laboratory)

Abstract: Analytics derived from large-scale analysis of satellite image data is a key driver for many geospatial models including population mapping, risk modeling, and critical infrastructure assessment. Making informed national-level decisions based on analytics such as the spatial distribution of mobile home parks or the development rate of new constructions requires repeated processing of Peta-scale high resolution image data. The process often comprises computationally expensive steps like extracting, representing, and identifying pixel patterns that corresponds to the physical location and man-made structures. Such processes can easily saturate conventional CPUs. Our settlement mapping system consists of computer vision and machine-learning techniques implemented on a multi-GPU architecture to extract, model, and interpret image data to characterize the spatial, structural, and semantic attributes of human settlements. It scales linearly with increase in the number of GPUs while delivering significant speedup enabling us to produce world scale human settlement mapping products at every 1-2 year’s.

Poster: pdf
Two-page extended abstract: pdf


Poster Index