Parallel Processing with MPI for Inter-band Registration in Remote Sensing

摘要

Inter-band registration is an essential processing method that is used to generate satellite imagery products with raw data obtained from a high-resolution satellite. The processing method requires considerable time owing to its high computational cost as well as the large-scale data to be processed. In this paper, we present a parallel inter-band registration method on windows-based clusters as part of an effort to reduce the total execution time during product generation. We use a blade system as clusters and the MPICH2 library for the parallel programming tool. We logically divide roles of the nodes into one root node, three sub-root nodes, and several compute nodes. We divide the inter-band registration into sequential and parallel processing areas to design the nodes suitable for inter-band registration. The root node and sub-root nodes control the sequential processing area and exchange data with compute nodes in the parallel processing area. In addition, we introduce an object-oriented pseudo code for easy parallelization using a Message Passing Interface (MPI). We apply our system on the KOMPSAT-2 product generation process. The experimental result shows that our system reduces the total execution time from 330 seconds to 79 seconds.