Hybrid Image Classication Using a Shared Memory Parallel Algorithm

TL;DR: To facilitate the transition to parallel processing from serial processing, this work modies the iterative guided spectral class rejection (IGSCR) classication algorithm for implementation on a parallel platform and uses the shared memory parallel programming paradigm, and the Fortran 95 language with OpenMP directives.

Abstract: Although the processing speeds of personal computers have increased dramatically during the past few years, the newer, faster computers cannot keep pace with the size increases in remote sensing datasets due to higher resolution sensors, larger study regions, and the like. A single workstation may not be sucien t to process these large datasets, but there are increasingly more powerful parallel computers available for use within the scientic community. To facilitate the transition to parallel processing from serial processing, this work modies the iterative guided spectral class rejection (IGSCR) classication algorithm for implementation on a parallel platform. The code utilizes the data format HDF5 (Hierarchical Data Format, version 5) and data access library to simplify data access while maintaining storage and input/output eciency , even for parallel access. This work uses the shared memory parallel programming paradigm, and the Fortran 95 language with OpenMP directives. Results are given for the SGI Altix 3300 shared memory computer with as many as 12 processors to demonstrate parallel speedup.