Parallel implementation of unmixing algorithm for variable-endmember linear mixture model

This study introduces a parallel implementation of unmixing algorithm for a variable-endmember linear mixture model (VELMM). The model had been developed to retrieve leaf area index (LAI) and fraction of vegetation cover (FVC) from remotely sensed surface reflectance. Since the model consists of a linear sum of two nonlinear functions, the unmixing algorithm involves a constrained optimization process which requires a major computation time. To examine its practicality, an implementation of the algorithm was conducted in a distributed memory multi-core system with STAR-P/MATLAB development environment. The results indicate that approximately two orders of speed up can be achieved using 256 cores to process a LANDSAT7-ETM+ full scene comparing to single core case. It was concluded that the algorithm can gain much practicality on a distributed parallel environment.