A Parallel Satellite Image Processing Approach Using a Distributed Memory MIMD System

Satellite remote sensing is a discipline in which data volumes are enormous and processing chains relatively complex. As more satellites are now being launched into space by different organizations and more sophisticated research is being undertaken in this field, the volume of data captured grows year by year with higher temporal, spectral and spatial resolutions. The volume and variety and quality of these data present a challenge to traditional processing approaches. Apart from that, the requirement of the analysis of a multispectral time series of images for the same geographical location in many applications, especially in the investigation of dynamic environmental changes, also has increased the need for the high performance processing power. This paper presents research in developing a parallel processing approach applied to satellite image processing on the distributed-memory MIMD parallel system. The Victorian Partnership of Advanced Computing (VPAC) funds this research project, as one of their expertise program grants. The objectives of the research are to investigate and determine the optimal data parallelism on the distributed-memory MIMD computer, develop the specifications required to map the image data onto the parallel processors and design the algorithms of parallel spatial input/output and spatial analysis to make best use of the chosen parallelism. The paper also presents an empirical test on the determination of the area susceptible to soil salinity by using satellite images. Several approaches that can be used for remote sensing image analysis are introduced and implemented in this empirical example. The result of the performance evaluation that occurred on the MIMD computer in VPAC demonstrates the potential advanced computing has for the development of a set of software tools that can quickly perform precision analysis on large volumes of spatial data.