Parallel ISODATA clustering of remote sensing images based on MapReduce

The ISODATA clustering algorithm is regarded as a common method in the field of analyzing remote sensing images. It is very effective to generate a preliminary overview of images. These kinds of clustering methods are currently done in personal computers. However, with the development of remote sensing technology, the spatial resolutions are increasing rapidly and the sizes of the data are becoming larger. Clustering large amounts of images is considerably time-consuming in personal computers because of the limitation of both hardware and software resources. Researchers have developed many kinds of variants of the ISODATA algorithm executing in parallel, and most of them are implemented by using MPI. Generally, writing programs in MPI requires sophisticated skills of the user. Different with the former studies, we propose in this paper to parallel ISODATA clustering algorithm on MapReduce, another parallel programming model that is very easy to use. The algorithm is mainly divided into two steps defined by the framework of MapReduce, and they are detailed by pseudo-codes. To improve the accuracies of the color values, the color space CIELAB is used instead of RGB. The experiment results demonstrates that our proposed algorithm possess a robust scalability and the computational time substantially reduced through increasing the number of nodes and it may inspire new solutions of other similar problems.