Spatial clustering using neighborhood for multispectral images

Abstract

Spatial data mining discovers patterns and knowledge in spatial data. The geospatial data analysis plays a decisive role in framing essential policies related to the environment at the national as well as global level. It helps in the prediction of weather, studying the impact of climate change, the effect of global warming, forest fire, deforestation, and other causes of changes in nature. Clustering is a process of assigning similar objects to a group and dissimilar objects to different groups. We proposed a clustering algorithm, spatial clustering using neighborhood (SCN), in which similar neighboring pixels groups together to form a cluster. It is a two-step process: first, identification of clusters of similar neighboring pixels and second, merging of spatially separated similar clusters. Experiments with the implementation of the proposed approach have been carried out on three Landsat 5 thematic mapper images of Delhi, Chilika lake, and Kolkata region. In terms of normalized cluster quality parameters, SCN produced better clusters than other discussed algorithms in the literature.