Satellite Image Classification Using KNN Algorithm

Abstract

The process of analyzing and classifying a satellite image in order to produce thematic maps involves categorizing the corresponding picture elements. To do this process in a more efficient way than the previous approaches, we have proposed to use kohonen maps which train itself using unsupervised and competitive learning techniques. These maps cluster the pictures more accurately than the existing method of K- medoid clustering. The Kohonen networks does the functionality of clustering by automatically comparing the similar features of the pixels and assigning it to the same class of its alike ones and also helps in dimensionality reduction of data. We have incorporated this with mostly widely used K- Nearest Neighbor (KNN) classification technique to finally the classify the processed data into four categories namely Irrigation land, Green land, Arid land and Aqua.