Classification of remotely sensed images

Abstract

For this research, the researchers examine various existing image classification algorithms with the aim of demonstrating how these algorithms can be applied to remote sensing images. These algorithms are broadly divided into supervised and unsupervised classification. Classification is a process of assigning pixels from a multidimensional image to discrete categories. The formation of a cluster of pixels for a remotely sensed image is a result of both the spectral value of the pixel, as well as the spatial location of the pixel. The research demonstrates a clustering algorithm that can incorporate both the spectral and spatial features of the pixels in the image resulting in better defined categories in terms of its homogeneity. A hyperspectral image is a remotely sensed image with a large number of spectral bands at various parts of the electromagnetic spectrum. The research also demonstrates the use of these algorithms on a hyperspectral image and show how a classification algorithm, which is specifically designed for hyperspectral images, can be used in forming categories