AREA ESTIMATION OF SATELLITE IMAGES USING MACHINE LEARNING ALGORITHMS

Abstract

Satellites have been used from past several years to obtain a large variety of information about the earth’s surface like in military applications, estimating global weather patterns, tectonic activity, surface vegetation, ocean currents and temperature, polar ice fluctuations, pollution, and many other problems. Image acquired from satellites are best choice to extract information regarding the natural resources of country that can be extremely useful for planning purposes. Resources include agricultural, geological and strategic resources. Such information can be conveniently extracted from satellite images with application of suitable digital image processing techniques. The low and high level computation using machine learning algorithms can be done for earth’s surface using satellite images. The main emphasis of this thesis is using various machine learning algorithms for area estimation with satellite images for five different classes namely soil, urban, water, vegetation and dark forest regions, spectral unmixing for mixed pixels for calculation of fractional vegetation cover, using different type of vegetation and built-up indices and also calculation textural features such as mean, entropy and roughness for proper classification of soil and urban region in optical satellite images. The study and techniques applied have been performed on region of Uttarakhand, India i.e. Roorkee city and surrounding area.