Fundamentals and Methods of Machine and Deep Learning

14Measuring Urban Sprawl Using Machine Learning

Keerti Kulkarni* and P. A. Vijaya

Dept of ECE, BNM Institute of Technology, Bangalore, India

Abstract

Urban sprawl generally refers to the amount of concrete jungle in a given area. In the present context, we consider a metropolitan area of Bangalore. The area has grown tremendously in the past few years. To find out how much of the area is occupied by built-up areas, we consider the remotely sensed images of the Bangalore Urban District. Each material on the earth’s surface reflects a different wavelength, which is captured by the sensors mounted on a satellite. In short, the spectral signatures are the distinguishing features used by the machine learning algorithm, for classifying the land cover classes. In this study, we compare and contrast two types on machine learning algorithms, namely, parametric and non-parametric with respect to the land cover classification of remotely sensed images. Maximum likelihood classifiers, which are parametric in nature, are 82.5% accurate for the given study area, whereas the k-nearest neighbor classifiers give a better accuracy of 85.9%.

Keywords: Urbanization, maximum likelihood classifier, support vector machines, remotely sensed images

14.1 Introduction

Urbanization is a key deciding factor for the government to provide various infrastructure facilities. It is an indirect indication of the amount of population staying in the cities. Although the census report does provide this information, ...