Publication

Abstract : Classification is a technique used for categorizing different features associated with the land cover. It is a very important and visually distinguishable method for understanding the land cover and land use pattern in any area. Classifying images is a part of every study that includes change detection. Various methods are used for classification. Classical, conventional and parametric methods of classification are always time consuming and chances for error are also high. Supervised and unsupervised classification techniques are useful inorder to classify the images into distinguishable forms. For fast and accurate classification it is very vital to build a model that can classify the satellite image. In this study two Machine Learning techniques are used to perform LULC classification and the accuracy of both techniques are compared and a performance assessment is done. For attaining the best classification results it is important to provide good training samples, correct data sets, the best algorithm for classification, etc. So here we use Landsat 8 Operational Land Imager (OLI) dataset and uniform training sites for all three classification techniques. Maximum Likelihood Classification (MLC) is done in ArcGIS software, Naive Bayes classification (NBC), and Random Forest Classification (RFC) is done in Google Earth Engine using the same training samples. All three classifiers worked well. To evaluate the accuracy of each classifier, the Kappa coefficient was calculated for each technique by forming a confusion matrix. The accuracy assessment gave a promising result that Random Forest is comparatively the best technique with an accuracy of 94.86% while NBC and MLC also gave satisfactory accuracy values. All three classification techniques applied had certain limitations and were not 100% accurate hence study can be elaborated to find more classifying techniques and their accuracy in order to get the most reliable classification technique.