Urban Area Extraction Using Machine Learning Algorithms

Abstract

Abstract Urbanization is the major concern nowadays for the whole world as it is increasing at a very tremendous rate. Several studies have already been conducted and new researches still going on in this particular field. Considering optical data for urban mapping is a challenging task using conventional supervised classification methods. A new method of classification needs to be developed to overcome this problem. In the study, Decision Tree, Random Forest (RF), Support Vector Machine (SVM) and Artificial Neural Network (ANN) based machine learning classifiers have been used for urban area classification. For the study, high resolution Sentinel-2A satellite image is considered so as to get the efficient urban map of area around Roorkee, Haridwar. Spectral features are good at discriminating classes to some extent but intermixing of pixels in few bands affects the accuracy. In this study, extraction of average spectral reflectance features of each class in different bands is considered as a feature attribute and combined with the geo-coordinates at the point locations in a data-frame to train the classifiers and urban area maps are created using these classifiers. Machine learning models such as Decision Tree, Random Forest (RF), Support Vector Machine (SVM) and Artificial Neural Network (ANN) models have been trained using the training dataset to classify the urban area and accuracy assessment is performed to get the best classifier. The overall accuracy for the above classifiers is in preferring order as 94.50, 93.00, 92.00 and 91.5% respectively for SVM, RF, NN and Decision Tree. Our result showed that SVM model performs best, followed by RF, ANN and decision tree. ANN and decision tree are relatively poorer in terms of urban area extraction.