Topographical Feature Extraction Using Machine Learning Techniques from Sentinel-2A Imagery

Abstract

The advancement in the satellite technology has made it possible to easily and frequently obtain the satellite images of most of the regions on the Earth. The satellite data contains the abundant amount of information which can be very useful for variety of societal applications. However, manual identification of the land cover in a particular area is a very challenging and time-consuming task. In this manuscript, an attempt has been made to better extract the landcover types from Sentinel-2A imagery using the popular classifiers such as random forest, SVM, Naive Bayes, Decision Tree (CART). The manuscript also validates the results obtained by the used models by computing the performance metrics. The analysis reveals that random forest classifier outperforms the rest of the classification methods in terms of better accuracy of 95.67%. This automated approach can be applied to large sets of data, reducing the need for manual labeling. © 2020 IEEE.