Analysis of Machine Learning Techniques for Sentinel-2A Satellite Images

Abstract

This article presents the comparative analysis of classification techniques to assign land use and land cover classes from different strategies (pixel-based, object-based, rule-based, distance-based, and neural-based) with a Sentinel-2A satellite image for 2016. The study area is the Sana’a city of Yemen which covers about 18,796.88 km2 land area. This research aims to present the fundamentals of supervised machine learning approaches, including their limitations and strengths and experimentation for twelve classifiers. The outcome of experimentation showed that the Random Forest could be a good choice as a classifier for object-based strategy. In contrast, DTC and SVM were efficient in rule-based and pixel-based strategies. Results also showed that the highest accuracy was with object-based strategy, followed by rule-based and then pixel-based and distance-based strategies.