Land-Use Land-Cover Prediction from Satellite Images using Machine Learning Techniques

Abstract

The objective of the proposed research is to estimate the land-use/ land-cover (LULC) changes by employing artificial intelligence techniques rather than doing it manually. For this purpose, Sentinel-2 satellite images are used; these images are overtly and can be accessible readily within the Earth observation Copernicus. Sentinel-2 images labelled as EuroSAT data are employed, these images cover 13 spectral bands and consist of ten categories. The presented model will ease the process of classification of images so that various usage types of lands will be revealed out of this. In this paper, classification using supervised machine learning (ML) techniques e.g. Random Forest, K-Nearest Neighbor (KNN), Support Vector Machine, Decision Tree, Gradient booster and Ensemble classifiers by stacking all these models are carried out. Furthermore, the results of all the six models supported by the metrics like accuracy, precision, F1 score, and recall are compared. Finally, it is identified that Ensemble classifier is the highly efficacy model which may be applied for classifying LULC cover in order to achieve a highly accurate result in ground data.