Land Use Land Cover Classification in Remote Sensing Using Machine Learning Techniques

Machine Learning techniques are popular and effective means for Land Use Land Cover (LULC) classification using remotely sensed data. These techniques are capable to deal with high-dimensional data, which is a major concern in the domain of remote sensing. Literature indicated the effectiveness of Machine Learning (ML) for complex classification of land cover classes. However, it is a challenging issue to select the appropriate classifier. The main objective of this paper is to implement and evaluate the most popular machine-learning techniques for LULC classification using remotely sensed data. The selected ML techniques are Artificial Neural Network (ANN), Support Vector Machine (SVM), k-Nearest Neighbor (kNN), and Random Forest. The chosen study area is Nanital district situated in Uttarakhand, India. In this research work, Sentinel-2 satellite data (10m spatial resolution) has been utilized. Results revealed that Random Forest achieved the highest accuracy (OA) of 91.56% (kappa value of 0.899). It is found that the ANN classifier produced the lowest classification accuracy of 88.16 and a kappa value of 0.859. Other classifiers like SVM and kNN obtained an OA of 90.72% and 89.64% and kappa values of 0.889 and 0.876 respectively.