Land Cover Classification Using Ensemble Techniques

Abstract

This paper aims at a Land cover classification of compact polarimetric RISAT-1 data using pixel-based and patch-based input to an ensemble model. In the pixel-based approach, Support Vector Machine (SVM), Multi-Layer Perceptron (MLP) and Random Forest (RF) are ensembled to shape a single voting classifier using the soft voting and majority voting method. In the patch-based approach, Convolutional Neural Network (CNN), Neural Network (NN) and RF are ensembled through soft voting, majority voting and average voting methods. The experimental results show that the ensemble method on patch-based input provides better test accuracy than the pixel-based input. Visual evaluation of classified images indicates that water bodies and urban region are well classified with patch-based input using majority voting method, while the forest class is less misclassified in contrast to soft voting classification on pixel-based input.