SVM-based Classification of Multi-temporal Sentinel-2 Imagery of Dense urban Land Cover

Abstract

Abstract The technological breakthrough and the availability of multispectral remote sensing data have given rise to an ambitious challenge for the classification of the multispectral images accurately to support administrative bodies in decision-making. In this paper, the multi-temporal medium resolution Sentinel-2 imagery of the densely populated urban area of Delhi-NCR is classified using SVM into five different land cover classes, namely water bodies, barren land, vegetative region, road network, and residential areas. Further, the effect of different kernel functions of SVM on land cover classification performance is contrasted and the radial basis function (RBF) leads to the best results. The experimental results are compared with the maximum likelihood classification (MLC) method on different evaluation metrics. The SVM with RBF kernel shows promising improvements in the overall accuracy by 10 percent relative to the polynomial kernel and by 3 percent compared to MLC. The analysis of multitemporal spectral imagery of the study area reflects the increase in a built-up area (road network, Buildings), water bodies, and decrement in the area of barren land and vegetation.