Urban land cover and land use classification using multispectral sentinal-2 imagery

In recent years, remote sensing data contains ample information about landcover due to the advancements in remote sensing technology. Humans can utilize this data in agriculture, forestry, disaster management, urbanization, and many more applications. The European Space Agency's Sentinel -2 satellite delivers freely accessible multispectral remote sensing data of different high spatial resolutions, which can be used in various remote sensing fields to extract meaningful information. In this work, the multispectral imagery of 10 m spatial resolution of a densely populated urban area, obtained from sentinel-2, is classified using Support vector machine (SVM), artificial neural network (ANN) and maximum likelihood classifier (MLC). The results obtained using the classifiers SVM, ANN and MLC are compared in terms of the kappa coefficient, overall accuracy and accuracy of users and producers. An area of 14 x 14 km(2) of the South-West district of Delhi (India) is chosen for this study with five urban land-cover and land-use(ULCLU) classes, namely roads, water, buildings, vegetation, and barren land, with a training sample size of 150 pixels per class. The highly complex nature (high population density, urbanization) of the study area makes the classification task challenging and appealing. All the classification methods have more than 90% accuracy, but SVM obtains the best performance with 98.05% accuracy. The work presented in this paper can support policymakers in making better decisions and extracting meaningful information about ULCLU.