31. Building Extraction of Kolkata Metropolitan Area Using Machine Learning and Earth Observation Datasets

Rural-to-urban migration and increasing population has created urban agglomeration, particularly in metropolitan cities. This agglomeration creates pressure on cities which interrupts the city to become sustainable. Evaluation of the pattern of urban growth pattern is a crucial task for a city’s long-term development. The building footprint is one of the most important features of a city form to support urban management and development. Previous studies show that high-resolution images are robust to extract building footprints using machine learning algorithms. The main objective of this study is to extract the building footprint from the satellite imagery using machine learning algorithms. In this study, Sentinel-2 multispectral satellite imagery and support vector machine (SVM) linear and radial basis function (RBF) have been used to extract the building footprints in Kolkata metropolitan area. In addition, both pixel-based and object-based image classification approaches have been applied and compared in this study. This result shows that in pixel-based image classification SVM linear gives a high accuracy than the SVM RBF. The accuracy level of SVM linear is 92.58% while Kappa is 0.89. On the other hand, object-based image analysis LULC classification has been done using the SVM ML algorithm. In this image classification, the SVM RBF kernel type gives high accuracy. The overall accuracy of this OBIA image classification is 91.58% and the Kappa is 0.87. For the building extraction in an urban area from the medium-resolution image Sentinel 2 using a machine learning algorithm with high accuracy gives a significant approach. Policymakers and planners can develop the city sustainably from this building footprint in an urban region and use sustainable urban planning to achieve the Sustainable Development Goal.