Spatio-temporal analysis of Indian urban infrastructure growth using deep learning and 3-channel RGB satellite images

Land cover detection and classification has been an important component of Geographic Information Systems. They are used in policy planning, socio-economic analysis, cartography and Government scheme planning and evaluation. Our study uses high-resolution time-series satellite images of Indian cities between years 2000- 2017 and measures the changes in area occupied by infrastructure such as buildings and hutments during that period. To detect buildings and hutments we train a U-Net model1 for image segmentation task and highlight the boundaries for man-made constructions i.e. buildings and hutments for each block in our New Delhi data collection. We have also provided sample contrast against the development information available on BHUVAN portal, made publicly available by Indian Space Research Organization (ISRO) study. Using the time-series data of building and hutment growth, we can enable urban planners and policy makers to identify necessity of supplementary resources like government hospitals, roads, gardens, etc.