Rooftop segmentation to estimate urban sprawl
Research Team: Peiyu Li, Kexiang Xu, Dr. Mohammad Swapan, Dr. Cecilia Xia
CIC Specialist: Shiv Meka
Spatial science researchers in collaboration with CIC, devised methods that enable accurate calculation of urban sprawl from three band satellite images. Urban sprawl is a direct measure of the overall habitable land cover, and when assessed in the context of urban planning, urban sprawl can provide reasonable guidance on the planning logistics based on yesteryears’ patterns. An accurate estimate of urban sprawl requires that buildings and industrial sections are precisely segmented. The overall objective of segmenting images was achieved by converting the segmentation problem to a classification task wherein training data is split into two classes – urban and rural. A custom deep convolutional neural network (DNN) was designed and trained in a way that, while the overall network performs a classification task, segmentation is a natural outcome when certain intermediary layers’ outputs of DNN are arithmetically combined. The project also serves as a prelude to a future research project – “Sustainability index” – a unit that attempts to quantify the sustainability of a locality based on evolution of geospatial entities.
This work was supported by resources provided by the Pawsey Supercomputing Centre with funding from the Australian Government and the Government of Western Australia.