Most of the advances in modern technology are driven by Artificial Intelligence (AI). Deep learning – a popular subfield of AI is gaining popularity in masses with respect to detecting objects/patterns in images. It not only enables your smartphone cameras to recognize scenes but also detect faces to enhance picture quality. Moreover, when used on satellite imagery – it can help us recognize features such as buildings, roads, parks, slums etc., thus helping us map our cities better.
This article aims to elaborate on building detection through satellite imagery. We have trained a Deep Learning model (UNET) to detect buildings using freely available satellite imagery from Google Earth that contains information only in 3 channels of visible light spectrum (Red,Green and Blue). Using this model we have mapped buildings in Punjab, Pakistan that enabled us to estimate building densities at a very high resolution. Refer to Figure 1 for a visual representation illustrated from our data, showing a comparison of building densities of Lahore with its neighbouring districts.