The assessment of population within a certain region holds significant importance in the field of urban planning and the allocation of resources. The primary objective of this proposed research endeavour is to aid in assessing the population of a region by estimating building counts through a two-stage framework that analyses low-resolution satellite images using convolutional neural network (CNN)-based models. The first phase of the framework focuses on the segmentation of built-up areas in a satellite image using a Mask-RCNN model, while the second phase employs a convolutional neural network (CNN)-based regression model to estimate the number of buildings within each segmented built-up area without the need for individual extraction of roof-tops.

The extraction of roof-tops from low-resolution satellite images for population estimation still poses a huge challenge to the researchers due to the lack of visual clarity. Further, in densely populated areas, the low contrast of the built-up areas causes huge difficulty in the detection of roof-tops individually. In view of such challenges, we develop a Mask-RCNN model for segmentation of probable built-up areas in low-resolution satellite images instead of attempting to extract every building individually. Subsequently, a CNN-based regression model is developed to estimate the count of buildings within the segmented built-up areas in low-resolution satellite images.

The proposed framework exhibited a promising level of accuracy while working with low-resolution satellite images. The experimental results indicated that the proposed framework can provide a cost-effective solution for estimating the population in a region, which is useful for the assessment of demographic variation, resource allocation for disaster management, smart city construction, and many other socio-economic planning activities.