Detection and Billing of Taxable Building Properties in Ghana: A Satellite-Based Approach

Abstract

The rise in the number of a nation's physical infrastructure is a key indicator of its development and, likewise, the proceeds from taxable infrastructure. However, in many sub-Saharan African countries (for example, Ghana), the amount of taxes realised from such infrastructure as buildings (property rates) fall short of expectation and hence hamper national development due to delays in the detection of such properties as well as logistical challenges. Therefore, the importance of accurate and timely detection and billing of such properties cannot be overemphasised. This work sought to evaluate and enhance the performance of some state-of-the-art segmentation models for building detection (U-Net, FPN, PSPNet, U-Net++, LinkNet and MAnet), primarily focusing on satellite images of Berekuso in the Akuapem South Municipal Area. Initial analysis showed that the U-Net++ outperformed the others on the Massachusetts Building Dataset (IOU = 0.8280, mean dice loss = 0.1241, precision = 0.8856, recall = 0.9260, F1 score = 0.9054, and accuracy = 0.9077) but achieved a lower accuracy on the Berekuso Dataset. However, a significant increase in performance is observed by combining enhanced (histogram equalisation) satellite images with post-training techniques. The study, therefore, proposes an incentivised billing system based on the above enhancement mechanism with U-NET++ for building detection towards climate sustainability. This study aligns well with the sustainability goals of SDG 11, 13 and 15.