Exploring urban densification with deep learning: Integrating remote sensing and electrical infrastructure data

Abstract

Rapid urban growth poses challenges for efficient city management, demanding the implementation of sustainable principles. In this scenario, the electrical sector faces significant changes with the introduction of new technologies, notably Artificial Intelligence (AI) and Deep Learning (DL), especially when applied to urban remote sensing. This study investigates the urban growth of a Brazilian state using semantic segmentation based on Convolutional Neural Networks (CNNs) in satellite images. Two CNN architectures, LinkNet34 and DLinkNet34, were trained with RGB satellite images from Google Earth, covering geographical and environmental diversity. Both models showed similar performance, with LinkNet34 slightly superior with an accuracy of 0.8457. Segmentation analysis was correlated with data from the Geographic Database of the Distributor's Infrastructure (BDGD), aiming to provide insights for sustainable urban development.