Deep learning in urban studies through multi-spectral satellite imagery

The analysis of multispectral satellite imagery using neural networks has emerged as a powerful tool for studying and classifying urban environments. Deep learning enables the processing of large volumes of data by identifying characteristic patterns used to define levels of urbanization, thereby facilitating the identification of distinct urban morphologies. In this work, a neural network is trained on Fractional Brownian Motion (FBM) surfaces for the classification of built-up environments. Multi-spectral satellite images of the WorldView-2 (WV2) database are analyzed. Feature extraction techniques combined with neural network models offer a robust approach to understanding and classifying urban environments, supporting the development of tools for territorial analysis and sustainability.