Building Semantic Segmentation Using UNet Convolutional Network on SpaceNet Public Data Sets for Monitoring Surrounding Area of Chan Chan (Peru)

A novel approach leveraging artificial intelligence (AI) and remote-sensing (RS) techniques demonstrates high accuracy in monitoring damage to archaeological sites, achieving around 70% performance in building footprint segmentation. The study focuses on the Chan Chan Archaeological Zone in Peru, utilizing a UNet convolutional network with an EfficientNet encoder trained on public SpaceNet datasets derived from WV2 and WV3 satellite images. Pre-processing equalization of images improved quality and mitigated overfitting, while high-resolution images from Google Earth Pro were used for segmentation analysis. This methodology aims to assist in drafting monitoring and conservation protocols to prevent future disasters, addressing challenges posed by urban growth, seismic events, and ecosystem deterioration.

Chicchon, M., Malinverni, E. S., Sanità, M., Pierdicca, R., Colosi, F., & Trujillo, F. J. L. (2024). Building Semantic Segmentation Using UNet Convolutional Network on SpaceNet Public Data Sets for Monitoring Surrounding Area of Chan Chan (Peru). Geomatics and Environmental Engineering, 18(3), 25–43. https://doi.org/10.7494/geom.2024.18.3.25