Archaeological site detection on remote-sensed images using U-net

Abstract This project aims to develop a model to identify archeological sites (tumulus and hillforts) on LiDAR data through multiclass semantic segmentation. The data used is from the Alto-Minho region in the north of Portugal and was obtained within the scope of the ODYSSEY project. In this work, it was used a U-net model, that achieved a recall and precision of 64% and 94% (class-wise metrics), respectively, regarding the identification of hillforts, and was not able to identify any of the tumulus presented on the data.