Semantic Segmentation (U-Net) of Archaeological Features in Airborne Laser Scanning—Example of the Białowieża Forest

A U-Net deep learning model successfully achieved semantic segmentation of archaeological, natural, and modern features in the Białowieża Forest using airborne laser scanning (ALS) data. The model demonstrated strong performance, with a Dice-Sorensen coefficient (F1 score) of 0.58 and an Intersect Over Union (IoU) metric of 0.5 after 83 epochs. Notably, the IoU for specific features like ancient field system banks, plots, and burial mounds reached 0.41, 0.62, and 0.62, respectively. This approach offers a promising method for automating the identification of complex archaeological structures in dense forested environments, potentially uncovering previously unknown monuments.

Banasiak, P., Berezowski, P., Zapłata, R., Mielcarek, M., Duraj, K., & Stereńczak, K. (2022). Semantic Segmentation (U-Net) of Archaeological Features in Airborne Laser Scanning—Example of the Białowieża Forest. Remote Sensing, 14(4), 995. https://doi.org/10.3390/rs14040995