The Cultural Landscapes Scanner (CLS) project aims to develop Artificial Intelligence (AI) methods for the identification of undiscovered Cultural Heritage (CH) sites on remote sensing data. The project will address the following needs.
- The need for creating automated means via AI for detecting and preserving archaeological objects and providing crucial information for landscape planners.
- The need for a comparison strategy for different AI methods by introducing a standard set of performance evaluation methods will enable cross-study comparisons.
- The need for a publicly available dataset containing different archaeological features which will enable the cross-fertilisation between AI and CH.
CLS project aims at setting a benchmark in the use of AI and remote sensing data for the automatic identification of various classes of undiscovered CH. This aim will be achieved through the following steps:
1) State-of-the-art object detection and semantic segmentation methods will explored to establish how the granularity of the detection affects the quality of the prediction from an archaeological point of view. A study of the generalisation capabilities will be conducted by comparing different transfer learning settings.
2) To assess the performance of the different models, a set of different metrics will be introduced with the collaboration of landscape archaeologists in order to set a first reference standard for promoting objective cross-study measurements.
3) Publish the first publicly available multi-modal dataset of labelled archaeological sites toward solving the problem of the non-standardisation of performance metrics. This benchmark dataset will contain Sentinel 2 multi-spectral images and LiDAR data of the archaeological landscape of Aquileia (Italy), a major city of the Roman Empire. A quantitative baseline will be provided by comparing state-of-the-art semantic segmentation methods.