ADAF is an open source software that enables fast and cost-effective archaeological investigation of airborne laser scanning DEMs. The reason for the development of ADAF was Transport Infrastructure Ireland’s need for early identification of previously unknown and potentially significant monuments. This was impossible to achieve without machine learning due to the extent of the survey areas, the constraints of the programme and the limited capacity of the private sector to provide this service.
The ADAF tool consists of Jupyter notebooks for data preparation, training (creation of specific machine learning models), and automatic recognition of archaeological features.
The project involved the creation of a comprehensive training dataset by digitising segmentation polygons for 10,718 recorded monuments and the processing and homogenisation of more than 200 individual ALS datasets. The machine learning models were thoroughly tested for the impact of patch size, visualisation technique, data augmentation and transformation methods, data quality and different architectures for object detection and semantic segmentation on performance. The currently implemented models are optimised to detect three classes of Irish archaeology (barrows, enclosures, ringforts), but other models can be inserted. The software requires minimal interaction and no prior knowledge of machine learning techniques, which greatly increases its accessibility to the archaeological community.
The project is an important step towards democratising and demystifying machine learning for archaeological applications. The work done so far shows that advanced machine learning models can effectively identify and classify archaeological features in ALS data, reducing the need for manual effort and expert intervention. This progress reflects the development of GIS in the 1990s, which has evolved from specialised tools to widely accessible technologies, encouraging innovation and wider application. Open source software, FAIR machine learning -ready dataset, and open access to methodology aim to opening up new ways for non-experts to engage with and benefit from these advanced technologies.
ADAF can be downloaded from the GitHub repository.
YouTube instructional video.
