The webinar “Using AI to monitor historic towns and landscapes: #Connecting Archaeology” on 23 June discussed :
“AI and satellite images to monitor historical (features of) landscapes”, Henk Alkemade
“3D and AI technologies for the development of automated monitoring of urban cultural heritage”, Tadas Žižiūnas
“Monitoring of Immovable Cultural Heritage using 3D and Artificial Intelligence Technologies”, Rimvydas Laužikas
The field of cultural heritage is moving forward the development of artificial intelligence, while progressing the field itself. Satellite imagery and drones are already in use to monitor landscapes for documentation and conservation, with the help of AI the processing of images and data can become automated and accurate.
What we learned:
Identifying cultural landscapes assures their conservation and impacts development decisions or alteration by informing policymakers and the public. Historic maps gave us insight into identifying heritage places; satellite imagery and artificial intelligence can simplify this process.
a) The classification of cultural protected landscapes can be long as it requires gathering and processing large volumes of information.
c) Mapping and monitoring the current state of heritage requires: noting several structures like monuments, buildings, streets, parks etc. and several aspects within them as architecture, composition and so on.
b) Cultural landscapes face several threats notably:
- Urban development
- Damage to heritage
- Natural Disasters
- Clean energy development
d) Recognising the threats and taking action is arduous and often requires consultation with several state and private agencies, which is time-consuming and problematic, especially when protecting heritage from disasters that require immediate conservation interventions.
Available technology can mitigate these issues. Satellites orbiting Europe gather extensive images, of spatial and temporal resolution, that can inform several aspects of heritage management when analysed. An A.I. can process massive amounts of data, identify patterns and analyse images more efficiently than any group of people.
Cost and time effective
Higher information access
Innovation, transparency and Standardisation for policy and regulation
Below is a closer look at the process:
Satellites remotely sense spectral signatures (recognising sunlight, atmosphere changes and earth futures, forests, towns, buildings, the height of surface) that in turn, provide us with images and information.
Where do I get them?
These sources offer free access to satellite data:
Landsat USGS Earth Explorer
Copernicus, Sentinel Open Access Hub
Google Earth Pro
How to train your A.I.
For an A.I. to process satellite imagery, it requires training. The complexity of tasks you wish the A.I. to accomplish defines the type of training and the subset of Machine Learning as Deep learning* you will employ. ( in layman’s terms an A.I. learns a new capability, from raw information, which it analyses, classifies and produces or conceptualises into new information)*
There are 3 phases to training and inference.
The first phase consists of identifying what you are looking for in the images and creating a training dataset for the A.I. to learn. The dataset is a set of images and their descriptions(labels), its size will depend on your requirements.
The A.I. will train over a few cycles and test the dataset, which calls for frequent monitoring of its performance and feedback to learn and correctly classify the data. Following this process, the A.I. will apply its newly learned skill to new data.
The webinar goes in-depth in describing the process, types of data, data collection, methodology AND MORE.
I want to learn more:
* ‘Applying and deploying Artificial Intelligence (AI) in GLAMs’ is a great resource to understand A.I. and Machine learning.
Our previous articles