Day 4

Detailed paper information

Back to list

Paper title The benefits of EO super-resolution techniques in land applications: Monitoring cultural heritage sites with enhanced Sentinel-2 data
  1. Mihaela-Violeta Gheorghe GMV Innovating Solutions Speaker
  2. Fredrik Samuel Nistor GMV Innovating Solutions
  3. Vlad Gabriel Olteanu GMV Innovating Solutions
Form of presentation Poster
  • C1. AI and Data Analytics
    • C1.04 AI4EO applications for Land and Water
Abstract text Preservation of historic monuments and archaeological sites has a strategic importance for maintaining local cultural identity, encouraging a sustainable exploitation of cultural properties and creating new social opportunities. Cultural heritage objectives are often exposed to degradation due to natural and anthropogenic impacts.
With its main objective being transferring research-based knowledge into operational environments, AIRFARE, a national funded project lead by GMV Romania, intends to implement, test and promote responsiveness solutions for effective resilience of cultural heritage sites against identified risks by exploiting the benefits of Earth Observation data wide availability and capabilities.
At a first iteration with potential users involved in cultural sites management in Romania, they manifested most interest for change detection capabilities to prevent illegal dumping of waste, illegal building and changes of land use/land cover within boundaries of large heritage sites (such as old fortresses), which often contain private owned properties with special construction regime. A monitoring service that would provide warnings in a timely manner to support intervention should be able to ensure at least monthly updates of information. While temporal resolution of Sentinel-2 data can easily respond to user needs in terms of frequency, the spatial resolution of 10 m provides limited capabilities in detecting changes that can be indicators of illegal activities at detailed scales: occurrence of new roads, new buildings, non-compliant waste sites on public areas and changes of land cover or destination within private properties. While very high resolution imagery would cover the needs in terms of spatial resolution, frequent acquisition costs are too prohibitive and would substantially reduce the economic benefits of the proposed solution.
In order to meet user requirements for spatial and temporal resolution, we employed a Super-Resolution Generative Adversarial Network (SR-GAN) inspired algorithm trained on SPOT-6 data to upscale and enhance Sentinel-2 imagery. The particularity of the model that we selected is that the loss function calculation is based on VGG network feature maps, which leads to a decreased sensitivity of the model to changes in pixel space.
As an initial approach, we used very high resolution SPOT imagery acquired over five cultural sites in Romania during each season of a year. Sentinel-2 data that was used for the initial training of the model was acquired in the same period as the SPOT images, in an attempt to reduce potential inconsistencies caused by changes in seasons between corresponding training datasets. The first results of the approach produced a year-long stack of synthetic images with a spatial resolution of 2.5 m, therefore upscaling the resolution of the Sentinel-2 imagery by four times. In order to improve the performance of our model, we intend to extend our training dataset in the future, the next step being implementation of a monitoring and risk prevention system based on automated change detection from synthetic imagery stacks.
Our project activities will rely on the Copernicus Earth Observation programme to support public authorities and private sectors involved in cultural heritage management by offering satellite-derived information in a timely and easily accessible manner. Although in an early stage, the work conducted so far demonstrates once again the operational and possible commercial potential of Earth Observation data in corroboration with AI techniques in becoming a viable solution that answers user-driven products and services that meet the day-to-day real needs arising in land management application sectors.
This work was supported by a grant of the Romanian Ministry of Education and Research, CCCDI – UEFISCDI, project number PN-III-P2-2.1-PTE-2019-0579, within PNCDI III (AIRFARE project).