Finland’s new national aerial photography and laser scanning programmes produce more accurate remote sensing data in a shorter cycle. Using the method developed in the project for automating topographic data production, more changes in aerial photos can be interpreted automatically in the future. This allows people to focus on the important work machines cannot do.
Customers can see this change through more quickly updated features in the National Topographic Database. Automation helps update data more quickly, which produces benefits for society at large.
– The significance of updated building and road data is increasing all the time. At the same time, new technologies offer opportunities to interpret any changes in our environment more automatically using remote sensing data. It’s been interesting to find out how we can increase the automation rate in updating key features in the National Topographic Database, such as buildings and roads, and in this way keep topographic data better updated, says Heli Laaksonen, Head of Cartography at the NLS.
The project will build a new, uninterrupted and ‘smart’ channel network, created automatically. The current channel network in the National Topographic Database, i.e. data about ditches, streams and rivers, is not topologically intact, and it does not include comprehensive information about the direction of waterflows. The channel network does not currently enable the analyses required in water protection, for example. The channel network produced in the project provides partners with valuable data for preparing various flow analyses.
– Channel data is difficult to access and collect for teaching AI. We’re still taking our first steps in this, Laaksonen says.
Three highlights of the project:
- We are using various source material, separately and together, in developing deep learning methods. We are using 2D and 3D data. In addition to remote sensing data (point clouds and aerial photos), we are also using various vector data – data that is already available and data collected specifically for this project.
- So far, the most promising results have been achieved by using true orthophotos and a deep learning method in identifying and vectoring buildings.
- A master’s thesis on the use of a transfer learning method has been completed as part of the project, and the method can be tested in practice shortly.
Cooperation in Finland and abroad
The project is being carried out with NLS topographic data production, the Finnish Geospatial Research Institute (FGI), and specialists representing different fields. The project is cooperating not only with the NLS’s Finnish partners, but also with other Nordic land survey agencies.
– Other land survey agencies are also working on similar challenges. We want to learn from each other so no one needs to reinvent the wheel, says Laaksonen.
The goal is to produce openly accessible Nordic material for teaching AI, with which other parties can also develop AI solutions in the geospatial data sector. Laaksonen has good news for everyone interested:
– The material produced in the project may benefit the geospatial data sector as a whole. Whenever we can validate the quality of different datasets, we aim to share them with everyone who is interested.
The project has been funded by appropriations of EUR 399,000 granted by the Ministry of Finance for the NLS. The aim of funding is to support projects that use AI, software robotics, data analytics and other emerging technologies, and improve productivity.
Heli Laaksonen, Head of Cartography, email@example.com,
tel. +358 29 531 527