2023-03-06, 10:45–11:15, Auditorium
In this talk, we present our journey in creating automated pipelines for data preparation and training GeoAI models.
Besides being responsible for land administration, the Dutch Cadaster is also the national mapping agency of The Netherlands. To address challenges related to topographic mapping like decreasing budgets and lack of trained experts in the future, automating as many processes as possible is a solution we are exploring.
In recent years, advances in GeoAI have made it possible to automate, or support, many of the manual processes involved in creating and maintaining topographic maps. In this talk, we present our journey in creating automated pipelines for data preparation and training GeoAI models. These models aim to improve the accuracy and efficiency of an otherwise labor-intensive process.
We will share how our deep learning models are trained with high resolution aerial photography and how we efficiently scale our (open source and Python based) pipelines to automatically detect features of interest nationwide. Specifically, we use GeoAI to perform object detection and instance segmentation. Examples of detected features are storage tanks, green houses, recreational housing, and forest trails.
These innovations allow our topographic experts to spend their time more efficiently. This enables us as the national mapping agency to achieve operational excellence and keep producing topographic maps more efficiently.