The Norwegian Coastal Administration is looking to create a new version of an already existing dataset. The existing dataset defines the coastline using segmented lines with a terrain type associated with each segment, but the dataset contains a lot of classification errors.

To that end, we were asked by Norkart and the Norwegian Coastal Administration to look into the possibilities of using machine learning for terrain classification to help create a new, more correct dataset. In this report we will explore a possible approach to this task and describe how we used work methods like Kanban and Scrum as well as Anaconda, QGIS and TensorFlow as tools for our process.

As it stands at the time of writing, with the achieved result, we find it likely that machine learning is a viable method for classifying terrain from aerial photos. The process of adding and using resource files is simple, and it is likely that better results can be achieved with more data.