Bruk av maskinlæring for å gjenkjenne bygninger i flybilder
Journal article, Peer reviewed
Published version
Åpne
Permanent lenke
http://hdl.handle.net/11250/2592915Utgivelsesdato
2018Metadata
Vis full innførselSamlinger
Originalversjon
Kart og Plan. 2018, 111 (3), 221-230.Sammendrag
This paper explores how machine learning can be used to recognize buildings in aerial images (orthophotos). The Norwegian map data: Joint Map Database (Felles kartbase) is used as «the true value» for training neural networks to detect buildings in aerial images. The paper introduces two models for machine learning. One basic model ( AirNet) and one model which include more layers in the network (AirNet-extended). The second model is supposed to detect more complex shapes in the images. Several parameters for training the neural network are tested. Complete building maps can still not be made using machine learning, but our results show that these methods can be used, for example, to study of city growth, find unregistered buildings etc.