Deep learning methods for naturalness evaluation of forests based on canopy height model
Typ
Examensarbete för masterexamen
Master's Thesis
Master's Thesis
Program
Complex adaptive systems (MPCAS), MSc
Publicerad
2024
Författare
Bauner, Andreas
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
Forest evaluation has historically been done through field surveys by experts from
national forest agencies or from forestry companies. This is costly manual labor that
consumes a lot of time. A solution could be to use remote sensing ecological data
and automate the naturalness evaluation of forests with the use of computers. The
aim of this thesis is to develop a machine learning model that could help automate
naturalness evaluation of forests. The remote sensing data is in the form of a Canopy
Height Model (CHM), that is height of trees obtained from airborne laser-scanning.
Ground truth data for forest naturalness is given in the form of annotated, georeferenced
polygons. The study area is limited to a 50×50 km2 area north-east of the
city of Jönköping in Sweden. After applying different processing steps on the data,
it is then used for training a convolutional neural network, based on U-Nets, on
this semantic segmentation task. The evaluation of the model shows good results,
achieving an accuracy of 94.1% on the test set. This performance is competitive with
currently used models for related tasks and shows the feasibility of using machine
learning in the relatively new field of automated naturalness evaluation of forests.
Beskrivning
Ämne/nyckelord
machine learning , artificial intelligence , U-Net , forests , canopy height model , remote sensing , semantic segmentation