Automatic extraction of Bursaphelenchus xylophilus-induced sporadic death trees on unmanned airborne digital photographs

Abstract

Bursaphelenchus xylophilus is an insect-spread disease resulting in severe mortality in pine forests. At present, the most effective way to control this infection is to timely remove and destroy the infected trees from the pine forests. This paper explores the approach to automatically extract the infected dead trees from unmanned airborne digital photographs, that is, to automatically identify the infected dead trees and their spatial distribution. The result can be used in guiding the field action. First, a peak-climbing algorithm was used to classify the spectral features into clusters with a small clustering measure. Secondly, the generated clusters were automatically merged with feature space-based Closeness Index and Close Mate. Finally, the analyst interactively merged the clusters of dead trees that cannot be automatically merged with the Closeness Index and Close Mate approach. This research indicated the userpsilas and producerpsilas accuracies based on the Closeness Index approach were 69.9% and 58.8%, 2% higher than that from ISODATA respectively. Both approaches can extract almost all infected dead trees, but other non-forest land covers could be misclassified as dead trees.