An assessment of support vector machines for crown delineation of pine single trees on unmanned aerial vehicle imagery

Abstract

The aim of this study was to compare the performance of the Support Vector Machine (SVM) algorithm with pixel and basic object-oriented approaches for identifying the crown of single pine trees in a man-made forest. For this purpose, the SVM algorithm was evaluated based on four different kernels: Linear, Polynomial, RBF and Sigmoid. In pixel base approach, the ROI obtained from the user's choice, and in the object-oriented approach, the ROI obtained from segmentation for part of the image. Then, the results of crown area estimation in both approaches were compared with each other and in-situ data. The results showed that using ROI from the object-oriented provides accurate result with less run time consuming. The SVM classification algorithm with RBF and ROI obtained from segmentation were showed the best performance in comparison to other approaches.