Multi-level Adaptive Support Vector Machine Classification for Tropical Tree Species

Abstract

High diversity of tree species in tropical forest is a constraint to achieve satisfactory accuracy in tree species classification, as accuracy reduces with the increasing of target tree species. A new multi-level adaptive classification procedure is introduced in the present study employing Support Vector Machine (SVM). The experiment handled 20 tropical tree species classification using in-situ hyperspectral data. Three levels of classification were carried out and the final overall classification accuracy was improved to 74.56% from the beginning accuracy produced by SVM itself. Result of SVM also has proven its better capability than Maximum Likelihood Classification (MLC) in tropical tree species classification.