Hybrid Hierarchical Classifiers for Hyperspectral Data Analysis

We propose a hybrid hierarchical classifier that solves multiclass problems in high dimensional space using a set of binary classifiers arranged as a tree in the space of classes. It incorporates good aspects of both the binary hierarchical classifier (BHC) and the margin tree algorithm, and is effective over a large range of (sample size, input dimensionality) values. Two aspects of the proposed classifier are empirically evaluated on two hyperspectral datasets: the structure of the class hierarchy and the classification accuracies. The proposed hybrid algorithm is shown to be superior on both aspects when compared to other binary classification trees, including both the BHC and the margin tree algorithm.