Efficient combination of multiple hyperspectral data processing chains using binary decision trees

According to the technical literature, there is no classification algorithm which is able to extract different classes with the same quality. In this paper we introduce a novel methodology to build a multi-stage, hierarchical data processing approach that is able to combine the advantages of different processing chains, which may be best suited for specific classes. The combination process is carried out using a binary decision tree (BDT) structure where at each node the most useful input information source, in the form of different processing chains, is used, according to the outcome of a simple learning mechanism on small training/validation subsets. Final results are instead achieved by applying the designed BDT to the whole data set. The usefulness of the procedure is proved by extensive analysis of a standard test data set, the Indian Pine AVIRIS set.