Hierarchical band selection method based on scalability tree structure multilayer classification label and HSPFiGs(H-STS-HSPFiGs).

In this paper, a band selection method based on a scalability tree structure multilayer classification label and HSPFiGs is proposed. To better describe the separability of ground objects with a mixed Gaussian distribution, we present an adaptive method to determine the number of subclass extensions, and extend a class of ground objects with mixed a Gaussian distribution to a series of single Gaussian distributions by subclass extension. Then, according to the result of subclass extension, we construct a multilayer classification label with a scalability tree structure. The second layer of the scalability tree is a mixed classification label, and the third layer contains a series of subclass extension labels. Then, the original classification labels of HSPFiGs are replaced by the scalability tree structure multilayer classification label, and the bands of the hyperspectral images are selected in hierarchical order to obtain a mixed optimal band subset and a series of subclasses of extended optimal band subset, which together constitute an optimal band subset family. Finally, the effectiveness of the proposed method is verified by comparing it with existing band selection methods. Experimental results based on datasets from Pavia University, Salinas, KSC and Botswana show that the method selects fewer bands and has higher overall accuracy, average accuracy and kappa coefficient than those of other methods. [ABSTRACT FROM AUTHOR]