Target detection approaches to hyperspectral image classification

Detection and classification are generally considered as two different separate areas. As a matter of fact, classification can be formulated as a multiple-target problem via a hypothesis testing problem from a statistical detection theory point of view where each class is specified by a particular target under a hypothesis. Two approaches can be taken to perform hyperspectral image classification (HSIC). One is to formulate an M-class HSIC as an M-hypotheses testing problem where each class is considered as a hypothesis to be tested. The other is a binary hypothesis testing problem where the null hypothesis specified by H₀ represents all classes other than the class to be classified and the alternative hypothesis specified by H₁ represents a class of interest (CI) to be classified. As a result, the class to be classified is considered as a signal to be detected under H₁ and all other classes are considered as noise under H₀. With this interpretation, this paper presents a hypothesis testing problem formulated by HSIC which extends the statistical detection theory to statistical HSIC.