Determining the number of endmembers of hyperspectral images using clustering

Abstract

Some applications require knowing how many materials are present in the scene represented by a hyperspectral Image. In a previous paper, we presented an algorithm that estimated the number of materials in the scene using clustering principles. The proposed algorithm obtains a hierarchy of image partitions and selects a partition using a validation Index; the estimated number of materials is set to the number of dusters of the selected partition. In this algorithm, the user must provide the Image and the maximum number of materials that can be estimated (P). In this paper, we have extended our algorithm so that It does not require P as input parameter. The proposed method Iteratively performs the estimation for several increasing values of P and stops the process when a certain condition is met. The results obtained with five hyperspectral Images show that our algorithm approximately estimates the number of materials in that images.