Experimental Study of Hierarchical Clustering for Unmixing of Hyperspectral Images

Estimation of the number of materials that are present in a hyperspectral image is a necessary step in many hyperspectral image processing algorithms, including classification and unmixing. Previously, we presented an algorithm that estimated the number of materials in the image using clustering principles. This algorithm is an iterative approach with two input parameters: the initial number of materials
$(P_{0})$
and the number of materials added in each iteration
$(\Delta)$
. Since the choice of
$P_{0}$
and
$\Delta$
can have a large impact on the estimation accuracy. In this paper, we made an experimental study of the effect of these parameters on the algorithm performance. Thus, we show that the choice of a large
$\Delta$
can significantly reduce the estimation accuracy. These results can help to make an appropriate choice of these two parameters.