Material Classification of Hyperspectral Images Using Unsupervised Fuzzy Clustering Methods

Abstract

This paper presents a novel approach in classifying materials in Hyperspectral images. In particular, unlike other similar approaches in which every pixel in the image is mapped to one of the reference spectra, the proposed methods use the data itself to create clusters of pixels with the same material. This is done by using unsupervised fuzzy clustering methods. Here, two fuzzy clustering approaches have been addressed: Fuzzy C-Means clustering (FCM) and fuzzy relational clustering (FRC). The proposed methods can also solve the problem of identifying the objects for which the radiance of light makes it barely hard to identify them as a single object e.g., a pitched roof. The proposed methods have been applied on the CASI image and the results show that they can successfully classify the materials in the image.