Comparative Study of Kernel Based Fuzzy Clustering Algorithms for Hyperspectral Data Clustering

Hyperspectral data, by precise sampling of objects reflectance in visible and near infrared spectrum in to numerous spectral bands, can prepare valuable data source for identification and recognition of different objects. One of the main applications of this data is classification for retrieving different land cover classes. However, numerous bands of this data, makes their classification a challenging task. Usual classifiers due to lack of sufficient amount of training datacannot have an acceptable performance.Thus, in recent years unsupervised classification algorithms have gained more attention. One of the emerging unsupervised algorithms in data mining context are kernel based fuzzy clustering algorithms (KFCMs),these algorithms, which are based on well known Fuzzy C-means algorithm and kernel function, usually, have better results for medical images and standard datasets. Dueto good performance of kernel based algorithms in hyperspectral image processing and the characteristics of FCM,the KFCMs algorithms seem good choices for hyperspectral data clustering. The objective of this paper is to study the performance of kernel based clustering algorithms for hyperspectral data clustering and comparing them with generic FCM algorithm. Because of great impact of kernel function on the performance of KFCMs