2nd IEEE Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing

| 8 Marzo 2012
14 giugno 2010@18:00–15 giugno 2010@19:00

Development of a Network Based Method for Classification of Hyperspectral Data

V. Karathanassi, D. Sykas

Laboratory of Remote Sensing,

National Technical University of Athens, 9 Heroon Polytechniou St, 15780 Athens, Greece


Abstract-The paper presents a new logarithmic mixed pixel classification method. The method considers that each pure class is a source that creates a constant field and each pixel of the image presents a charge according to its relative position in the field. The value of each field is provided by a network which links the source with the other pure classes. Based on the principle that natural targets do not consist of equally distributed components, the Network Based Method (NBM) alerts the user for non-sampled pure spectral classes in the image scene. Experiments showed that the method is robust with clear and interpretable results and provides reliable abundance fractions. The results of these experiments were compared with the Sum to one Constraint Least Square Method (SCLS) in order to further evaluate the performance of the proposed algorithm.