Classification And Reduction Of Hyperspectral Images Based On Motley Method

The Hyperspectral image is a substitution of more than one hundred images of the same region called bands. To exploit the richness of this image it is necessary to reduce its dimensionality. The question is how to reduce it? This high dimensionality is also a source of confusion and adversely affects the accuracy of the classification bands and increases the compute time. Many methods are introduced to solve this confusion. Our proposed approaches are based on three-step. As a first principle, we start by scheduling according to a Mutual information criterion. Then the selection according to the filter in which we retried two filters MidMax for the first algorithm and MIFSU for the second. In the last step we added a wrapper strategy which is based on the probability of error. The classifier we had chosen is the Support Vector Machine (SVM). This algorithm gets high performance of classification accuracy. The study is established on HSI AVIRIS 92AV3C.