Dimensionality Reduction and Classification feature using Mutual Information applied to Hyperspectral Images : A Filter strategy based algorithm

摘要

Hyperspectral images (HIS) classification is a high technical remote sensingtool. The goal is to reproduce a thematic map that will be compared with areference ground truth map (GT), constructed by expecting the region. The HIScontains more than a hundred bidirectional measures, called bands (or simplyimages), of the same region. They are taken at juxtaposed frequencies.Unfortunately, some bands contain redundant information, others are affected bythe noise, and the high dimensionality of features made the accuracy ofclassification lower. The problematic is how to find the good bands to classifythe pixels of regions. Some methods use Mutual Information (MI) and threshold,to select relevant bands, without treatment of redundancy. Others control andeliminate redundancy by selecting the band top ranking the MI, and if itsneighbors have sensibly the same MI with the GT, they will be consideredredundant and so discarded. This is the most inconvenient of this method,because this avoids the advantage of hyperspectral images: some preciousinformation can be discarded. In this paper we'll accept the useful redundancy.A band contains useful redundancy if it contributes to produce an estimatedreference map that has higher MI with the GT.nTo control redundancy, weintroduce a complementary threshold added to last value of MI. This process isa Filter strategy; it gets a better performance of classification accuracy andnot expensive, but less preferment than Wrapper strategy.