Using a Laboratorial Hyperspectral Image for the Evaluation of Feature Reduction Methods for the Classification of High Dimensional Data

Thee rapid advances in hyperspectral sensing technology have made it possible to collect remote sensing data in hundreds of bands. However, the data analysis methods which have been successfully applied to multispectral data are often limited to achieve satisfactory results for hyperspectral data. The major problem is the high dimensionality, which deteriorates the classification due to the Hughes phenomenon. In order to avoid this problem a feature reduction process is inevitable. There are currently many different methods for feature reduction process in hyperspectral data. The feature selection methods pick the most informative features and discard the redundant features from the total set of features. Feature extraction methods, on the other hand, transform a large amount of information into a small number of transformed features. The decision boundary feature extraction (DBFE) and nonparametric weighted feature extraction method (NWFE) are two important approaches for feature extraction. Another group of feature reduction algorithms are based on the theory of multiple classifiers. Thus far, many different methods for the feature reduction process have been proposed but the validation of these algorithms has not yet been done on an appropriate image dataset. The main goal of this study is to have a good evaluation of these different feature reduction algorithms based on a laboratorial hyperspectral data. Selection of classes for the simulated target was based on the challenging point of different algorithms which are classifying targets with very similar spectral characteristics, targets with different shapes, targets with high different spectral characteristics or targets with high spatial variability. In this respect following the aforesaid criteria 22 classes were considered in the final simulated target. The feature reduction methods were compared using the test image. The consistency between the various methods is discussed as well as the implication of feature reduction on image classification.