Modular eigen subspace scheme for high-dimensional data classification with NASA MODIS/ASTER (MASTER) airborne simulator data sets of Pacrim II project

In this paper, a novel filter-based greedy modular subspace (GMS)technique is proposed to improve the accuracy of high-dimensional remote sensing image supervisor classification. The approach initially divides the whole set of high-dimensional features into several arbitrary number of highly correlated subgroup by performing a greedy correlation matrix reordering transformation for each class. These GMS can be regarded as a unique feature for each distinguishable class in high-dimensional data sets. The similarity measures are next calculated by projecting the samples into different modular feature subspaces. Finally, a supervised multi-class classifer which is implemented based on positive Boolean function (PBF) schemes is adopted to build a non-linear optimal classifer. A PBF is exactly one sum-of-product form without any negative components. The PBF possesses the well-known threshold decomposition and stacking properties. The classification errors can be calculated from the summation of the absolute errors incurred at each level. The optimal PBF are found and designed as a classifer by minimize the classification error rate among the training samples. Experimental results demonstrate that the proposed GMS feature extraction method suits the PBF classifer best as a classification preprocess. It signifcantly improves the precision of image classification compared with conventional feature extraction schemes. Moreover, a practicable and convenient "vague" boundary sampling property of PBF is introduced to visually select training samples from high-dimensional data sets more effciently.