The detailed vegetation classification for airborne hyperspectral remote sensing imagery by combining PCA and PP

The feature extraction and dimensionality reduction is one of the core problems in hyperspectral remote sensing imagery processing. For the detailed vegetation classification, a projection index is established. It describes the separability of easy mixed classified vegetation objects. By optimizing the index, the projection directions may be calculated and the directions are orthogonal each other. The feature subspace of full data space may be constructed by combining principal components directions and the projection pursuit directions. The classification is completed on the feature subspace. It is hopeful to increase the classification accuracy especially the accuracy of easy mixed classified objects by the strategy. To verify the conclusion, a classification experiment is completed on an airborne hyperspectral imagery, the result shows that the overall classification accuracy promote 7% and the accuracy of easy mixed classified objects promote more than 20%.