A Novel Fuzzy Linear Feature Extraction for Hyperspectral Image Classification

In this paper, a novel fuzzy linear discriminant analysis feature extraction method has been proposed for classifying hyperspectral image data. We used the fuzzification mechanism proposed by Keller, et al. to discover samples close to the class boundary or not, and different weights are assigned on samples when estimating the within-class and between-class scatter matrix. The effectiveness of the proposed feature extraction scheme as compared to nonparametric weighted feature extraction (NWFE) and linear discriminant analysis (LDA) is demonstrated using Washington DC Mall data, and achieve a satisfactory performance.