Local Averaging Based Feature Extraction on Hyperspectral Image Data

Abstract

This paper focuses on the land cover/usage area classification problem by using local averaging for feature extraction method. In hyperspectral image classification tasks, spatial information is also useful as much as spectral information. A pipeline of methods is utilized using Fisher's discriminant analysis for dimension reduction, z-score value for central limiting and support vector machines and extreme learning machines for classification. The classification accuracies on transformed data set are outperforming previous works by achieving % 99.51 success ratio on for support vector machines and % 99.73 for extreme learning machines on 10-fold cross validation. the proposed method increases classification accuracy significantly while reducing the dimension of the original data by % 95.