Ensemble Classification of Hyperspectral Images by Integrating Spectral and Texture Features

Hyperspectral images provide abundant spectral information of the land surface materials, which make it possible to distinguish those materials with subtle difference. Researches about improvement in classification accuracy of hyperspectral images have been conducted from two aspects. The first one is trying to integrating other features such as texture features, geometry features and geographical information. The second one focuses on the employment of advance classifiers such as random forest classifier, support vector machine classifier, decision tree classifier. This paper demonstrated a recent study about the ensemble classification of hyperspectral image by integrating spectral features and texture features. Morphology texture features were extracted from the principal components of the hyperspectral bands. Multi-size of structure elements was used to get the morphology texture images by implementing the closing and opening operation. Texture features extracted from the gray-level co-occurrence matrix were also utilized to classify the hyperspectral images. Four classifiers were trained using spectral features, texture features, combined spectral and texture features, respectively. It was found that a single feature induced relatively poor classification accuracy in all of the four classifiers. Integrated spectral–texture features generated improved classification results. On the other hand, ensemble classification could produce much better classification effects compared with a single classifier. Further works will focus on the classification performance of other features such as wavelet texture feature and context information.