| Summary: | Spectral-spatial feature extraction methods present an effective way for classification of hyperspectral images. However, performances of these methods may decrease depending on different data sets, classifier type, number of training samples, noise and smoothness level of data sets. In this paper, these terms are called as robustness criteria. This paper first discusses the effects of each robustness criteria for the feature extraction methods in the state-of-the-art. Secondly, a robust feature extraction method that can solve the aforementioned problems is proposed for hyperspectral image classification. The proposed method first performs dimensionality reduction and then extracts features using multiple edge preserving filters and intrinsic image decomposition method. Each robustness criteria is thoroughly examined in experiments and the results are compared with the state-of-the-art feature extraction methods. Experimental results demonstrate that the proposed method not only achieves better classification accuracies for different classifiers at lower training samples but also presents robust performance on different data sets that includes noise and smoothness effects.
|
|---|
