Hyperspectral image classification is a critical issue in hyperspectral data processing. However, the task has been acknowledged as extremely challenging due to its characteristics including high dimensionality in data, spatial variability of spectral features and scarcity of marked data. In this paper, we propose a new classification method combined with Local Binary Patterns (LBP) and Singular Value Decomposition Networks (SVDNet). Linear Prediction Error is first employed to select informative spectral bands. Then LBP is utilized to extract the texture features. After that, the extracted features of a specified field are transformed to 2-D images. Finally, SVDNet classifies the obtained images and then the classification result can be obtained. Experimental results on the real hyperspectral dataset demonstrate that the proposed method is capable to achieve higher classification accuracy or at least comparable to existing methods.
|