A Multi-Layer Binary Pattern Based Method for Hyperspectral Imagery Classification Using Combined Spatial-Spectral Characteristics

It is essential to make full use of the spatial features for hyperspectral imagery classification. A multi-layer binary pattern based method for hyperspectral remote sensing imagery classification using combined spatial-spectral characteristics is proposed. Firstly, the hyperspectral remote sensing imagery was transformed to local binary pattern image reflecting the micro-structures of the imagery, and multi-layer feature vectors were acquired from the local binary pattern image reflecting a wider range of the imagery structure. Here, in order to verify the proposed method, the PaviaU, the Salinas and the Chikusei imageries were used and nine kinds of features such as spectral features, local binary pattern spatial features, the multi-resolution local binary pattern features, etc., were employed in the experiment for hyperspectral remote sensing imagery spatial-spectral classification.The classifier of this algorithm used the kernel extreme learning machine classifier. The overall classification accuracies separately reached 97.31%, 98.96% and 97.85% based on the multi-layer binary patterns spatial-spectral features which were superior to the traditional methods such as spectral classification and spatial-spectral classification using 3Gabor features. The results indicate that the proposed approach can provide the effectively distinguished features showing the higher classification accuracy and the smoother classification map.