Hyperspectral image classification method combining fast guided filtering and spatial neighborhood information

In order to solve the phenomenon of edge pixel disorder and different spectra characteristics with the same object caused by traditional hyperspectral classification methods which only pay attention to spectral information but ignore spatial information, a hyperspectral image classification algorithm combining fast guided filtering and spatial neighborhood information (FGF-SVM-SNI) is proposed. The method first uses a fast guide filter to extract the spatial texture information of hyperspectral images after dimensional reduction by principal component analysis and keep the edge details, the filtered spatial texture information is combined with the spectral information to form the spatial spectrum information, which is then classified by the support vector machine. Then a kind of hyperspectral pixel neighborhood information is designed to construct hyperspectral spatial correlation information to reduce "pocking mark" phenomenon and further optimize the classification results. The experimental results in The Indian Pines data set and Pavia University data set show that the proposed method (FGF-SVM-SNI) is significantly more accurate than the SVM classification method using only spectral information, which fully demonstrates the effectiveness of the proposed method.