Abstract: In order to better make full use of the texture information of hyperspectral image, a hyperspectral image classification method is proposed based on Multiscale Local Binary Pattern and Composite Kernel function (MLBPCK). Firstly, two optimal scales of the LBP are used to extract the texture features of the hyperspectral image, then the above spatial texture features are introduced into Gaussian kernel function for obtaining two spatial kernels. Moreover, the two spatial kernels and the spectral kernel obtained by introducing spectral information into Gaussian kernel function are combined to form a composite kernel. Finally, the composite kernel is input into a support vector machine (SVM) for classification to obtain classification results. Experimental results show that the classification accuracy on Indian Pines and Pavia University are 0.994 8 and 0.991 8, respectively, which is significantly better than other similar hyperspectral image classification methods.