Hyperspectral image classification has become one of important hyperspectral applications, but how to get good classification results in the case of small sample size is still an open issue and attracts many research attentions. In recent year, deep learning has been used in the context of remote sensing image analysis. In this paper, we propose a new hyperspectral image classification method based on EMAPs (extended multiattribute profile) and polynomial networks (PN). Firstly, EMAPs can extract multilevel structures of morphological features by a series of attribute filters, which integrate the spatial and spectral information of remote sensing data. Then, the spatiospectral features are fed as the input to the deep PN composed of multilayer feed forward structure. PN decreases the training error layerbylayer in the sense to obtain good classification results. Classification experimental results on different hyperspectral image sets demonstrate that the proposed method outperforms other methods.