 This paper presents a novel deep learning framework called by CLSTM for automatically extracting spectral and spatial features from hyperspectral images, HSICE. The proposed network uses a recurrent connection operator across the spectral domain to address the issue of spectral feature extraction, while a convolution operator is incorporated into the network to extract spatial features. A bi-directional recurrent connection is also proposed to capture the spectral information. In the classification phase, the learned features are concatenated and fed to a softmax classifier via a fully connected operator. The bi-CLSTM framework outperforms six state-of-the-art methods, including 3D, CNN, by improving the classification performance by almost 1.5%. This article was authored by Qin Shanlu, Fom Zhou, Ra Longhang, and others.