 The proposed BICLSTM network is a novel deep learning framework for automatic spectral spatial feature extraction from hyperspectral images. It combines the advantages of bi-directional recurrent connections and convolutional neural networks to achieve better spectral spatial feature extraction than existing methods. Experimental results demonstrate that BICLSTM outperforms other state-of-the-art methods in terms of classification accuracy. This article was authored by Qing Shan Lu, Feng Zhou, Ren Long Heng, and others.