Hyperspectral Image Classification Based on Bidirectional Recurrent Neural Network

In recent years, the rise of machine learning algorithms provides a good tool for processing hyperspectral data. A series of machine learning algorithms have served for the classification of hyperspectral images. Derived from these methods that regarding spectral segments of each pixel as a spectral sequence. Recurrent Neural Network (RNN) showing better processing capability for sequence data play an important role in hyperspectral data classification. The standard unidirectional RNN, however, only focus on the current input and the memory state of the past, and cannot connect to the future memory. Alternatively, in this paper, bidirectional RNN(BiRNN) is employed for the classification of hyperspectral images for the future memory. BiRNN can integrate the past memory and future memory state. The proposed method is applied to a classical hyperspectral data set, the performance of classification is better.