Joint multilayer spatial-spectral classification of hyperspectral images based on CNN and ConvLSTM

In this paper, a novel method based on convolutional long short-term and convolutional nerual network (MCNN-ConvLSTM) is proposed for hyperspectral image (HSI) classification. Firstly, due to powerful hierarchical feature extraction ability, CNN is devised to extract spatial features of HSIs. Then, shallow, middle and deep spatial features of CNN are used as the input of several ConvLSTMs. ConvLSTMs are devised to extract different layers of joint spatial-spectral features due to the ability of global spectral feature extraction. Finally, multilayer spatial-spectral features are fused to achieve an end-to-end classification, which learn complementary information among the shallow layers with basic information and the deep layers with abstract information. The experimental results demonstrate that the proposed algorithm can yield competitive classification performance compared with existing methods.

Recommended citation: Feng, J., Wu, X., Chen, J., Zhang, X., Tang, X., & Li, D. (2019, July). Joint multilayer spatial-spectral classification of hyperspectral images based on CNN and ConvLSTM. In IGARSS 2019-2019 IEEE International Geoscience and Remote Sensing Symposium (pp. 588-591). IEEE.
Download Paper