Exploiting Spatial-Spectral Feature for Hyperspectral Image Classification Based on 3-D CNN and Bi-LSTM

Hyperspectral remote sensing has been gaining more and more attention in recent years because of the rich spectral and spatial information contained in hyperspectral image (HSI). With the rapid development of deep learning, many deep learning methods have been applied to classify HSI. In the existing 3-D convolution methods, a widely used method is to project the original data into a low-dimensional subspace, so a small amount of the useful spectral information can be lost. To solve this problem, this paper propose a unified network framework using band grouping-based bidirectional long short-term memory (Bi-LSTM) network and 3-D convolutional neural network for HSI classification. In this framework, the issue of spectral feature extraction is considered as a sequence learning problem, and the Bi-LSTM as a spectral feature extractor is adopted to address it. To evaluate the performance of the proposed method, the Indian Pines remote sensing data sets are used for HSI classification experiments. The results demonstrate that the performance of proposed method is better than the state-of-the-art HSI classification methods.