A Spectral-Spatial Classification Network for Hyperspectral Images using a Residual Attention Network

In recent years, hyperspectral image (HSI) classification has been considered by the remote sensing community for its academic significance and potential wide applications. Deep learning algorithms, particularly convolutional neural networks (CNNs), have significantly improved the performance of hyperspectral classification. However, due to the high dimensionality of HSI and limited training samples, the deep neural network causes model overfitting. Additionally, the increased model parameter, considering all the bands of HSI datasets equally for feature learning, reduces classification accuracy. Thus, in this paper, we propose an end-to-end spectral-spatial residual attention network for hyperspectral image classification (S2RANHC). The residual attention network maintains the original spectral and spatial information intact without increasing the feature operations. It constructs deeper neural networks to overcome the vanishing gradient problem. The proposed method has been evaluated on three different datasets to illustrate state-of-the-art performance for various land use land cover scenarios. A comprehensive qualitative and quantitative analysis of the results shows that the proposed method outperforms other state-of-the-art methods.