Deep Fusion of Hyperspectral and LiDAR Images Using Attention-Based CNN

In this paper, an accurate and fast fusion technique for combining the spectral and spatial contents of hyperspectral images (HSI) with the height information of light detection and ranging (LiDAR) is proposed to increase the classification accuracy of HSI. First, a sub-network consists of a hybrid 3D/2D convolutional neural networks (CNNs) and two attention modules is applied to HSI to extract the spectral-spatial features. These modules are used to enhance CNN’s focus on more informative features. Second, the representative elevation features of LiDAR are exploited using a sub-network that is constructed by a 2D-CNN and an attention module. Third, the spectral-spatial information of HSI and the obtained elevation features of LiDAR images are integrated and classified using the softmax classifier. The proposed method is applied to two real datasets and compared with some state-of-the-art fusion algorithms. The experimental results show that the proposed model increases classification accuracy and decreases computational complexity. © 2022, The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd.