Morphological Convolution and Attention Calibration Network for Hyperspectral and LiDAR Data Classification

Reasonable fusion of multimodal data can increase the accuracy of remote sensing classification. In this article, an effective morphological convolution and attention calibration network is proposed for the joint classification of the hyperspectral image (HSI) and light detection and ranging (LiDAR). First, we devise a morphological convolution block, which combines the dilation and erosion operations in morphology with convolution to better capture the feature from the HSI and LiDAR. Next, we designed a dual attention module that uses self-attention to calibrate features and cross attention to combine multisource complementary information, respectively. Finally, considering the features of semantic inconsistency and different scales, the adaptive feature fusion module is introduced to dynamically fuse multimodal features. To verify the progressiveness of the proposed network, we experiment on three common datasets and one self-made dataset. The result shows that our network performs better than the state-of-the-art models.