Multi-scale hierarchical cross fusion network for hyperspectral image and LiDAR classification

As the satellite sensor technology continues to advance, acquiring hyperspectral images (HSI) and Light Detection and Ranging (LiDAR) data have become increasingly accessible. Consequently, the fusion of these two types of data has become very popular in land cover classification tasks. However, there are still some deficiencies in the current methods, such as insufficient extraction of distinctive features of multi-source data, and inadequate fusion caused by poor use of the complementarity and cooperation in multi-source data. Therefore, we propose a multi-scale hierarchical cross fusion network (MSHCFNet) for the fusion classification of HSI and LiDAR. It includes four parts. Firstly, the multi-scale cascading feature extraction module is used to extract fully multi-scale spatial features. Secondly, the hierarchical fusion module is used to integrate spatial features by cross fusion and group fusion. Thirdly, the spectral signature enhancement module is used to enhance the spectral signature, so that the discriminant information in the spectral signature can be used effectively. Finally, the decision fusion module is used to classify the types of ground objects. A large number of experiments on three publicly available datasets show that the proposed method has the most advanced classification efficacy compared with existing methods.