Multi-perspective regional continuity alignment network for hyperspectral and LiDAR image fusion and classification

In these decades, remote sensing (RS) techniques have been widely used when facing diverse real-world tasks. These unique RS theories tend to observe on-ground objects and materials by several techniques from different perspectives. Generally, several works have verified that land cover classification tasks are always hampered when using one single RS modality owing to the lack of diverse information. In this paper, we propose a multi-perspective regional continuity aligning network (MPRCAnet) for hyperspectral and light detection and ranging (LiDAR) image fusion and classification. Our motivation is to capture and align the regional spatial-spectral continuities from different perspectives for better performance. Specifically, the whole network consists of three parallel streams, including 1) the HS spatial stream, 2) the HS spectral stream and 3) the LiDAR spatial stream. This network aims to discuss the potential of the continuous information extraction strategy in related tasks. Multiple experiments are conducted on two publicly available hyperspectral and LiDAR datasets to prove the effectiveness and feasibility of the proposed MP-RCAnet. Compared with several state-of-the-art techniques, MP-RCAnet is capable of generating fusion and classification results with higher evaluation accuracies. Its classification map has less scattered noise than other techniques.