Nearest Neighbor-Based Contrastive Learning for Hyperspectral and LiDAR Data Classification

The joint hyperspectral image (HSI) and light detection ranging (LiDAR) data classification aims to interpret ground objects at more detailed precise level. Although deep learning methods have shown remarkable success in the multisource task, self-supervised has rarely been explored. It is commonly nontrivial build a robust model for classification, due fact that semantic similarities of neighborhood regions are not exploited existing contrastive framework. Furthermore, heterogeneous gap induced by inconsistent distribution impedes performance. To overcome these disadvantages, we propose nearest neighbor-based network (NNCNet), which takes full advantage large amounts unlabeled learn discriminative feature representations. Specifically, augmentation scheme use enhanced relationships among nearby regions. intermodal alignments can be captured accurately. In addition, design bilinear attention module exploit second-order even high-order interactions between HSI LiDAR data. Extensive experiments on four public datasets demonstrate superiority our NNCNet over state-of-the-art methods. source codes available https://github.com/summitgao/NNCNet .