Research Article
Cross-Stage Fusion Network Based Multi-modal Hyperspectral Image Classification
@INPROCEEDINGS{10.1007/978-3-031-36011-4_7, author={Yuegong Sun and Zhening Wang and Ao Li and Hailong Jiang}, title={Cross-Stage Fusion Network Based Multi-modal Hyperspectral Image Classification}, proceedings={6GN for Future Wireless Networks. 5th EAI International Conference, 6GN 2022, Harbin, China, December 17-18, 2022, Proceedings, Part I}, proceedings_a={6GN}, year={2023}, month={7}, keywords={Multi-modal Feature Fusion Hyperspectral Image Classification Remote Sensing}, doi={10.1007/978-3-031-36011-4_7} }- Yuegong Sun
Zhening Wang
Ao Li
Hailong Jiang
Year: 2023
Cross-Stage Fusion Network Based Multi-modal Hyperspectral Image Classification
6GN
Springer
DOI: 10.1007/978-3-031-36011-4_7
Abstract
With the development of satellite technology and airborne platforms, there are more and more methods to acquire remote sensing data. The remote sensing data acquired by multiple methods contain different information and internal structures. Nowadays, single-mode hyperspectral image (HSI) data are no longer satisfactory for researchers’ needs. How to apply and process the information of multimodal data poses a great challenge to researchers. In this paper, we propose a deep learning-based network framework for multimodal remote sensing data classification, where we construct an advanced cross-stage fusion strategy using a fully connected network as the backbone, called CSF. Like the name implies, CSF incorporated two separate stages of fusion strategies for more effective fusion of multimodal data: fusion at the pre-structure and fusion at the tail of the network. This strategy prevents the preservation of excessive redundant information in the pre-fusion and the details of information lost due to late fusion. Moreover, a plug-and-play cross-fusion module for CSF is implemented. On the Houston 2013 dataset, our model strategy outperformed the fusion strategy of each stage and the single-modal strategy, which also demonstrated that multimodal feature fusion has promising performance.
