STSNet: A cross-spatial resolution multi-modal remote sensing deep fusion network for high resolution land-cover classification

We have released an open-source dataset combining hyperspectral and high spatial resolution data with multi-temporal data (named WHU-H2SR-MT). To our knowledge, this is currently the largest dataset available for spatio-temporal-spectral interpretation. This dataset serves not only to validate the effectiveness of the algorithms presented in this paper but also to contribute to the research community in this field.The dataset has eight land-cover categories: paddy field, dry farmland, forest land, grassland, building, highway, greenhouse, and water body.  

The H2SR images was acquired in September 2020, with a spectral range between 391 nm and 984 nm, comprising 249 bands. The image was preprocessed by relative radiometric calibration and atmospheric correction. The H2SR images was resampled from 0.75 m to 1 m, and the spectral resolution of each band is ≤ 5 nm.

Multi-temporal Sentinel-2 images for the year 2020 within the experimental area were acquired, resulting in a total of 31 Sentinel-2 images after cloud removal operations. Ten land-cover relevant bands (B2, B3, B4, B5, B6, B7, B8, B8a, B11, B12) were selected, and resampled the bands with coarser spatial resolution (B5, B6, B7, B8a, B11, B12) to 10 m.

Citation:

Yu, B., Li, J., & Huang, X. (2024). STSNet: A cross-spatial resolution multi-modal remote sensing deep fusion network for high resolution land-cover segmentation. Information Fusion, 102689.