Dr. Wenyuan Li’s recent publication on Remote Sensing of Environment

Fine-grained hierarchical crop type classification from integrated hyperspectral EnMAP data and multispectral Sentinel-2 time series: A large-scale dataset and dual-stream transformer method

Congratulations to Dr. Wenyuan Li, Professor Shunlin Liang, and the research team on publishing a new article in Remote Sensing of Environment (Vol. 344, 115525). The paper is titled “Fine-grained hierarchical crop type classification from integrated hyperspectral EnMAP data and multispectral sentinel-2 time series: A large-scale dataset and dual-stream transformer method” and is available at https://doi.org/10.1016/j.rse.2026.115525. The product is freely accessible at www.glass.hku.hk.

Fine-grained crop type classification serves as the fundamental basis for large-scale crop mapping and plays a vital role in ensuring global food security. However, existing methods predominantly rely on multispectral satellite imagery (such as Sentinel-2), whose limited spectral resolution struggles to differentiate crops with highly similar morphological and phenological traits. To address this bottleneck, the team constructed the first Hierarchical Hyperspectral Crop dataset () by integrating 30-m EnMAP hyperspectral data with 10-m Sentinel-2 time series. They also proposed an innovative dual-stream Transformer architecture that coordinates two specialized pathways: a Spectral-Spatial Decoupled Vision Transformer to extract fine-grained biochemical signatures from EnMAP data, and a temporal Swin Transformer to capture crop growth patterns from Sentinel-2 time series.

The  dataset establishes a massive and unprecedented benchmark featuring over one million annotated field parcels organized in a four-tier crop taxonomy (ranging from 6 to 101 crop types). Extensive validation experiments demonstrate that incorporating hyperspectral EnMAP data into Sentinel-2 time series yields a 4.2% average F1 score improvement, peaking at an impressive 6.3% increase for fine-grained classification levels. The proposed dual-stream Transformer method significantly outperforms conventional deep learning architectures (such as UNet, 3DCNN, and CNN-LSTM) and proves highly effective at distinguishing subtle spectral-temporal interactions even under varying temporal windows and dynamic crop rotation scenarios.

Spatial distribution and representative samples of the  dataset across France, showcasing Sentinel-2 imagery, hierarchical crop labels, and EnMAP spectral curves.

The overall dual-stream Transformer network architecture, featuring a spectral-spatial decoupled Vision Transformer and a multispectral temporal Transformer for precise hierarchical crop classification.

Reference:

Li, W., Liang, S., Zhang, Y., Liu, L., Chen, K., Chen, Y., Ma, H., Xu, J., Ma, Y., Guan, S., & Shi, Z. (2026). Fine-grained hierarchical crop type classification from integrated hyperspectral EnMAP data and multispectral Sentinel-2 time series: A large-scale dataset and dual-stream transformer method. Remote Sensing of Environment, 344, 115525. https://doi.org/10.1016/j.rse.2026.115525

Keywords:  Crop type classification, Precision agriculture, Hyperspectral data, Sentinel-2 time series, Deep learning