FY-4B MultiModal: A Chinese Geostationary Meteorological Satellite Image-Text Retrieval Benchmark Dataset

It contains 828 RGB composite images (1024x1024 pixels) from the FY-4B AGRI instrument, spanning  13 weather-phenomenon categories: clear sky, cumulus, stratus, cirrus,    cumulonimbus, typhoon, rainstorm clouds, dust storm, fog/low cloud, snow  cover, severe convection, cloud systems, and others.

Each category is paired  with two Chinese-language meteorological text descriptions, yielding 26 cross-modal retrieval queries. Images were synthesized from FY-4B AGRI L1  channels 1 (0.47 μm), 2 (0.65 μm), and 3 (0.825 μm) via Lanczos resampling and percentile contrast stretching. Annotations were performed by domain experts at the Yunnan Provincial Meteorological Data Center.