As the MOSAIC Research Group, we are pleased to announce that our new study has been published in the IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (JSTARS):

📌 Introducing MOSAIC-SEN2-CC: A Multispectral Dataset and Adaptation Framework for Remote Sensing Change Captioning

🔗 https://ieeexplore.ieee.org/document/11181102

The MOSAIC-SEN2-CC study has successfully filled an important gap in remote sensing by bringing the task of remote sensing image change captioning (RSICC) to Sentinel-2 multispectral (MS) data for the first time. In this work conducted by the MOSAIC research group, a new MOSAIC-SEN2-CC dataset is introduced, consisting of 5,232 MS image pairs and 26,160 change captions covering seven change types and one no-change class. The dataset includes events such as floods, fires, agricultural and urban changes between 2017 and 2023, spanning 89 countries and 6 continents.

The paper proposes the MSICC framework, which redesigns baseline models such as RSICCformer, Chg2Cap, and PSNet with MS encoders pretrained on BigEarthNet data, and through systematic experiments, reveals the contribution of out-of-visible-spectrum information to the change captioning task.