Ph.D. student, The University of Tokyo
I work on remote sensing agents and recursive self-improvement (RSI).
I am currently pursuing my Ph.D. degree at the University of Tokyo, advised by Prof. Naoto Yokoya. Before that, I received my M.S. degree from Nanjing University, where I worked under the supervision of Prof. Xueliang Zhang and Prof. Pengfeng Xiao. My current research interests include remote sensing agents, geospatial reasoning, disaster remote sensing, and Earth observation. Feel free to send me an email if you would like to chat with me for any reason.
* equal contribution † corresponding author
IEEE Transactions on Geoscience and Remote Sensing (TGRS), 2026
This research introduces TSSUN, a novel deep learning model that unifies diverse remote sensing data and dense prediction tasks. It addresses data heterogeneity challenges by standardizing input/output, achieving state-of-the-art performance across various applications without task-specific modifications.
arXiv, 2026
OpenEarth-Agent is a tool-creation agent framework for open-environment Earth Observation, designed to handle diverse multi-source data and heterogeneous tasks beyond the limits of closed, predefined tool-calling systems. It adaptively plans workflows, creates task-specific tools, integrates multi-stage tools and cross-domain knowledge, and is evaluated with OpenEarth-Bench, demonstrating robust full-pipeline EO performance across multiple application domains.
International Conference on Machine Learning (ICML), 2026
This research introduces WLA, a novel deep learning model that compresses massive weather datasets into a compact latent space. This innovation significantly reduces data storage and computational costs, while improving the accuracy and adaptability of weather task models across diverse scenarios.
IEEE Transactions on Geoscience and Remote Sensing (TGRS), 2025
This research introduces VegeDiff to forecast future vegetation states by employing a novel diffusion model to probabilistically capture uncertainties in vegetation change. It accurately models dynamic meteorological and static environmental impacts, providing clear, precise predictions, outperforming existing deterministic methods.
IEEE Transactions on Geoscience and Remote Sensing (TGRS), 2024
This research introduces Remote Sensing Mamba (RSM) to efficiently model global context in large remote sensing images. RSM overcomes the quadratic complexity of transformers by using an omnidirectional selective scan, achieving state-of-the-art dense prediction performance on VHR images.
IEEE Transactions on Geoscience and Remote Sensing (TGRS), 2023
This research introduces a novel exchanging dual encoder-decoder structure for binary change detection. It addresses limitations of existing methods by fusing bitemporal features at the decision level and leveraging bitemporal semantic features. The proposed model achieves superior performance and high efficiency across various change detection scenarios.
The University of Tokyo
Ph.D. in Department of Complexity Science and Engineering (Supervised by Prof. Naoto Yokoya)
Nanjing University
M.S. in School of Geography and Ocean Science (Supervised by Prof. Xueliang Zhang and Prof. Pengfeng Xiao)
Nanjing University
B.S. in School of Geography and Ocean Science
Reviewer for journals TCSVT, TGRS, JSTARS, and for conferences CVPR, ICLR, AAAI.
