Runmin Dong

Runmin Dong ｜ 董润敏

PhD candidate in
Department of Earth System Science
Tsinghua University

Email: drm17@mails.tsinghua.edu.cn

Education Background

Sep. 2017 – Present
Tsinghua University, Ph.D. candidate

• Supervisor: Prof. Haohuan Fu
• My research interests include remote sensing image understanding, deep learning, land cover mapping, image super-resolution, and self-supervised representation learning.

Sep. 2013 – Jun. 2017
Beijing Jiaotong University, B.S.

Intern Experience

May. 2019 – Sep. 2020
SenseTime, Remote Sensing Algorithm Intern

Jun. 2018 – Sep. 2018
Meituan Dianping, Algorithm Intern

Research Projects

Reference-based Super-Resolution for Remote Sensing Image

• This work aims to alleviate the problem that it is difficult for the single image super-resolution methods to reconstruct the fine texture of high-resolution images at large upscaling factors.
  
• We build an open-source reference-based remote sensing super-resolution dataset and propose a novel end-to-end reference-based super-resolution network (RRSNet).RRSNet can extract the Ref features and align them to the LR features and then transfer the texture information in the Ref features to the reconstruction of HR images.
  
• This work is accepted by IEEE Transactions on Geoscience and Remote Sensing.
  

Improving 3-m Resolution Land Cover Mapping through Efficient Learning from an Imperfect 10-m Resolution Map

• The high-resolution land cover mapping over large areas is a challenging task for lack of high-quality labels. A potential solution is to leverage the existing knowledge contained in the freely available lower-resolution land cover products. This work aims to produce a novel 3-m resolution land cover map through efficient learning from imperfect 10-m resolution maps without human efforts being involved.
• This work is published in Remote Sensing.

Oil Palm Plantation Mapping from High-resolution Remote Sensing Images

• In order to promote the development of oil palm plantation mapping from high spatial-resolution satellite imagery, we release a high-quality and pixel-level Malaysian oil palm plantation dataset (MOPPD).
• We propose a Residual Channel Attention Network, including the Residual Channel Attention Unit and the post-processing strategy. Our proposed method achieves the overall accuracy (OA) of 96.8% and mean IoU of 90.58%, improving the OA by 2.03%-3.96% and the mean IoU by 2.13%-5.44% compared with FCN,U-Net and FC-DenseNet.
• Our works are published in International Journal of Remote Sensing and SPIE Conference 2019.

Deep Learning Based Oil Palm Tree Detection from High-Resolution Satellite Images

• We propose a two-stage convolutional neural network (TS-CNN)-based oil palm detection method using high-resolution satellite images, including a CNN for land cover classification and a CNN for object classification.
• Our proposed approach achieves a much higher average F1-score of 95% in our study area compared with existing oil palm detection methods, and much fewer confusions with other vegetation and buildings in the whole image detection results.
• This work is published in Remote Sensing (Second Author).

Integrating Google Earth Images with Landsat Data to Improve 30-m Land Cover Mapping

• We integrate free and public Google Earth high-resolution images with Landsat data to improve 30m-resolution land cover mapping in China.Considering the characteristics of the Google Earth imagery and the Landsat data, we design a novel deep convolutional neural network based land cover mapping approach which takes full advantages of different sources of remote sensing data.
• Our proposed method achieves the classification accuracy of 84.4% on the whole validation dataset in China, improving the previous state-of-the-art accuracy by another 4.5%, with much fewer confusions between different vegetation types and the impervious type.
• Related work is published in Remote Sensing of Environment (Second Author).

Publications

[1]Runmin Dong, Lixian Zhang and Haohuan Fu*. RRSGAN: Reference-based Super-Resolution for Remote Sensing Image. IEEE Transactions on Geoscience and Remote Sensing. (Accepted)

[2]Runmin Dong, Cong Li, Haohuan Fu*, Jie Wang, Weijia Li, Yi Yao, Lin Gan, Le Yu, and Peng Gong*. Improving 3-m resolution land cover mapping through efficient learning from an imperfect 10-m resolution map. Remote Sensing, 2020, 12(9), 1418.

[3]Runmin Dong, Weijia Li*, Haohuan Fu, Lin Gan, and Maocai Xia. Oil palm plantation mapping from high-resolution remote sensing images using deep learning. International Journal of Remote Sensing, 2019, 41, 2022–2046.

Competitions & Awards

National Scholarship, Tsinghua University, 2020
2nd Place, Facebook AI Self-Supervised Learning Challenge, ICCV Workshop, 2019
Outstanding Intern, SenseTime, 2019