Awesome

BigImageMapper (Landuse_DL)

Land use classification or landform delineation from big remote sensing imagery using Deep Learning. This repo contains codes for mapping thermokarst landforms in cold regions and others.

Citation

If the scripts here are useful for your project, please consider citing our papers:

The scripts for producing our results in the paper Huang et al. 2020 can be found at the RSE2020paper Branch.

@article{huang2023identifying,
  title={Identifying active retrogressive thaw slumps from ArcticDEM},
  author={Huang, Lingcao and Willis, Michael J and Li, Guiye and Lantz, Trevor C and Schaefer, Kevin and Wig, Elizabeth and Cao, Guofeng and Tiampo, Kristy F},
  journal={ISPRS Journal of Photogrammetry and Remote Sensing},
  volume={205},
  pages={301--316},
  year={2023},
  publisher={Elsevier}
}

@article{huang2022accuracy,
  title={Accuracy, Efficiency, and Transferability of a Deep Learning Model for Mapping Retrogressive Thaw Slumps across the Canadian Arctic},
  author={Huang, Lingcao and Lantz, Trevor C and Fraser, Robert H and Tiampo, Kristy F and Willis, Michael J and Schaefer, Kevin},
  journal={Remote Sensing},
  volume={14},
  number={12},
  pages={2747},
  year={2022},
  publisher={MDPI}
}

@article{huang2020using,
  title={Using Deep Learning to Map Retrogressive Thaw Slumps in the Beiluhe Region (Tibetan Plateau) from CubeSat Images},
  author={Huang, Lingcao and Luo, Jing and Lin, Zhanju and Niu, Fujun and Liu, Lin},
  journal={Remote Sensing of Environment},
  volume = {237},
  pages={111534},
  year = {2020},
  publisher={ELSEVIER},
  doi = {https://doi.org/10.1016/j.rse.2019.111534}
}

@article{huang2018automatic,
  title={Automatic Mapping of Thermokarst Landforms from Remote Sensing Images Using Deep Learning: A Case Study in the Northeastern Tibetan Plateau},
  author={Huang, Lingcao and Liu, Lin and Jiang, Liming and Zhang, Tingjun},
  journal={Remote Sensing},
  volume={10},
  number={12},
  pages={2067},
  year={2018},
  publisher={Multidisciplinary Digital Publishing Institute}
}

How to install and use

See tutorial and the script: working_dir/exe.sh

Contributions

Please let me know or pull a request if you spot any bug or typo. Thanks! Any enhancement or new functions are also welcome!

updates

Ocotber 2021: Add CUT to the workflow, which uses the power of generative adversarial networks to translate images from target domain to source domain.

January 2021: Clean or re-organize the scripts; re-define the input parameter files, separate them into main ini, network defined ini, and regions defined ini, making it easy to different regions or networks for training and inference; split data into training (90%) and validation (10%) for checking the overfitting issue. output mIOU during training and allow early stopping.

March 2019: Many scripts for producing figures of manuscript: Using Deep Learning to Map Retrogressive Thaw Slumps in the Beiluhe Region (Tibetan Plateau) from CubeSat Images, Remote sensing of Environment.

January 2019: Supporting of Mask RCNN

August 2018: Delineate retrogressive thaw slumps from Planet CubeSat images.

March 2018: Land use classification using the data from 2018_IEEE_GRSS_Data_Fusion. Also submitted the result. codes in "grss_data_fusion". The method utilized Deeplab V4(+3), a semantic segmentation algorithm, to classify land use (20 classes).

Dependencies and setting:

Python package: Numpy, rasterio, GDAL 2.3, tensorflow-gpu 1.6, pyshp 1.2.12, pillow, imgaug
Other: GDAL, OTB, ASP, CUDA 9.0, cudnn 7.0.
More information on the setting can be found in 'docker_ubuntu1604/run_INsingularity_hlctest.sh' and 'tutorial/Setting_running_on_ITSC.md'

Disclaimer

This is a personal repo that we are actively developing. It may not work as expected. We have all the settings on our workstations and servers. You need to spend some efforts on environment settings on your computers before running these codes. These codes are only for research, and please take risks by yourself.

TODO

We will update some of the codes, including bug fix and enhancement because they are used in other projects.

Better documents