Deep learning pipeline to cloud mask VHR imagery
Project description
vhr-cloudmask
Python library to perform semantic segmentation of clouds and cloud shadows using very-high resolution remote sensing imagery by means of GPUs and CPU parallelization for high performance and commodity base environments.
- GitHub repo: https://github.com/nasa-nccs-hpda/vhr-cloudmask
- Documentation: https://nasa-nccs-hpda.github.io/vhr-cloudmask
Objectives
- Library to process remote sensing imagery using GPU and CPU parallelization.
- Machine learning and deep learning cloud segmentation of VHR imagery.
- Large-scale image inference of very high-resolution imagery.
Background
The detection of clouds is one of the first steps in the pre-processing of remotely sensed data. At coarse spatial resolution (> 100 m), clouds are bright and generally distinguishable from other landscape surfaces. At very high-resolution (< 3 m), detecting clouds becomes a significant challenge due to the presence of smaller features, with spectral characteristics similar to other land cover types, and thin (partially transparent) cloud forms. Furthermore, at this resolution, clouds can cover many thousands of pixels, making both the center and boundaries of the clouds prone to pixel contamination and variations in the spectral intensity. Techniques that rely solely on the spectral information of clouds underperform in these situations.
In this study, we propose a multi-regional and multi-sensor deep learning approach for the detection of clouds in very high-resolution WorldView satellite imagery. A modified UNet-like convolutional neural network (CNN) was used for the task of semantic segmentation in the regions of Vietnam, Senegal, and Ethiopia strictly using RGB + NIR spectral bands. In addition, we demonstrate the superiority of CNNs cloud predicted mapping accuracy of 81–91%, over traditional methods such as Random Forest algorithms of 57–88%. The best performing UNet model has an overall accuracy of 95% in all regions, while the Random Forest has an overall accuracy of 89%. We conclude with promising future research directions of the proposed methods for a global cloud cover implementation.
Getting Started
The main recommended avenue for using vhr-cloudmask is through the publicly available set of containers provided via this repository. If containers are not an option for your setup, follow the installation instructions via PIP.
Downloading the Container
All Python and GPU depenencies are installed in an OCI compliant Docker image. You can download this image into a Singularity format to use in HPC systems.
singularity pull docker://nasanccs/vhr-cloudmask:latest
In some cases, HPC systems require Singularity containers to be built as sandbox environments because of uid issues (this is the case of NCCS Explore). For that case you can build a sandbox using the following command. Depending the filesystem, this can take between 5 minutes to an hour.
singularity build --sandbox vhr-cloudmask docker://nasanccs/vhr-cloudmask:latest
If you have done this step, you can skip the Installation step since the containers already come with all dependencies installed.
Installation
vhr-cloudmask can be installed by itself, but instructions for installing the full environments are listed under the requirements directory so projects, examples, and notebooks can be run.
Note: PIP installations do not include CUDA libraries for GPU support. Make sure NVIDIA libraries are installed locally in the system if not using conda.
vhr-cloudmask is available on PyPI. To install vhr-cloudmask, run this command in your terminal or from inside a container:
pip install vhr-cloudmask
If you have installed vhr-cloudmask before and want to upgrade to the latest version, you can run the following command in your terminal:
pip install -U vhr-cloudmask
Running Inference of Clouds
Use the following command if you need to perform inference using a regex that points to the necessary files and by leveraging the default global model. The following is a singularity exec command with options from both Singularity and the cloud masking application.
Singularity options:
- '-B': mounts a filesystem from the host into the container
- '--nv': mount container binaries/devices
vhr_cloumask_cli options:
- '-r': list of regex strings to find geotiff files to predict from
- '-o': output directory to store cloud masks
- '-s': pipeline step, to generate masks only we want to predict
singularity exec --nv -B $NOBACKUP,/explore/nobackup/people,/explore/nobackup/projects \
/explore/nobackup/projects/ilab/containers/vhr-cloudmask.sif vhr-cloudmask-cli \
-o '/explore/nobackup/projects/ilab/test/vhr-cloudmask' \
-r '/explore/nobackup/projects/3sl/data/Tappan/Tappan16*_data.tif' '/explore/nobackup/projects/3sl/data/Tappan/Tappan15*_data.tif' \
-s predict
To predict via slurm for a large set of files, use the following script which will start a large number of jobs (up to your processing limit), and process the remaining files.
for i in {0..64}; do sbatch --mem-per-cpu=10240 -G1 -c10 -t05-00:00:00 -J clouds --wrap="singularity exec --nv -B $NOBACKUP,/explore/nobackup/people,/explore/nobackup/projects /explore/nobackup/projects/ilab/containers/vhr-cloudmask.sif vhr-cloudmask-cli -o '/explore/nobackup/projects/ilab/test/vhr-cloudmask' -r '/explore/nobackup/projects/3sl/data/Tappan/Tappan16*_data.tif' '/explore/nobackup/projects/3sl/data/Tappan/Tappan15*_data.tif' -s predict"; done
Infrastructure
The vhr-cloudmask package is a set of CLI tools and Jupyter-based notebooks to manage and structure the validation of remote sensing data. The CLI tools can be run from inside a container or from any system where the vhr-cloudmask package is installed.
The main system requirements from this package are a system with GPUs to accelerate the training and inference of imagery. If no GPU is available, the process will continue as expected but with a large slowdown. There are no minimum system memory requirements given the sliding window procedures implemented in the inference process.
Package Structure
├── archives <- Legacy code stored to historical reference
├── docs <- Default documentation for working with this project
├── images <- Store project images
├── notebooks <- Jupyter notebooks
├── examples <- Examples for utilizing the library
├── requirements <- Requirements for installing the dependencies
├── scripts <- Utility scripts for analysis
├── vhr_cloudmask <- Library source code
├── README.md <- The top-level README for developers using this project
├── CHANGELOG.md <- Releases documentation
├── LICENSE <- License documentation
└── setup.py <- Script to install library
Data Locations where this Workflow has been Validated
The vhr-cloudmask workflow has been validated in the following study areas using WorldView imagery. Additional areas will be included into our validation suite as part of upcoming efforts to improve the scalability of our models.
- Senegal
- Vietnam
- Ethiopia
- Oregon
- Alaska
- Whitesands
- Siberia
Development Pipeline Details
When performing development (training a model, preprocessing, etc.), we want to run from the dev container so we can add the Python files to the PYTHONPATH. The following commmand is an example command to run inference given a configuration file.
singularity exec --env PYTHONPATH="$NOBACKUP/development/tensorflow-caney:$NOBACKUP/development/vhr-cloudmask" \
--nv -B $NOBACKUP,/explore/nobackup/people,/explore/nobackup/projects \
/explore/nobackup/projects/ilab/containers/vhr-cloudmask.sif \
python $NOBACKUP/development/vhr-cloudmask/vhr_cloudmask/view/cloudmask_cnn_pipeline_cli.py \
-c $NOBACKUP/development/vhr-cloudmask/projects/cloud_cnn/configs/production/cloud_mask_alaska_senegal_3sl_cas.yaml \
-s predict
If you do not have access to modify the configuration file, or just need to perform small changes to the model selection, the regex to the files to predict, or the output directory, manually specify the arguments to the CLI file:
singularity exec --env PYTHONPATH="$NOBACKUP/development/tensorflow-caney:$NOBACKUP/development/vhr-cloudmask" \
--nv -B $NOBACKUP,/explore/nobackup/people,/explore/nobackup/projects \
/explore/nobackup/projects/ilab/containers/vhr-cloudmask.sif \
python $NOBACKUP/development/vhr-cloudmask/vhr_cloudmask/view/cloudmask_cnn_pipeline_cli.py \
-c $NOBACKUP/development/vhr-cloudmask/projects/cloud_cnn/configs/production/cloud_mask_alaska_senegal_3sl_cas.yaml \
-o '/explore/nobackup/projects/ilab/test/vhr-cloudmask' \
-r '/explore/nobackup/projects/3sl/data/Tappan/Tappan16*_data.tif' '/explore/nobackup/projects/3sl/data/Tappan/Tappan15*_data.tif' \
-ib B G R N G1 G2 \
-ob B G R N \
-ps sieve smooth fill dilate \
-s predict
Manual Testing
For manual testing, you can always call the pytests component using the dev container for development. The following is an example of manually testing the package components.
singularity exec --env PYTHONPATH="$NOBACKUP/development/tensorflow-caney:$NOBACKUP/development/vhr-cloudmask" --nv -B $NOBACKUP,/explore/nobackup/people,/explore/nobackup/projects,/css/nga /explore/nobackup/projects/ilab/containers/vhr-cloudmask.sif pytest $NOBACKUP/development/vhr-cloudmask/tests
Authors
- Jordan Alexis Caraballo-Vega, jordan.a.caraballo-vega@nasa.gov
- Caleb S. Spradlin, caleb.s.spradlin@nasa.gov
- Margaret Wooten, margaret.wooten@nasa.gov
Contributors
- Andrew Weis, aweis1998@icloud.com
- Brian Lee, brianlee52@bren.ucsb.edu
Contributing
Please see our guide for contributing to vhr-cloudmask. Contributions are welcome, and they are greatly appreciated! Every little bit helps, and credit will always be given.
You can contribute in many ways:
Report Bugs
Report bugs at https://github.com/nasa-nccs-hpda/vhr-cloudmask/issues.
If you are reporting a bug, please include:
- Your operating system name and version.
- Any details about your local setup that might be helpful in troubleshooting.
- Detailed steps to reproduce the bug.
Fix Bugs
Look through the GitHub issues for bugs. Anything tagged with "bug" and "help wanted" is open to whoever wants to implement it.
Implement Features
Look through the GitHub issues for features. Anything tagged with "enhancement" and "help wanted" is open to whoever wants to implement it.
Write Documentation
vhr-cloudmask could always use more documentation, whether as part of the official vhr-cloudmask docs, in docstrings, or even on the web in blog posts, articles, and such.
Submit Feedback
The best way to send feedback is to file an issue at https://github.com/nasa-nccs-hpda/vhr-cloudmask/issues.
If you are proposing a feature:
- Explain in detail how it would work.
- Keep the scope as narrow as possible, to make it easier to implement.
- Remember that this is a volunteer-driven project, and that contributions are welcome :)
References
Tutorials will be published under Medium for additional support and development, including how to use the library or any upcoming releases.
Please consider citing this when using vhr-cloudmask in a project. You can use the citation BibTeX to site bot the software and the article:
Paper
@article{caraballo2023optimizing,
title={Optimizing WorldView-2,-3 cloud masking using machine learning approaches},
author={Caraballo-Vega, JA and Carroll, ML and Neigh, CSR and Wooten, M and Lee, B and Weis, A and Aronne, M and Alemu, WG and Williams, Z},
journal={Remote Sensing of Environment},
volume={284},
pages={113332},
year={2023},
publisher={Elsevier}
}
Software
@software{jordan_alexis_caraballo_vega_2021_7613207,
author = {Jordan Alexis Caraballo-Vega},
title = {vhr-cloudmask},
month = dec,
year = 2021,
publisher = {Zenodo},
version = {1.0.0},
doi = {10.5281/zenodo.7613207},
url = {https://doi.org/10.5281/zenodo.7613207}
}
Additional References
[1] Raschka, S., Patterson, J., & Nolet, C. (2020). Machine learning in python: Main developments and technology trends in data science, machine learning, and artificial intelligence. Information, 11(4), 193.
[2] Paszke, Adam; Gross, Sam; Chintala, Soumith; Chanan, Gregory; et all, PyTorch, (2016), GitHub repository, https://github.com/pytorch/pytorch. Accessed 13 February 2020.
[3] Caraballo-Vega, J., Carroll, M., Li, J., & Duffy, D. (2021, December). Towards Scalable & GPU Accelerated Earth Science Imagery Processing: An AI/ML Case Study. In AGU Fall Meeting 2021. AGU.
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file vhr_cloudmask-1.3.2.tar.gz
.
File metadata
- Download URL: vhr_cloudmask-1.3.2.tar.gz
- Upload date:
- Size: 29.5 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/5.1.1 CPython/3.12.7
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 3414a888ddd4e0c8f3cc5824e9bb59f58b4dde4c2b79f9f5eef1ac48017f521e |
|
MD5 | c6e9a4678d2246499d583363a88ba53f |
|
BLAKE2b-256 | 7b1ee498d19529a57bf408f16e9def5479c40f2a6a1b0f7ca237efb0c1a99986 |
File details
Details for the file vhr_cloudmask-1.3.2-py3-none-any.whl
.
File metadata
- Download URL: vhr_cloudmask-1.3.2-py3-none-any.whl
- Upload date:
- Size: 25.7 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/5.1.1 CPython/3.12.7
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | b7338e9c013d4fea7973d1f0ff0de06cc72da49c27f13752ab97fbeb8243daa4 |
|
MD5 | f45114712f7e650d24809580e061a660 |
|
BLAKE2b-256 | 57da6aca16be9dd853234780df565383a5830e293e0cc924e04a4cd385fe416a |