This repository contains source code used in

[1] López-Puigdollers, D., Mateo-García, G., Gómez-Chova, L. “Benchmarking Deep Learning models for Cloud Detection in Landsat-8 and Sentinel-2 images” Submitted pre-print

The following code creates a new conda virtual environment with required dependencies.

conda create -n dl_l8s2_uv -c conda-forge python=3.7 tensorflow=2 matplotlib --y

conda activate dl_l8s2_uv

python setup.py install

Expects an L1T Landsat-8 image from the EarthExplorer. The --landsatimage attribute points to the unzipped folder with a GeoTIF image for each band.

python inference.py CloudMaskL8 --l8image ./LC08_L1TP_002054_20160520_20170324_01_T1/ --namemodel rgbiswir

The folder ./LC08_L1TP_002054_20160520_20170324_01_T1 will contain a GeoTIF with the cloud mask.

Expects an L1C Sentinel-2 image from the OpenHub. The --s2image attribute points to the unzipped SAFE folder. The --resolution attribute select the output resolution of the product (10, 20, 30 or 60)

python inference.py CloudMaskS2 --s2image ./S2A_MSIL1C_20160417T110652_N0201_R137_T29RPQ_20160417T111159.SAFE/ --namemodel rgbiswir --resolution 30

The folder ./S2A_MSIL1C_20160417T110652_N0201_R137_T29RPQ_20160417T111159.SAFE will contain a GeoTIF with the cloud mask.

If you use this work please cite:

 @article{lopez-puigdollers_benchmarking_2021,
	title = {Benchmarking Deep Learning models for Cloud Detection in {Landsat-8} and {Sentinel-2} images},
	volume = {},
	issn = {},
	doi = {},
	journal = {Submitted},
	author = {López-Puigdollers, Dan and Mateo-García, Gonzalo and Gómez-Chova, Luis},
	month = {},
	year = {2021},
	pages = {},
}

• Multitemporal cloud masking in the Google Earth Engine
• Landsat-8 to Proba-V transfer learning and Domain adaptation for cloud detection