j-desloires / sentinel2-process Goto Github PK

The purpose of this module is to download and preprocess Sentinel-2 satellite imagery data directly from Theia-land from given dates and a tile name or bounding box.

• Configure an account on Theia https://sso.theia-land.fr/theia/register/register.xhtml to configure credentials and pull https://github.com/olivierhagolle/theia_download. The folder theia_download available on this git repository is an example of what you should get.
• Download OTB https://www.orfeo-toolbox.org/CookBook/Installation.html for data preprocessing.

Using anaconda, create the environment using :

cd sentinel2-process
conda env create -f environment.yml
conda activate sentinel2-process

Code to download and preprocess a tile for Land Use and Land Cover analysis is summarized into the script pipeline_land_cover.py.

You must provide the following paths to execute a code from the following files :

• A vector folder where you have :

  • Ground truth shapefile : Objects ID and labels to use for land cover training. The projection should be in your UTM (e.g 'espg:32631' for the tile TC31J (France)).
  • Polygon of the Area of Interest : shapefile that we be used to cropping the images with respect to your study zone.

• Folder where Orfeo Toolbox is installed

• Path to save the output GEOTIFFS images

The tile is cropped with respect to the extent of a given shapefile (script main.py, with Haute-Garonne as example in the variable mask_data) :

• mask_R10_crop.tif : Binary cloud masks at 10 meters (int16)
• mask_R20_crop.tif : Binary cloud masks at 20 meters (int16)
• GFstack_X_crop.tif : Gap filled images with a given band X at 10 meters using linear interpolation (OTB) (float32)
• GFstack_SI_X_crop.tif : Gap filled images with a given band X at 20 meters, superimposed at 10 meters, using linear interpolation (OTB) (float32)
• Class_ID_crop.tif : ID of the class from the training dataset to predict for supervised task (script main.py) (int16)
• Object_ID_crop.tif : ID of the objects from the training dataset (int16)
• dates.csv : Csv file with acquisition dates that corresponds to the time index of each .tif image (except Class_ID and Objects_ID)