App Censipam

This project provides a Python environment that downloads SAR images (Sentinel-1 satellite) through zip files, pre-process it using SNAP, post-process it using SAR enhancing techniques and then feed it to a pre-trained CNN (U-Net implemented here) to detect newer deforestation in a georeferenced (CRS - WGS84) TIFF images.

This application performs the task of classifying the same region on different dates, usually separated by months, then performs the difference between the two and saves the output in a shapefile (.shp), using the same CRS as the original TIFF image, so it is compatible with QGIS, PostGIS, ArcGIS, etc.

  pred.shp = pred_after.shp - pred_before.shp

1 - Install Anaconda

Install conda as instructed in Anaconda.

Download the installer

    wget https://repo.anaconda.com/archive/Anaconda3-2022.10-Linux-x86_64.sh

Run the installer. You will be asked to provide a installation folder, for instance: $HOME/conda

    bash Anaconda3-2022.10-Linux-x86_64.sh

Or install conda from the mini version already downloaded here

    bash installation/Miniconda3-py39_4.11.0-Linux-x86_64.sh

In order to run conda commands in a terminal, save the conda installation path to $PATH environment variable. After that, you may need to close and reopen the terminal.

2 - Create the environment using conda

For convience all dependecies are provided in the file censipam.yml. To create the environment with all packages needed to run the application, run the following command:

    conda env create -f app/installation/censipam.yml

Once the environment is created, you may need to activate the environment:

    conda activate

Otherwise you will need to call the python by the full path like ~/conda/envs/<env_created>/bin/python

3 - Install ESA-SNAP

The application requires the installation of the ESA-SNAP for pre-processing the SAR images. The ESA-SNAP may be downloaded from ESA-SNAP or if the link is still valid:

** You might need the get the link to latest version at the ESA-SNAP website.

    wget https://download.esa.int/step/snap/8.0/installers/esa-snap_all_unix_8_0.sh

Run the installer

    bash esa-snap_all_unix_8_0.sh

Stick with default installation options

3 - Clone the application repository

Clone the application repository. You need only the /app folder

    git clone https://github.com/eliasqueirogavieira/app-censipam

3-1 - Additional package

You also must to install the package inside the folder packages/segmentation/. To do that, go the folder and execute the following command:

    conda activate <conda_env_created> # <conda_env_created> the conda env created previously
    pip install -e .

4 - Running the App - Sentinel data

Go the folder /appa and execute the application by runnning:

    ~/conda/envs/<env_created>/bin/python main_prediction.py \
                --data_path <folder>
                --root_output <folder>
                --config configs/cfg_sentinel.yml

• data_path: is the root folder with *.zip files of the scenes

• root_output: is the root folder where to save the processing results and the serialized object that keeps track of what have been processed. This folder will consume considerable disk space, choose thougly.

• config: this is a file with several settings, take a look at the file to see its content. Make sure the files indicated there exist and are properly set up. In config there are definitions for ESA-SNAP processing like the xml graph files and the model settings.

Depending on the number of .zip files encountered in data_path, the execution will take a while. It might be better to first pass a data_path with a few files to test. The application will: i) process each zip file with ESA-SNAP; ii) find out the image pairs; iii) process the pairs together with ESA-SNAP; and iv) execute the learned model to generate the predictions.

4-1 Structure of the root_ouput folder

The root_ouput folder will be have:

• partial_sentinel: keeps a folder for each zip file processed. Each folder's name is the name of the zip found

• app_results: keeps a folder for each image pair found and processed. Each folder's name is a combination of image pair.

  • stacked.tif: (VV1, VH1, VV2, VH2) stacked image
  • pred.shp: shapefile with prediction

• sentinel_obj: a serialized object to keep track of what have been processed