SENTINEL2 FOREST FIRES DETECTION

Demo using a deep learning model to detect forest fires using sentinel-2 multi-spectral satellite imagery

Table of Contents

1. About this repository
2. Getting started
3. Usage
   • Demo running
   • Data structure
4. License
5. Contact

1. About this repository 💼

This git repository contains a demonstration of active forest fire detection using deep learning. The demo consists of showing how to load the model and make the inference on some sample images.

This demo is built with

• Anaconda
• Python
• Tensorflow
• Skimage
• Matplotlib
• NumPy

(back to top)

2. Getting Started 📚

Follow these instructions to locally setup up your project.

• Clone the repo

  git clone "https://github.com/hxfdanger/S2WDSV2.git"
  
  cd S2WDS/

• Create the dedicated python environment using the S2WDS_env.yml file

  conda env create -f S2WDSV2_env.yml

🚨 If you have problems while installing the environment, please create it manually, then you can manually install the packages mentioned here.

• Activate it:

     conda activate S2WDSV2

🚨 To run conda command you have to install Anaconda distribution).

(back to top)

3. Usage 📒

Demo running

The jupyter nootebook is not installed in the S2WDS env created by S2WDS_env.yml file.

🚨 To run the notebook demo.ipynb you have to install jupyter package in your active (S2WDS) environment

conda install -c anaconda jupyter

🏁 Run the notebook demo.ipynb

jupyter notebook demo.ipynb

✍ NOTES:

• Demo.ipynb will use some function declared in utils.py.
• All images used for the demo are located in data folder. The structure of the data described (bellow).
• The model used in the demo does not exist in this repository. you can download all models from link. Do not forget to put it in a folder called models.

Data structure

data/
├── false_color
│         ├── Sentinel-2_L1C_from_2018-08-23_Mendocino_13.tif
│         ├── Sentinel-2_L1C_from_2018-08-23_Mendocino_27.tif
│         ├── Sentinel-2_L1C_from_2018-11-16_Chico_5.tif
│         ├── Sentinel-2_L1C_from_2019-03-04_Traralgon_22.tif
│         ├── Sentinel-2_L1C_from_2019-03-04_Traralgon_71.tif
│         └── Sentinel-2_L1C_from_2019-03-04_Traralgon_76.tif
├── image_distribution
│         └── test.txt
├── masks
│         ├── Sentinel-2_L1C_from_2018-08-23_Mendocino_13.tif
│         ├── Sentinel-2_L1C_from_2018-08-23_Mendocino_27.tif
│         ├── Sentinel-2_L1C_from_2018-11-16_Chico_5.tif
│         ├── Sentinel-2_L1C_from_2019-03-04_Traralgon_22.tif
│         ├── Sentinel-2_L1C_from_2019-03-04_Traralgon_71.tif
│         └── Sentinel-2_L1C_from_2019-03-04_Traralgon_76.tif
└── RGB
    ├── Sentinel-2_L1C_from_2018-08-23_Mendocino_13.tif
    ├── Sentinel-2_L1C_from_2018-08-23_Mendocino_27.tif
    ├── Sentinel-2_L1C_from_2018-11-16_Chico_5.tif
    ├── Sentinel-2_L1C_from_2019-03-04_Traralgon_22.tif
    ├── Sentinel-2_L1C_from_2019-03-04_Traralgon_71.tif
    └── Sentinel-2_L1C_from_2019-03-04_Traralgon_76.tif

✍ NOTES:

• The data folder only contains examples used to run the demo.
• false_color folder contains 6 images composed by [B12,B11,B04] (sentinel2 bands) used as the input of the model for inference.
• RGB contains 3 images corrsponding to false_color images composed by [B04,B03,B02] (sentinel2 bands) used only for visualizing.
• masks folder contains 6 binary masks (value : 0=no-fire; value : 1=fire) corresponding to the images in the false_color folder.
• Image_distribution folder contains 1 files :
  • test.txt : contains the name of 6 images corrsponding to (false_color, RGB and masks) used for the inference.

🚨 To download the full data (link)

👉 For more information about the data please read this (Readme)

(back to top)

4. License 📑

Distributed under the Attribution 4.0 International (CC BY 4.0) License.

Contains modified Copernicus Sentinel data [2016-2020] for Sentinel data

(back to top)

5. Contact 📭

• Houssem FARHAT 📧
• Benjamin FRANCESCONI 📧
• Michael BENGUIGUI 📧
• Adrien GIRARD 📧
• Lionel DANIEL 📧

Project Link: https://github.com/hxfdanger/S2WDSV2

(back to top)