Check our website for documentation and more info - https://sgsaram.github.io/fguard/.
Our goal is to look for deforestation area and track them in time period on satellite images. For example:
We tried to use unsupervised and supervised learning on this problem and got two best algorithms:
- K-Means
- U-Net model
Of course we are taking information from date of photos (It takes a long time for a forest to grow back). The main problem are clouds and "even human can't be 100% sure if this pixel is forest or deforestation".
We download data from Sentinel Hub (https://www.sentinel-hub.com). We work only with L1C and L2A satellites. You need API token for this.
- Just install the package:
pip install fguard- Clone the repository (it will download to
fguardfolder):
git clone https://github.com/Sgsaram/fguard.git fguard- Go to the folder:
cd fguardUse venv!
- Install requirements using this command:
pip install -r requirements/prod.txt- Wake the
setup.pyup:
pip install .This will give you the opportunity to use fguard cli tool.
- Add credentials to a programm:
fguard config [–f “path/to/config/settings.toml”] | [--id “YOUR_ID” --token “YOUR_TOKEN”]- Request:
fguard request [–f “path/to/config/settings.toml”] | [-c c1 c2 c3 c4 -t t1 t2] “output/folder” {-s "size" –d "detector" --isolate}This will save images in output folder. Also saves json files of all events (new, add, merge).
- Erase cache, config file, etc:
fguard delete [--cache] | [--config] | [--model]- To use
netdetector you need to download model:
fguard net --updateIt can be also used to update model (we still train it).
For more info check our website - https://sgsaram.github.io/fguard/.
config-settings-example.toml
[config]
id = ""
token = ""request-settings-example.toml
coordinates = [ 42.65614, 59.58242, 42.73579, 5952529 ]
time = [ "2017.12.01", "2023.02.28" ]
