FSDL Deforestation Detection

Detecting deforestation from satellite images: a full stack deep learning project

Description

A deep learning approach to detecting deforestation risk, using satellite images and a deep learning model. We relied on Planet imagery from two Kaggle datasets (one from the Amazon rainforest and another on oil palm plantations in Borneo) and trained a ResNet model using FastAI. For more details, check the following links:

• Streamlit dashboard for testing the model and exploring the data

• Model training notebook in Colab

• Project management workspace in Notion

• Loom video about the project

• Medium article about this project

This is the result of a group project, made by André Ferreira and Karthik Bhaskar, for the Full Stack Deep Learning - Spring 2021 online course.

Very first steps

Initial

1. Initialize git inside your repo:

git init

2. If you don't have Poetry installed run:

make download-poetry

3. Initialize poetry and install pre-commit hooks:

make install

4. Upload initial code to GitHub (ensure you've run make install to use pre-commit):

git add .
git commit -m ":tada: Initial commit"
git branch -M main
git remote add origin https://github.com/karthikraja95/fsdl_deforestation_detection.git
git push -u origin main

Initial setting up

• Set up Dependabot to ensure you have the latest dependencies.
• Set up Stale bot for automatic issue closing.

Poetry

All manipulations with dependencies are executed through Poetry. If you're new to it, look through the documentation.

Poetry's commands are very intuitive and easy to learn, like:

• poetry add numpy
• poetry run pytest
• poetry build
• etc

Makefile usage

Makefile contains many functions for fast assembling and convenient work.

make download-poetry

make install

If you do not want to install pre-commit hooks, run the command with the NO_PRE_COMMIT flag:

make install NO_PRE_COMMIT=1

make check-safety

This command launches a Poetry and Pip integrity check as well as identifies security issues with Safety and Bandit. By default, the build will not crash if any of the items fail. But you can set STRICT=1 for the entire build, or you can configure strictness for each item separately.

make check-safety STRICT=1

or only for safety:

make check-safety SAFETY_STRICT=1

multiple

make check-safety PIP_STRICT=1 SAFETY_STRICT=1

List of flags for check-safety (can be set to 1 or 0): STRICT, POETRY_STRICT, PIP_STRICT, SAFETY_STRICT, BANDIT_STRICT.

The command is similar to check-safety but to check the code style, obviously. It uses Black, Darglint, Isort, and Mypy inside.

make check-style

It may also contain the STRICT flag.

make check-style STRICT=1

List of flags for check-style (can be set to 1 or 0): STRICT, BLACK_STRICT, DARGLINT_STRICT, ISORT_STRICT, MYPY_STRICT.

Codestyle uses pre-commit hooks, so ensure you've run make install before.

make codestyle

make test

make lint

the same as:

make test && make check-safety && make check-style

List of flags for lint (can be set to 1 or 0): STRICT, POETRY_STRICT, PIP_STRICT, SAFETY_STRICT, BANDIT_STRICT, BLACK_STRICT, DARGLINT_STRICT, ISORT_STRICT, MYPY_STRICT.

make docker

which is equivalent to:

make docker VERSION=latest

More information here.

make clean_docker

or to remove all build

make clean

More information here.

🛡 License

This project is licensed under the terms of the MIT license. See LICENSE for more details.

📃 Citation

@misc{fsdl_deforestation_detection,
  author = {Karthik Bhaskar, Andre Ferreira},
  title = {Predicting deforestation from Satellite Images},
  year = {2021},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/karthikraja95/fsdl_deforestation_detection}}
}