README.md

metadata

title: Oil Spill Detection
emoji: 🌊
colorFrom: blue
colorTo: indigo
sdk: docker
pinned: false

Oil Spill Detection & Monitoring System

This is a full-stack AI-driven application that detects oil spills from satellite imagery using a deep learning segmentation model (U-Net).

System Architecture

• AI/ML: TensorFlow / Keras (U-Net)
• Backend: FastAPI (REST API with Python)
• Frontend: React + Vite + TypeScript (Modern UI)
• Containerization: Docker & Docker Compose

Repository Directory Structure

• backend/ - FastAPI predict/health endpoints and Inference Engine
• frontend/ - React Vite App for the UI
• model/ - U-Net architecture, Custom losses, and Training logic
• utils/ - Data loading and augmentation pipelines utilizing tf.data.Dataset
• data/ - Holds your images and masks
• notebooks/ - Place to run exploratory code

Getting Data (Google Drive)

Download the dataset using the following Google Drive link provided: https://drive.google.com/drive/folders/1j35RM-5uZbTOfDv0mBTeRRmOUPvvg28t?usp=sharing

1. Download the images and masks folders inside it.
2. Place them inside the ./data/ folder in the root directory like so:

   d:/Oill_SPLIT/data/images/
   d:/Oill_SPLIT/data/masks/
   

Local Setup (Without Docker)

Backend

1. cd backend
2. pip install -r requirements.txt
3. uvicorn main:app --reload The API will be available at http://localhost:8000

Frontend

1. cd frontend
2. npm install
3. npm run dev The React UI will be available at http://localhost:5173

Setup (With Docker)

To run the entire suite using Docker:

docker-compose up --build

Model Training Instructions (Google Colab / Kaggle)

You can train the model on Google Colab leveraging the free T4 GPU.

1. Zip the model/, utils/, and data/ directories and upload them to Google Drive or directly to Colab/Kaggle.
2. Open a Colab Notebook.
3. Install dependencies: !pip install tensorflow scikit-learn pillow numpy
4. Run the training script:

   !python model/train.py --data_dir ./data --save_dir ./model/saved_models --epochs 50 --batch_size 16
   

5. Once trained, download oil_spill_unet_best.keras from ./model/saved_models and place it in your local folder structure.

(Note: During initial development, if oil_spill_unet_best.keras isn't found, the backend will auto-initialize an untrained "stub" model so you can immediately test the UI and API offline).