AESS-challenge-ENSI-SB : Oil Spill Detection and Monitoring

Overview :

This project leverages satellite SAR images from the Copernicus dataset and a SegNet deep learning model to detect oil spills, calculate their size and surface area. It also includes a web platform as a prototype for visualizing and alerting response teams about oil spill incidents.

Video Demo Link

Table of content :

1. Features
2. Project structure
3. Project setup
4. User Manual
5. Model Details
6. Web Platform

Features :

This project aims to address the environmental challenge of oil spills by combining AI and a web platform:

• AI Model:
  • Detect oil spills from Copernicus SAR images using a deep learning SegNet model by generating Ground Truth Masks (GTM)
  • Calculate the size and surface area of detected spills.
• Web Platform:
  • Display detected oil spills on an interactive map.
  • Send real-time alerts to response teams.
  • Maintain a history of past oil spills for analysis.

Project Structure :

The project consists of two main parts:

1. AI Model

• Dataset: European Space Agency (ESA) via the Copernicus Open Access Hub SAR image data.
• Model: SegNet-based deep learning architecture for semantic segmentation.
• Output: Ground truth masks for oil spills, spill size, and surface area calculations.

View on Kaggle

2. Web Platform

• Frontend: Built with React.js for user interaction.

Getting Started

Installation Steps

1. Clone the repository:

   git clone https://github.com/AhmedTrb/AESS-challenge-ENSI-SB.git
   cd AESS-challenge-ENSI-SB

2. Set up the environment for the AI model:

   pip install -r requirements.txt

3. Set up the web platform:

   cd web-platform
   npm install

Usage :

1. Running the AI model :

• Importing necessary libraries.
• Loading and processing data.
• Splitting the dataset.
• Loading pretrained model weights.
• Testing the model and displaying predictions.

2. Running the web platform :

cd web-platform
npm run dev

Open the platform in your browser at http://localhost:3000

Model Details :

overview :

In this project, a Conditional Generative Adversarial Network (CGAN) is used for oil spill detection by generating segmentation masks from SAR images. The CGAN consists of two main components: a generator and a discriminator. The generator is a U-Net-like architecture that predicts oil spill masks from input images, while the discriminator is a PatchGAN model that evaluates whether the predicted mask is realistic by comparing it with the ground truth.

Dataset :

The dataset is provided from this paper [1] and consists of preprocessed raw images and Ground Truth Masks (GTM). It was collected from the Sentinel-1 mission satellite, which provides level-1 Ground Range Detected (GRD) data, equipped with a C-band Synthetic Aperture Radar (SAR) system.

• Raw Images: Contains the original SAR images.
• Ground Truth Masks (GTM): Corresponding binary masks indicating the oil spill regions.

1. Seg-Net Generator :

The generator is a Convolutional Neural Network (CNN) that consists of an encoder-decoder structure to process the input images and generate segmentation masks.

• Encoder: Uses convolutional layers with down-sampling to extract features.
• Decoder: Uses transposed convolutional layers to reconstruct the segmentation masks.

2. Patch-GAN Discriminator :

The discriminator evaluates whether a given pair of input image and generated mask is real or fake.

Training :

During training, the generator learns to create accurate masks by minimizing a combination of adversarial loss (to fool the discriminator) and L1 loss (to ensure similarity to ground truth). This collaborative adversarial setup allows the CGAN to generate high-quality segmentation masks, even for complex SAR image patterns.

Web Platform :

Landing Page:

Main :

References :

[1] Deep neural network for oil spill detection using Sentinel-1 data: application to Egyptian coastal regions by Samira Ahmed,Tamer ElGharbawi,Mahmoud Salah & Mahmoud El-Mewafi link

[2] Mapping oil pollution in the Gulf of Suez in 2017–2021 using Synthetic Aperture Radar link

[3] Mediterranean Sea region briefing - The European environment — state and outlook 2015 link

[4] Global maritime traffic link

[5] More than 750 oil slicks spotlight pollution risks in Mediterranean Sea link

[6] Mediterranean Sea Chronic Oil Pollution Analysis: July 2020-January 2024 link