Skip to content

A Hackathon in partnership Mckinsey & Company: Quantum Black to detect methane leaks and develop a business case for a startup Clean R

Notifications You must be signed in to change notification settings

annabelleluo/methane-leak-detection-in-satellite-imagery

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

59 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Methane Leak Detection in Satellite Imagery Using ResNet

Authors

The project is done by Leonardo Bassili, Antoine Cloute, Karim El Hage,Yasmina Hobeika, Annabelle Luo, Ali Najem, Amine Zaamoun for a Hackathon in partnership with Mckinsey & Company Quantum: Black. The team has awarded Second Place for it's work by the jury.

Objective

The objective of the project is to develop an application capable of detecting methane leaks in provided satellite imagery to assist with MRV: monitoring, reporting, and verification

Current Build

The current build of the application is only capable of image classification (methane leak or no methane leak). There is also a GradCAM functionality capable to give a form of interpretability to the results.

Project Contents

1 - Models: Contains the training methdology of the ResNet model used for classification in the application. This includes train-test split methdology, image augmentation training feature and actual model training. The notebook within the folder contains the training of several different models. However, the model with the best AUC was selected also considering the most reasonable train test split that avoids data leakage.

2 - streamlit_web_app: Contains the Initial build of the application using StreamLit and having the best found model.

3 - Reports: Contains the final deck presented to the jury.

About

A Hackathon in partnership Mckinsey & Company: Quantum Black to detect methane leaks and develop a business case for a startup Clean R

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 99.2%
  • Python 0.8%