Binary classifier that predicts presence of methane emissions in an image. Model integrated in a Streamlit Web App for seamless workflows.
by: Clara Besnard, Ian Moon, Marina Pellet, Łukasz Pszenny, Adel Remadi, Lasse Schmidt
within: MS Data Sciences & Business Analytics
at: CentraleSupélec & ESSEC Business School
This repository contains our work during the 3-day hackathon by QuantumBlack at ESSEC Business School. It includes all the code required for our binary image classification model to identify methane emissions as well as the Streamlit web app to get model predictions as a non-technical user.
Within the hackathon, we work together with CleanR, a fast-growing start-up specialized in Methane emissions reporting. CleanR founders team met for the first time in 2020 in Geneva. Today, they count already 50 team members. Their mission is to help diminish Methane emissions by providing a clear method for MRV: monitoring, reporting and verification.
Using the gathered satellite images [data set – 64 x 64 images in greyscale] of different locations, we want to identify whether each location contains a methane plume or not. On top, we want to detect use cases where such a model can be used to drive positive impact.
- Predicted labels for test set (csv file)
- Model code (github link sufficient)
- Web app code (with two features -- [1] uploading satellite image, [2] model prediction that indicates presence of methane plume)
- Pitch presentation ([1] approach, [2] web app demo, [3] use cases) --> https://docs.google.com/presentation/d/1Y1krles5VHxe1MJsJfR4c9I9gJADnv-lHLUK-CELl7I/edit?usp=sharing