This project, named sky, aims to detect and localize in satellite photos four specific types of cloud formations: Flower, Fish, Gravel and Sugar, so named because of their characteristic pattern.
The sky project is based on a 2019 Kaggle competition, initiated by the Max Planck Institute.
Main purpose was to mobilize the data scientist community on the realization of models capable of
identifying shallow cloud locations, considered as potential key factors in climate’s understanding.
In this frame, we developed a convolutional neural network model trained to identify and localize the
above four cloud formations.
More details on methodology and results are provided inside the article: Silver-lining clouds with AI published on Towards Data Science.
The repository structure reflects project sequence. The sky project was organized in 3 phases:
a first exploratory analysis, the second phase aimed at implementing a multi-classification model and the third one
aimed at implementing a segmentation model. Libraries have also been developed and are grouped under the "packages" folder (associated documentation
is available under "docPackages" folder).
The 3 related jupyter notebook are visible from the «colab» links below.
-To-be-done
-To-be-done
The project sky was carried out as student project for Data Scientist training course at the DataScientest institute (datascientest.com)
Project members:
Hans Schauer 
Georg Wolf
Project mentor:
Victor (DataScientest)