whigg/nz_space_challenge

A single end-to-end deep learning solution to identify and traverse crevasses in Antarctica for safer, more efficient navigation

New Zealand Space Challenge 2018

Detecting crevasses in Antarctica for safer, more efficient navigation as an analogue for future space missions.

Experimental (alpha) leaflet map demo using tensorflowjs here.

Youtube video giving a quick overview explanation here.

CrevasseNet model architecture

Consists of a classifier module seamlessly joined to a navigator module, trained using supervised learning and reinforcement learning respectively.

Note that the classifier component is actually much deeper, but has been abbreviated in the above diagram for simplicity.

Sample predictions

Crevasse Classifier

Input image (satellite/aerial)--> Intermediate Output (crevasse map)

Route Navigator

Intermediate output (crevasse map) --> Action quality outputs

Getting started

Quickstart

Launch Binder, data will be loaded via Quilt. Cheers to data2binder!

Installation

Start by cloning this repo-url

git clone <repo-url>
cd nz_space_challenge
conda env create -f environment.yml

Running the jupyter notebook

source activate nz_space_challenge
python -m ipykernel install --user  #to install conda env properly
jupyter kernelspec list --json      #see if kernel is installed
jupyter notebook

Data used

Name	Data Source
MOA-derived Structural Feature Map of the Ronne Ice Shelf, Version 1	NSIDC-0497
MODIS Mosaic of Antarctica 2003-2004 (MOA2004) Image Map, Version 1	NSIDC-0280