README.md

Planet Pieces

Convolutional Neural Network and image segmentation for land classification of high resolution multispectral Planet imagery

Collaborators

Nga (Nancy) Nguyen

Claire Miles

Joshua Driscol

Matt Olson

Mike Leech

Shashank Bhushan

Michelle Hu

Team Lead:

Michelle Hu

Data Science Leads:

Claire Miles, Joshua Driscol - Tensorflow, Keras
Shashank Bhushan - Planet data and rasters
Nga Nguyen, Matt Olson - thresholding and visualization

Introduction

Background reading:

• Ensemble CNN for land classification
• RGB-NIR Image Classification
• Original U-Net paper
• U-Net architecture borrowed heavily from this repository.
• Tensorflow 2.0 Documentation

Files

• .gitignore
  Globally ignored files by git for the project.
• environment.yml
conda environment description needed to run this project.
  To preserve the conda environment across sessions, please add this line of code into your ~/.condarc file, or create that file if it does not exist:

envs_dirs:
 - /home/user/conda-envs/

Then run:

conda env create -f environment.yml -n planetpieces

And to activate the environment:

conda activate planetpieces

• README.md
  Description of the project and personnel.

We will be using 3m resolution data from Planet Labs at Mount Rainier in Washington state. The purpose of this repository is to implement novel techniques for analyzing geospatial data sets using neural networks:

• snow vs. land cover segmentation
• techniques for stacking coarse resolution images to train them for upsampling like in DeepSUM

Sample

Folders

contributors

Each team member has their own folder under contributors, where they can work on their contributions. Having a dedicated folder for one-self helps to prevent conflicts when merging with master.

notebooks

Notebooks that are considered delivered results for the project should go in here.

scripts

Helper utilities that are shared with the team

Team Wiki

Further information can be found on our team wiki page