This folder contains a notebook tutorial for the Nasa Harvest Rwanda field boundary detection competition Nasa Harvest Field Boundary detection Challenge.
The dataset contains field boundaries for smallholder farms in eastern Rwanda. The Nasa Harvest program funded a team of annotators from TaQadam to label Planet imagery for the 2021 growing season for the purpose of conducting the Rwanda Field boundary detection Challenge. The dataset includes rasterized labeled field boundaries and time series satellite imagery from Planet's NICFI program. Planet's basemap imagery is provided for six months (March, April, August, October, November and December). The paired dataset is provided in 256x256 chips for a total of 70 tiles covering 1532 individual fields.
The starter notebook focuses on a baseline model using UNet to walk you through the process of loading and structuring the data, training the model, and exporting the predictions to the sample CSV file.
The notebook Nasa_Harvest_Model_MLHub.ipynb can be run locally using the requirements.txt file on Python 3.8.13 with minimum RAM : 8GB
There is a version of the same notebook on google colaboratory Google Colab
About Radiant Earth Foundation (radiant.earth)
Radiant Earth Foundation is a nonprofit technology company working to create a more sustainable ecosystem of machine learning (ML) and Earth observation (EO) data, standards, and tools. Guided by core characteristics of neutrality, collaboration, and innovation, Radiant maintains Radiant MLHub, an open-access geospatial training data library to ease the discovery of ML-ready data and models. Radiant also supports the development of community standards that enable interoperability of EO data and ML tools, and provides information and training to help advance the capacity of those working on global challenges using ML and EO. Visit Radiant Earth on Twitter and GitHub
About University of Maryland, Nasa Harvest (nasaharvest.org)
NASA Harvest is NASA’s Food Security and Agriculture program based at the University of Maryland. NASA Harvest’s mission is to enable and advance the adoption of satellite Earth observations by public and private organizations to benefit food security, agriculture, and human and environmental resiliency in the US and worldwide. They accomplish this through a multidisciplinary and multisectoral Consortium of leading scientists and agricultural stakeholders at the University of Maryland and implement it together with partners across the globe. Visit NASA Harvest on Twitter and Github.
About USDA(fas.usda.gov)
About USAID(usaid.gov)
This challenge and the training dataset collection and curation are based upon work by NASA Harvest supported by USDA-Foreign Agricultural Service and USAID, under Award Number FX22TA10960R004 and Project Title Earth Observations for Field Level Agricultural Resource Mapping (EO-Farm): Pilot in Rwanda in Support of NISR.