SYH-pixel/cropfield-polygonization

Work related to crop field processing completed for ClarkU Spring26 Intro to Python Directed Study and Mapping Africa projects # cropfield-polygonization Work related to crop field processing completed for ClarkU Fall26 Intro to Python Directed Study's Final Project. Completed by Solana Huang, Abrianna Culligan, and Zachary Yildiz-Raslan. # Notes on setup Zambia Cropfield Shape statistics file is ~1.9 G - please download locally only at this link: drive.google.com # Background This work is based on Mapping Africa work completed by Prof. Lyndon Estes's AgroImpacts Research crop at Clark University. As part of their research, they have used CNNs to generate cropfield polygons across the entirety of states such as Zambia using high resolution Planet Imagery. This project is performed in parallel with other projects also attempting to create country wide large area field boundary maps, such as Fields of the World. Below is an example of the code used by Fields of the World to generate vector files from raster imagery: github.com One challenge of this work is the creation of false "thin-necked" polygons and false "islands" within cropfield polygons. See example image below with false islands circled in red and thin necks connecting separate fields outlined in blue. For our final project, we propose building upon this research by finding an algorithmic way to clean up these cropfield polygons. # Research Question For our final project, we propose building upon this research by finding an algorithmic way to clean up these cropfield polygons. We hope to identify and fix (in order of priority): Thin necks connecting crop fields and false cutouts within crop fields. This entails splitting the polygons at the narrowest part of the neck and deleting all false cutouts. # Data Sources Our primary data source for this project will be shape statistics generated for the state of Zambia using 2023 Planet Imagery generated using `src/instancemaker/computeinstances …