Ag-Net is a bundle of customized deep neural networks for recognizing crops based on spectral characteristics. The main idea is to customize the current neural networks like U-Net to make it more suitable for extracting and recognizing the season-changing features of various crop over the growing season. The ultimate goal is to generate high-quality accurate crop maps at any stage of crops rather than waiting for next year. The training and validation ground truth data comes from USGS EarthExplorer and USDA NASS.
We have created a free sample dataset for direct training and testing of neural network. Ag-Net-Dataset
Ayush Patel, Jaypee University of Information Technology : Agriculture Net.ipynb
SANKALP MITTAL, BITS Pilani : Ag-Net.ipynb Training Accuracy: 91% Validation Accuracy: 84%
AgNet_NE_v1 dataset: Agricultural Map Archive of Nebraska, 1985 ~ 2000, Download here
We have several talented students voluntarily working on this project as their summer projects.
If you would like to contribute to this project or list existing jupyterbooks or github repos in this page, please feel free to contact Dr. Ziheng Sun (zsun@gmu.edu) or Dr. Annie Burgess (annieburgess@esipfed.org).