Menu Close
 

Deriving Tillage Practices and Irrigation Area/Methods from National Agriculture Imagery Database using Deep Learning and Supercomputing

Project Team

  • Adam Stewart, Ph.D. Student, Dept. of Computer Science
  • Dr. Jian Peng, Assistant Professor, Dept. of Computer Science
  • Dr. Kaiyu Guan, Assistant Professor, Dept. of Natural Resources and Environmental Sciences

Abstract

Accurate information on the areal distribution of irrigation and tillage practices is critical for precision and sustainable agriculture. Such information provides the basis for government policy-making and industry decision-making regarding conservation practices and water resource management. However, granular information on tillage practices and irrigation area at high resolution is currently missing. Previous work using 30-meter resolution Landsat imagery and random forest classifiers has provided a good baseline, but there is a huge room for improvement in terms of both spatial resolution and accuracy. We propose the use of 1-meter resolution imagery from the National Agriculture Imagery Program (NAIP) and convolutional neural networks to tackle this challenging task, and aim to develop regional maps of annual tillage practices and annual irrigation infrastructure at 1-meter resolution for the 12 major states of the U.S. Corn Belt. This proposed work is deeply rooted in many related research tasks we tackle, and thus we are confident in our ability to achieve it. We will create prototypes using NCSA’s supercomputers and also test for scalability on commercial cloud systems (e.g. AWS). This project is expected to have the potential to influence the broader community of remote sensing researchers, swaying the relevant field towards deep learning. With expertise in crop science, remote sensing, and computer science, our interdisciplinary research group is uniquely positioned to push the boundaries of what is possible in remote sensing for agriculture.