Being able to identify crop problems early can make the difference between saving a crop and losing it, but high-tech solutions can be costly. An interdisciplinary team of researchers thinks a new approach leveraging existing technology may be part of the solution.
Specifically, NC State researchers in the Department of Crop and Soil Sciences and the Department of Electrical and Computer Engineering are launching an inexpensive camera system that can monitor crop stress remotely.
Corn and soybeans are important commodities for North Carolina and the world. Both are eaten fresh, processed into a variety of foodstuffs and turned into animal feed. A lack of water at certain stages stress the plants and can make a significant dent on yields.
Paula Ramos-Giraldo, a computer vision and machine learning expert in the Department of Crop and Soil Sciences, has spent the past year working on a camera system that costs less than an average smartwatch to track drought stress in corn and soybean fields.
“Our goal, specifically, was to construct a low-cost sensor to track the soil moisture level in the field through plant behavior,” Ramos-Giraldo said.
These low-cost sensors can help researchers studying ways to make agricultural systems more resilient; plant breeders breeding more drought-tolerant varieties; and someday may be able to alert farmers when their fields need to be irrigated.
The StressCam system — constructed from parts that cost about $150 — is based around a Raspberry Pi and leverages its attached camera to capture photos every 15 minutes to watch for wilting. The system is solar-powered, with a back-up battery for cloudy days.
The tiny computer runs a machine learning algorithm on the photos to analyze them for indications of drought stress. Then it sends this information to a web platform for researchers, breeders or farmers, she said.
Both the machine learning algorithm and the web platform were constructed with help from students in the ECE.
During the fall 2019 semester, Ramos-Giraldo worked with Edgar Lobaton, an associate professor in the Department of Electrical and Computer Engineering, to enlist the students in his Neural Networks class to design machine learning algorithms capable of looking at photos of soybean fields and score the severity of drought stress.
Machine learning algorithms can find patterns in data without being explicitly programmed what important features to look for. Instead, they are “trained” on pre-defined data – in this case 5,000 photos of soybean fields showing different amounts of drought stress annotated by Anna Locke, a U.S. Department of Agriculture- Agricultural Research Service (USDA-ARS) soybean expert in the Department of Crop and Soil Sciences.
The class’s algorithms were then tested on thousands of other photos taken by the StressCam during the summer of 2019. One of the best algorithms was programmed into the StressCam.
Also during the fall of 2019, Ramos-Giraldo started working with a team of senior Electrical and Computer Engineering students to design a cloud-based web platform to allow farmers, soybean breeders and researchers to manage their StressCams and monitor their fields.
This team included Artem Minin, Nathan Libner, Stephanie Sierra and Manish Goud.
“It must have taken us the first three weeks to get our heads wrapped around how we could build a platform to solve the problem that Paula Ramos-Giraldo had,” said Minin, who will be returning to NC State in the fall to pursue a master’s degree in computer engineering. “That was one of the really difficult parts of the project because none of us had ever really built such a diverse system with so many different components, technologies, and requirements.”
The web platform allows the users to check that the StressCam is on and not overheating, change the photo scheduling and importantly, look at the past images from the StressCam and the drought stress severity scores. Overall, the platform will save time and increase the precision of drought data collection for plant breeders and researchers studying resiliency.
“Before the pandemic, our group would meet in person once a week, and that’s when our best results got created,” Minin said. “COVID-19 also added a difficulty that we couldn’t test our system like we wanted to. Originally, we planned to have a test site set up at the Sandhill Research Station, but obviously that didn’t happen. Instead, Paula set up a test field in her backyard.”
Despite all of the COVID-19-related challenges, Minin found working on an agricultural problem very rewarding.
“It is an amazing effort that we made with the ECE department, and will continue,” Ramos-Giraldo said. “The most important thing is that the students were so enthusiastic about their results. And on the way we learned a lot of things, not only the students, but us as well. It is amazing the results we can produce when we work together.”
The StressCam platform uses IBM’s Internet of Things cloud platform at its core. Additionally, IBM sponsored the senior design project and provided technical mentorship from two researchers based at the IBM Innovation Center, on NC State’s Centennial Campus.