Engineered ‘Brown Thumb’

In a world where interdisciplinary research is increasingly necessary to solve complex challenges, the ability to step outside one’s field has become essential. Nowhere is this more evident than in efforts to ensure global food security, where plant science, engineering and artificial intelligence must converge to help growers combat crop diseases and improve yields.

Traditional disease identification methods rely on manual surveying, which is time-consuming, labor-intensive and prone to human error. Meanwhile, global crop losses due to biotic stressors amount to an estimated $60 billion annually, making timely mitigation efforts crucial to reducing yield loss. As agricultural challenges grow in complexity, there is a continued need for data-driven solutions to provide faster, more accurate disease recognition.

My background blends computer science, math and data, which eventually led me to an Electrical Engineering Ph.D. focused on developing AI (artificial intelligence, not artificial insemination—sorry, animal scientists) and computer vision models for agricultural and environmental applications. I have had the opportunity to work with various tools and sensors (like satellites, hyperspectral cameras and stereoscopic systems) to analyze plant health. When I started my Ph.D., I never imagined I would find myself running a pathogenic greenhouse experiment, but research has a way of leading us down unexpected paths.

As a Ph.D. student and FFAR Fellow at NC State University, my research focuses on developing efficient deep-learning models for crop disease recognition and assessment so that growers can employ effective mitigation efforts, such as using field robots to apply fungicides or remove diseased plants. Typically, growers and researchers use different methods to recognize the type of disease and the severity of that disease. One of our projects explores the possibility of combining the methods into one efficient model for tomato foliar diseases. This multi-task approach will simultaneously classify the type of disease and estimate its severity. It’s kind of like a 2-in-1 shampoo and conditioner: when you have the time, space and budget, separate products (or models) are often preferred. But when you’re out in the field with an unmanned ground vehicle (UGV) dealing with limited compute and connectivity, the 2-in-1 becomes practical and efficient.

Of course, before deploying an AI system to the field robots, we need to ensure it is adequately trained. To develop and refine our algorithms, we need a controlled environment to capture high-quality data, free from unpredictable variables like weather. And so, my journey into the world of plant science began, and I had to develop a “brown thumb”; learning how to grow and deliberately kill tomatoes.

To my surprise, tomato plants don’t thrive on Python scripts and caffeine. I had to actually learn how to properly grow them, give myself a crash course in fungal and bacterial life cycles and adopt strict contamination protocols.

I worked closely with plant pathologists and horticulture experts at NC State to select disease-susceptible tomato varieties and coordinate pathogen culture development. While they worked on growing the pathogens, I became a regular at the NC State Phytotron. Now, a Phytotron is not the newest member of the Transformers Autobots, but rather a high-tech controlled-environment facility with greenhouses and growth chambers to study plants under specific conditions. Working with the Phytotron technicians, we strategized on both growth conditions and containment tactics. We eventually landed on the “dome method”, aka putting plastic tents over our plant and hoping for controlled chaos.

Once all our variables were defined, we got to work planting dozens of seeds and multiple replicates. Like many trials, there were hiccups; one batch didn’t make it and met their end at the bottom of a trash can.

Still, in a few short weeks, it was time to harvest pathogen spores and create inoculum. This is the disease’s starter pack; just enough of the pathogen to infect the plant and get things spreading. Shockingly, this is a task they don’t teach in an undergraduate-level “Intro to Biology” online course. Again, with the help of fellow graduate students in the Department of Plant Pathology, I learned how to use a spectrophotometer, which measures the amount of light a solution absorbs to figure out the concentration of our pathogen mix, and a compound microscope, which provided a front-row seat of the tiny spores to count them and calculate just how infectious our ‘starter pack’ really was.

After we created the inoculum and sprayed the tomato canopies in their domes, my collaborators reassured me that sometimes symptoms don’t develop even if everything was done correctly. But to my surprise (and honestly, delight), our tomatoes got sick. Really sick. The symptoms developed quickly, and finally, I could get back to the part of the project I was most comfortable with: imaging. At last—back behind a computer screen!

This journey taught me that solving real-world problems often means crossing disciplines and comfort zones. While the intersection of AI and agriculture is inherently interdisciplinary, I had previously operated within a bubble (which is behind a computer screen, writing code). This experience pushed me beyond that boundary and helped me understand the depth and hands-on collaboration that interdisciplinary research requires. Sometimes, it means stepping away from your desk, picking up a watering can, and learning how to grow plants just so you can give them fungal infections (with love and scientific purpose, of course). Turns out, having a “brown thumb” is just part of the process when you’re engineering solutions for the future of agriculture.

The FFAR Fellows program has been instrumental in my professional development, offering opportunities to strengthen both my project management and scientific communication skills. These tools have been essential in tackling interdisciplinary challenges like the one described in this project.

The program’s mentorship component has been extremely valuable not only for having industry professionals who could speak to the interaction between autonomy and controlled experiments, but also for having trusted voices to turn to when challenges arose. We experienced several hiccups, but prepared with the interpersonal communication and project management skills I learned as a FFAR Fellow, I was able to actively participate in moving the project forward.

I would like to express my gratitude to my sponsors, Bayer and the FFAR Fellows program, as well as my advisor, Cranos Williams, for their support and for the opportunity to tailor my Ph.D. experience and the encouragement to be curiosity-driven.