“I’ve come to realize that AI isn’t a magic tool that works everywhere. It needs careful data preparation, special algorithms designed for each task and real-world testing to make sure it’s reliable,” Sun said. “In agriculture, where data is often messy and unpredictable, AI models need to be strong, easy to understand and based on the knowledge of experts in the field.”
NDSU’s FEWS research projects study how soil, crops, water, and energy interact. AI is used to identify patterns and support decision-making. This enables the processing of data from sensors, soil samples, and satellite or drone imagery in real-time, providing producers with fast and accurate recommendations.
For the FEWS research Sun is conducting, AI is proving to be a good fit. Sun’s team utilizes AI for soil health assessment, employing autonomous, uncrewed ground vehicles to classify soil salinity and organic matter levels. Computer models trained on field imagery help with identifying weeds and crops; these models allow researchers to differentiate between crop species and invasive weeds, such as Palmer amaranth and kochia. AI-powered image recognition determines symptoms of diseases with models deployed on edge devices for near real-time diagnosis.
“We’ve taken soil and crop monitoring to the next level by bringing AI into the mix,” Sun said. “In the past, collecting and analyzing samples took weeks, but now we can get real-time insights right in the field. Our models are super accurate, consistently predicting soil salinity and plant disease. This means our field trials can help us make better management decisions faster.”
For decades, many public breeding researchers at NDSU relied on pen and paper to record data obtained from the field. At the same time, private breeders had already begun using automation, cutting-edge analytics, and machine learning to accelerate genetic gain.
“Bridging this gap is not just about technology, it’s about equity, efficiency, and innovation. Empowering public breeding programs with modern analytics and digital infrastructure will level the playing field, enabling breeders everywhere to make faster, smarter, and more impactful decisions for global food security,” Heilman Morales said.
Heilman Morales’s team utilizes AI in AgSkySight, a software platform designed to streamline the process of drone image stitching and vegetation index calculations for agricultural research. AGSkySight turns drone images into detailed data for each field plot.
AGSkySight converts UAV mosaics into plot-level data products using a custom YOLO detector enhanced with SAHI (Slicing Aided Hyper Inference). The YOLO model is trained on research-field layouts (ranges × rows, alleys, borders) to recognize plot boundaries even under variable lighting, dense canopy, or stitching artifacts.
Analyzing omics data is challenging due to its large and complex nature, which necessitates the use of advanced tools and expertise. There is a need for systems that combine traditional statistics with AI/ML to help plant breeding and biological research. PredictPro is a user-friendly software that facilitates working with complex genomics and phenomics data. It allows users, even without coding skills, to run genomic prediction models using standard statistical methods or AI/ML algorithms.
“AI/Machine Learning facilitates workflows and speeds up performance and tasks, also reduces cost as time management and by using prediction instead of actual plots in the field as physical experiments translate into cost savings and speeding up the breeding pipeline,” Heilman Morales said.
Dey’s background includes designing antennas, RFID sensors and other electromagnetic devices. When working with those sensors, Dey saw that many data was generated, but there was a need to translate the data into clear, concise information. It turns out that the same holds for the FEWS research project he is working on.
“I decided to bring AI into our FEWS project because the kinds of data we deal with in food, energy, and water systems are complex, dynamic, and often pretty noisy,” Dey said. “We’re collecting information from different types of sensors, things like soil moisture measurements, electromagnetic signals, environmental data streams, and these signals don’t always behave in neat, predictable ways. Traditional analysis methods can overlook subtle trends and often struggle to adapt when field conditions change. That’s where AI really adds value.”
An example of AI's influence on Dey’s research is soil moisture sensing. His team developed a low-cost, passive RF sensor that can be distributed throughout fields and paired with machine learning models to calibrate its readings against soil properties, such as bulk density and electrical conductivity.
Source : ndsu.edu