SIU Builds Soybean-Scanning Robot, With Disease-Detection AI Still to Come

A soybean disease can hide where a drone struggles to look: beneath the leaves and near the stem. Southern Illinois University Carbondale researchers have built a field robot to collect images from those angles. They hope the data will eventually support AI that spots frogeye leaf spot early enough to guide treatment, but the detection models remain unfinished.

The view from above fell short

Assistant professor Billy Ram and doctoral student Samuel Singh are working in a soybean field at SIU’s University Farms that has frogeye leaf spot. The disease causes leaf lesions and can spread to stems and pods. Their goal is to gather enough plant data to predict disease before symptoms appear on leaves and stems.

They first flew a drone over the field to make maps using light wavelengths beyond human vision. Even at low altitude, its camera did not gather enough information to detect frogeye leaf spot. It struggled to capture lesions beneath leaves and lower on the plants. That gap led the team to build a machine that could travel among the rows.

A machine for collecting closer views

The battery-powered robot has four wheels, GPS and cameras mounted at several positions. A user can upload a field map and set it to follow the soybean rows autonomously. The cameras collect the close views the drone missed; the robot is not yet an automated diagnosis tool.

Ram described the immediate challenge as an engineering one: collecting a large amount of useful data while equipment moves through a field. More camera angles address what the researchers could not see from above. Whether those images can support the early predictions they want is the next part of the project, not a result they have reported.

The warning system comes next

Ram and Singh plan to develop AI models to detect frogeye leaf spot from the plant data. Ram wants the eventual system to track individual plants, identify whether they have a disease and report how much of a crop is affected. Those are intended functions, not capabilities the team says its current models have achieved.

Singh wants maps that point farmers to disease hot spots so they can treat selected areas rather than spray a whole field. The distinction matters to Ram because treatment can come too late: he said a heavily affected field may yield only 60% to 70% of what it otherwise would. The project has received funding through the Illinois Soybean Center.

The researchers hope farm-equipment companies will eventually adapt the technology for tractor or sprayer attachments.