Cornell researchers develop system that detects grape mildew days before symptoms appear

Researchers at Cornell AgriTech in Geneva, New York, say they have developed an imaging platform that can detect grape diseases several days before symptoms appear on leaves, a step that could help breeders and vineyard managers respond sooner to two of the industry’s most damaging infections.

The system, called HyperBird, uses hyperspectral imaging to identify early signs of powdery mildew and downy mildew, according to Cornell University, which announced the work Monday. The diseases are a major source of crop losses in vineyards around the world, and earlier detection could give growers more time to adjust fungicide programs and other management decisions before damage becomes visible.

The research also has potential implications for the drinks business because grapes are the raw material for wine, and faster disease diagnosis could support more precise vineyard management. In the longer term, the same high-throughput screening method could help breeding programs move more quickly in developing disease-resistant grape varieties, which may reduce losses and lower chemical use in wine production if the approach proves effective at larger scale.

Cornell said HyperBird is designed to observe microscopic plant traits that ordinary cameras cannot capture. Standard cameras record three visible-light bands, red, green and blue. Hyperspectral imaging records hundreds of narrow wavelength bands, including information beyond the visible range. That allows researchers to detect internal changes in leaf tissue before disease marks reach the surface.

The project brings together plant pathology, engineering and grape breeding. The senior scientists involved include Katie Gold, an assistant professor of grape pathology at Cornell AgriTech; Yu Jiang, an assistant professor and systems engineer in Cornell’s School of Integrative Plant Science; and Lance Cadle-Davidson, a research plant pathologist with the U.S. Department of Agriculture’s Agricultural Research Service and an adjunct associate professor at Cornell.

Cornell said the platform can process hundreds of leaf samples within a few hours. That kind of speed matters in breeding and disease screening, where labs often need to compare very large numbers of samples. According to the university, the machine functions as a high-throughput phenotyping microscope, meaning it can measure observable traits linked to disease resistance across many samples in a short period.

One of the technical hurdles the team said it addressed was spatial resolution at the microscopic level. In hyperspectral imaging, a disease signal can be lost if the affected area is too small and its reflected light gets mixed with light from nearby healthy tissue. Jiang said in Cornell’s report that the challenge is similar to trying to isolate the signature of a small object in a large room if the image lacks enough detail. HyperBird was built to distinguish those tiny diseased areas from surrounding healthy cells.

The work builds on an earlier Cornell prototype known as Blackbird, which the team introduced in 2021. Blackbird was a robotic imaging system that recorded in red, green and blue wavelengths and was used to speed up assessment of grape leaf samples for evidence of infection. Cornell said HyperBird extends that platform with about 200 times more spectral resolution per pixel, while maintaining microscope-level spatial detail.

A May 20 study in the journal Plant Disease, led by former Cornell postdoctoral researcher Saeed Hosseinzadeh, served as an early proof of principle. Cornell said the system collected nine terabytes of hyperspectral images in a single day and showed that disease could be detected before visible symptoms appeared. The same study also found that the technology could identify fungicide residues on leaves, distinguish among treatments used against mildew, and evaluate how well those treatments worked.

That study also suggested the imaging data could help identify grapevine lineages being bred for disease resistance. Cornell said the system was able to detect subtle optical signatures tied to a plant’s genetic background, including differences in pigments, wax structure, water content and internal leaf anatomy. For grape breeders, that could offer a faster way to sort large populations and find promising plants earlier in the process.

HyperBird itself was formally described in a Sept. 8 paper in the journal Plant Phenomics. The first author was Jinhong Yu, a doctoral student in Jiang’s lab. Cornell said that paper detailed how the robotic system delivers high-throughput hyperspectral imaging at a microscopic scale and can compare small diseased regions with the average spectra across the rest of the leaf to track the progress of downy mildew and powdery mildew.

Gold said in Cornell’s report that the system can now reach, by day three, the level of accuracy that earlier methods achieved only by day six or day nine. If that performance holds up in broader use, it would mean researchers can move through disease evaluations two to three times faster than before.

A third study, published June 29 in Plant Disease, tested the system under field conditions. That paper, led by Lorenzo Pippi, a former visiting scientist in Gold’s lab from the University of Pisa, used leaf discs collected from grapevines in vineyards to compare conventional fungicides, biofungicides and untreated controls against downy mildew. Cornell said the imaging method detected early optical changes linked to infection, distinguished among sprays based on residues left on leaves, and then helped assess which treatments were working best.

That could be useful in commercial vineyards because mildew pressure often varies from one part of a field to another, making it hard to judge whether a product worked or whether infection levels were simply uneven. A tool that picks up disease earlier and separates infection signals from spray residues could help growers make more targeted decisions, though the technology would still need practical deployment beyond research settings.

For now, one of the slower steps remains sample collection. Technicians still need to cut dime-sized discs from leaves and bring them to the machine for analysis. Cornell said that is the most time-consuming part of the process and that robotics students are working to automate it.

The university said it is working with Moblanc Robotics to manufacture HyperBird. Cornell also said the platform could be adapted for other crops and diseases, and that researchers have started discussions with a company in the cheese industry, suggesting the imaging approach may have uses beyond viticulture.

The project is part of VitisGen3, a grape breeding program funded through the U.S. Department of Agriculture’s National Institute of Food and Agriculture. Cornell said other funding came from federal capacity funds awarded to the university, the President’s Council of Cornell Women Research Fund, the New York Wine and Grape Foundation, and the USDA Grape Genetics Research Unit.