← All research storiesFarming & food

Peer-reviewedFarming & food

An AI that “looks” at both the whole leaf and the sick spot can name grapevine diseases, and explain why

Researchers paired a full-leaf photo with an automatically cropped close-up of the damage, helping a language-and-vision AI identify grapevine diseases with 98.7% accuracy in tests.

How far alongPeer-reviewedReviewed by independent experts for a journal before publication. Peer review is a quality check, not a grade of evidence strength: confidence still depends on study design, limitations, replication and the wider evidence.
How soon could it reach you?A few years outPromising, but needs more testing or engineering first. Our rough editorial guess is several years, with a lot of uncertainty.
Where it comes fromDiscover Artificial IntelligenceIndustrial University of Ho Chi Minh City (Vietnam) · Oct 7, 2026Read the original →

What the scientists did

Grape growers lose crops to leaf diseases such as black rot and leaf blight, and catching them early matters. Many AI tools can classify leaf photos, but they act like black boxes and often get distracted by the background in real field photos.

The researchers fed an AI two views of each leaf: the whole leaf, and a close-up of the damaged spot that the computer found and cropped by itself using color differences. The AI was a vision-language model, meaning it handles both images and text, so it can name the disease and point to what it was looking at. They trained and tested it on 9,027 grapevine leaf images from a public dataset called PlantVillage.

Words to know

Vision-language model

An AI that can look at pictures and also read and write text about them.

  • Describing what’s in a photo
  • Answering “What disease does this leaf have?”

Plant disease classes

The categories the AI had to choose between in this study.

  • Black rot
  • Esca
  • Leaf blight
  • Healthy leaf

Attention map

A heat map showing which parts of an image the AI focused on when it made its decision.

  • Highlighting the brown spots on a leaf
  • Showing the AI ignored the soil in the background

What they found

On test images it hadn’t seen, the system was 98.7% accurate overall. Using only the full leaf gave 97.8%, and using only the close-up gave 90.3%, so combining both views helped. The untrained starting model managed just 18.6%.

On messier real-world photos from a different dataset, the system scored 80.5%, while a standard image classifier that scored 99.8% on the lab-style images fell to 54.9%.

98.7%accuracy on lab-style test images
80.5%accuracy on real field photos it wasn’t trained on
9,027grapevine leaf images used

Why it matters to you

Tools like this could eventually let a farmer, gardener or student snap a phone photo of a sick leaf and get both a likely diagnosis and a highlighted picture showing why, so they can act sooner and use fewer chemicals. Healthier vines mean steadier supplies of grapes, raisins and juice.

Where this could lead

Phone-based crop checks

An app could identify common grape diseases from a photo and show the spots it used to decide.

3–10 years

Targeted spraying

Knowing which vines are sick could help growers treat only those plants.

3–10 years

These are possibilities the research points toward, not promises. Most early findings take years of testing before they reach everyday life, and some never do.

Why scientists care

Farmers and plant experts are more likely to trust an AI that can show its reasoning. This study is part of a push to make agricultural AI both accurate and explainable, especially in messy field conditions.

Keep in mind

Most images came from PlantVillage, which shows single leaves on plain backgrounds, unlike real vineyards. Accuracy dropped to about 80% on field photos, and as low as 60.9% for black rot. Each image took over 2 seconds to analyze on the graphics card used. It covers only four categories and one crop.

How far along is the science?

  • Peer-reviewedReviewed by independent experts for a journal before publication. Peer review is a quality check, not a grade of evidence strength: confidence still depends on study design, limitations, replication and the wider evidence.
  • PreprintA manuscript shared publicly, often before or alongside journal review. Check its separate journal and review status (each story says what we know). Results may change.
  • Conference talkPresented to other scientists at a meeting. Usually short and early; a full paper may come later.
  • Agency reportPublished by a science agency such as NASA, NOAA or NIH. Reviewed internally, but not by an outside journal.
  • Press releaseAn announcement from a university or company. Useful, but always check the research it describes.

How soon could it reach you?

  • Could reach you soonCould show up in products, services or advice soon. Our rough editorial guess is about 1 to 3 years, not a forecast from the research.
  • A few years outPromising, but needs more testing or engineering first. Our rough editorial guess is several years, with a lot of uncertainty.
  • Long-term scienceBasic research that builds knowledge. Real-world uses, if any, are far off.
  • Small stepA small update to earlier work. Useful to researchers, but not a change you would notice yet.

Every story links to its original source: the paper itself, or the conference program when no paper is out yet. Explanations are written with AI, usually from the paper’s abstract (each story says what it was written from), then checked against the source. We only use journals, recognized preprint servers, science agencies, scientific conferences and established science news outlets.