What computer vision actually sees in a stressed leaf

Draft: a launch-seed article. It reflects our genuine approach and contains no invented results.

Chlorosis, lesions, wilting, mottling — the visual signatures of stress are real, consistent, and learnable. Here is a plain-language tour of how our models read a canopy.

Stress has a look

A healthy canopy has a characteristic colour, texture, and structure. Stress changes those in patterned ways: yellowing between veins points one direction, brown lesions another, a limp canopy another still. Humans read these cues instinctively; computer vision learns them from labelled examples.

Classification and segmentation

Verdyn uses two complementary kinds of model. Classification answers "is this plant showing a stress signature, and which one?" Segmentation draws the boundary — which pixels, which plants, how much. Together they turn a photo into a map.

Change over time

One flight is a snapshot. Several flights are a story. A change-detection layer compares them to show whether a hotspot is spreading or settling — often the most decision-relevant signal of all.

Why it keeps improving

Every field that is labelled and confirmed strengthens the models. Accuracy is not fixed; it compounds as the dataset grows.

This is a launch draft. Model details will be expanded as coverage grows.