A combination of AI and photography is helping wine makers keep their grapes free of disease.
听颈蝉听a fungus鈥攁lso known as grey rot鈥攚hich has听spores that pierce and infect wine grapes, causing them to shrivel and sweeten. It can add complexity and longevity to sweet wines, like Sauternes; in these cases it is known as 鈥淣oble Rot鈥�.
But it is a problem for producers of dry reds, for example, causing widespread losses of harvest in many wine regions鈥攑articularly the more humid ones where the rot proliferates.
Advertisement
Pesticides are the most common defence. Though an environmentally-kinder method is to find and then cultivate grapes that have a natural resilience. 鈥淧art of the armoury a grape possesses against Botrytis is a good bloom of epicuticular wax,鈥� says wine expert .
A little light drinking
But manually checking the wax coverage of grapes is laborious, time consuming and subject to human error. So a team at the University of Bonn and the in Germany created a mobile laboratory to automate and accelerate the process.
Grapes are placed under a light, before a series of high-resolution photos are taken that pick up patterns of scattered light鈥攊ncluding the diffuse reflections given off by the roughness of the wax.
The pictures are then analysed by two AI algorithms: one singles out the grape in the image; the other determines the distribution of waxiness over the grapes.
The team trained the AIs on 90 images with over six and a half million labelled pixels. They then took 180 images of six different grape varieties to train and test the system. In just a few seconds, it could detect the wax distribution on a grape with an accuracy of around 95 per cent.
The team would like to scale the process and say the system could be mounted onto a robot called PhenoBot, which is used to anlayse grapes in vineyards.
, a viticulture and small fruits advisor at the , says the idea is promising, and it will be very useful if it can be optimised and used in the field. It is an example of the techniques emerging from the blend of computer and plant sciences, he says.
鈥淚n my opinion, it is just the beginning. The near future will bring other un-destructive automated applications for plant phenotyping as a result of the collaboration among different branches of science.鈥�
