The Eyes Have It: Using Human Iris Identification Technology to Improve Upon Wine Grape Harvest Date Estimations

The decision on when to harvest wine grapes depends on many factors, with the most important being the physiochemical maturation of the berries (total soluble solids – TSS, phenolic compounds, sugars, acids, tannins, anthocyanins, etc.). Winemakers can get a good sense of when it is time to harvest by sampling grapes as well as using technologies such as refractometers to measure sugar levels in the vineyard. They can also measure pH and titratable acidity (TA) in the lab for additional information regarding the physiochemical maturity of the grapes.

Since a color change also occurs during the maturation of wine grapes (particularly red grapes), researchers suggest that it may be possible to determine grape quality (and thus a more precise time to harvest) based upon analysis of images in the visible RGB (red-green-blue) spectrum. Some studies have shown correlations between RGB images and values of anthocyanins, total phenols, flavonoids, antioxidant activity, etc. for bayberry juices as well as blackberries.  Theoretically, the same might be possible with grapes.

The traditional method for using RGB images to analyze physiochemical characteristics involves acquiring average pixel values for the red, green, and blue channels. One big problem with this method is that any color gradients present in the fruit will be hidden, which would be problematic if using it for maturation purposes, as grapes (and other fruits) undergo a relatively lengthy color change period before full maturation is achieved. Additionally, there can be areas of greater pixel “noise” in images of grapes, specifically in the center where there is greater LED light reflection from the camera/flash/lighting that can result in less-than-accurate pixel number estimates.

A new study, published January 2020 in the journal Computer and Electronics in Agriculture, aimed to evaluate a new technique for analyzing RGB images as it relates to wine grape maturation.  Specifically named IRIS-GRAPE, this new method treats the image of the grape like it would the image of the iris of the human eye.

The iris in the human eye is the colored circular part, the complexities of which vary from person to person and is thus able to be utilized in biometric recognition of individuals.  The very center of the eye, the pupil, is not used in this analysis, as there is no color or other complexities in the black pupil and thus would be ineffective in individual recognition.

RGB images of grapes are similar to pictures of the human eye – if you could imagine the grape representing the iris, the center LED reflection from the camera portion representing the pupil, and the background portion representing the white part of the eye (sclera). IRIS-GRAPE utilizes similar techniques as biometric recognition in humans and applies it the RGB image of the grape.

iris grape
Fig. 3 from Costa et al, 2020

The new study mentioned above aimed to analyze the RGB images of wine grapes using the new IRIS-GRAPE method by comparing the results with actual chemical measurements of the grapes.

Brief Methods

Due to the insanely technical details of the computer systems involved with this study, I plan to leave those out of this post.  If you have specific questions regarding this technical side of things, just ask in the comments and I’d be happy to look into it for you.

Syrah and Cabernet Sauvignon from the sub-middle São Francisco Valley, Brazil, were used in this study. Berries were collected 6 times over the course of the growing season, representing different stages of maturation (432 Syrah and 432 Cab Sauv grapes total collected).

Images of grapes were taken, and chemical analysis performed (total soluble solids – TSS, total anthocyanins, and yellow flavonoids).

Image analysis were performed using two different methods, the first “traditional” RGB image analysis method, and the second new IRIS-GRAPE image analysis method.

Multiple regression analysis was performed on the image data to help build the computer model that would ultimately predict the quality attributes of the grapes (and thus were compared to the real chemical analysis for accuracy).

Brief Results and Conclusions

  • For both Syrah and Cabernet Sauvignon grapes, the IRIS-GRAPE method had greater predictive power for TSS, total anthocyanins, and yellow flavonoids (all parameters studied) than the traditional image method.
  • The mean squared error (MSE) using the IRIS-GRAPE method was significantly lower than the MSE calculated using the traditional image method.
    • NOTE: MSE is a measure of the quality of a predictor.  The closer to zero it is, the better.  For example, the MSE for the IRIS-GRAPE method for Syrah grapes was 4.264, while the MSE for the traditional image method for Syrah grapes was 6.652).
  • The predictive models for total anthocyanins based on the IRIS-GRAPE method were better than the predictive models for TSS using this same method.
    • Despite the better performance for total anthocyanins, the IRIS-GRAPE method was still better than the traditional image method for all parameters studied.

Overall, it appears as though the IRIS-GRAPE method for analyzing images of wine grapes might be a viable method for determining grape ripeness and ultimately optimum harvest date.  Of course, this being the first time this method was tested in wine grapes, much more research should be done, however, the results of this initial study are promising. 

For two red grapes, this method of analysis was superior to more traditional image analysis techniques, with the IRIS-GRAPE method serving as a potential program for future electronic devices that could be easily used in vineyard without causing any damage to the grapes (while still providing decent accuracy).

I would be interested to see the IRIS-GRAPE image analysis technology tested on many more wine grapes, including white wine grapes.  Would they even work with white wine grapes?  Yes, they undergo a slight color change during the maturation process, but not nearly as obvious as red grapes.  Would this IRIS-GRAPE technology still be accurate?  Theoretically, if they can use this same software to identify individual people, even those with very light-colored eyes, I imagine it would still work with white wine grapes.  More research is certainly needed.

In terms of cost, the researchers noted that the cost of running this sort of analysis is much cheaper than using more traditional spectrometer methods, so developing a device that can utilize this computer program would be beneficial not just for accuracy but also for financial reasons.

This IRIS-GRAPE image analysis technology for predicting a more accurate harvest time for wine grapes certainly has potential, and hopefully more work will be done in this area to develop a user-friendly handheld device that can be used quickly in the vineyard.

Source:

Costa, D.d.S., Neto, R.F.d.O., Ramos, R.P., Oliveira, V.G.d.S., and Teruel, B. 2020. IRIS-GRAPE: An approach for prediction of quality attributes in vineyard grapes inspired by iris biometric recognition. Computers and Electronics in Agriculture 168: 105140.

1 comment for “The Eyes Have It: Using Human Iris Identification Technology to Improve Upon Wine Grape Harvest Date Estimations

Comments are closed.