Currently at work to develop robotic technology for crop pollination, Washington State University professor Manoj Karkee has taken on yet another task — development of a smartphone app to help wine grape growers estimate tonnage of their crop.

Recognizing that brain-power often beats brawn-power, Karkee and co-researchers are using a $1 million USDA grant to adapt a robotic pollinator, a job now done mainly by honey bee hives.  As envisioned, machine learning systems would locate flower blossoms, determine the best stage to pollinate a given crop, and develop a hand and arm to spread that pollen.

“This could be a huge help for the agriculture industry in the future,” he admitted before discussing a concurrent project of specific interest to grape growers.

Speaking before a Washington Advancement in Viticulture and Enology seminar on how smartphones in the field can help growers guesstimate their wine grape crop, he said: “Automating the labor-intensive task of crop estimation has long been a research goal of the wine grape industry.  Each summer, growers spend thousands of hours counting and weighing berries and clusters to estimate a yield, a vital step to guide vineyard management decisions.

“But that doesn’t provide real-time data because estimates represent only one point in time while weather events that take place after that data collection can skew final numbers,” said the WSU biological systems engineer.

“We’re working to develop a low-cost smartphone app that can detect and count the number of clusters — and the berries within each cluster — by using the phone’s camera to acquire images uploaded to a cloud computing platform.”

The concept and application is nothing new for Karkee, whose research team has already developed a similar smartphone app for yield estimation in apple orchard, a process that tested out at 98% accuracy in yield estimation of orchard trials.  That app is being expanded for field scale crop estimation and to include other features such as plant stress detection.

Simple approach

“This could be a simple, practical approach for efficient vineyard management,” Karkee told his WAVEx webinar audience.  “It offers in-hand, quick solutions because it’s portable, scalable, and affordable, another example of how mobile technology can help in the process of counting and sizing berries.

“While further validation is needed, this mobile vision tool for wine grapes is a step towards the goal of running something down the row and getting an accurate count.

“Smartphones today are ubiquitous and are equipped with cameras, GPS, 3D, and a variety of sensors so machine vision for Smartphone platforms is easily adopted for a simple, Click and Go operation of taking images to help determine planting density, the number of producing vines per acre, the average number of clusters per vine, and the average cluster weight.”

Noting that current manual counting efforts were often time-consuming, labor-intensive, and inaccurate and that earlier efforts at automated estimation tools were “generally complex and costly machines that lacked portability and found no commercial success,” Karkee said the first phase of trials — acquiring images of sample vines with berries and clusters in near real-time and developing calibrated correlation models for size and weight — have gone well in vineyard testing in Prosser, more tests are to be conducted as the current season progresses and next-phase development research continues with new capabilities to be added.

“We’ve found the fruit and sizing counts to be quite reliable in terms of accuracy.  This will be a promising tool for growers.”