Physical and chemical bunch characterisation of table grapes using machine vision

PROJECT TITLE: Physical and chemical bunch characterisation of table grapes using machine vision

Project leader:  Dr Carlos Poblete-Echeverría

Duration: 1 September 2021 – 31 December 2024

Project Summary:

The current methods for determining chemical and physical quality parameters of table grapes are destructive and time-consuming. In this sense, physical parameters are evaluated using generally subjective means (colour scorecards” and other visual classification systems) which can be prone to human errors. Additionally, the inherent variability which exists among berries on a grape bunch further complicates the matter of accurate evaluation of quality parameters and is one reason for the necessity of collecting large representative samples (a very time-consuming action). The destructive nature of the methods also means that the same sample cannot be evaluated over time. Therefore, there is a need for a reliable robust and non-destructive method for assessment and quantification of chemical and physical bunch characteristics – both in the vineyard and the pack store. This project will be a starting point for developing such a robust method with an in-field application for the table grape industry. It will also build on existing knowledge and adapt previously researched methodologies and algorithms for in-filed application.
The industry will benefit from a method in which grape quality evaluation can be carried out for multiple parameters simultaneously and automatically. Such a system will allow for accurate and timeous decision making related not only to post-harvest aspects, but also to management practices (particularly bunch shortening/ preparation actions), crop estimation and harvest related decision. Accuracy of these decisions will prevent the harvest and packing of grapes that do not meet the minimum export requirements, which in turn reduces waste and economic losses. Machine vision has been investigated as an alternative method to physical and chemical measurements in various crops. The challenge is developing accurate models and suitable protocols for image acquisition under field conditions. The determination of optimal protocols and models will be the first step in developing new and improved technologies for application in research and practice (in both the vineyard and the pack store). Established algorithms from previous studies (e.g., Daniels et al., 2019) will be evaluated in the context of this study and may be key in the development of a machine vision system to automate the evaluation of visual observations of grape bunches.
This project aims to develop a practical and robust methodology using machine vision on intact bunches for determining multiple chemical and physical quality parameters simultaneously.