The project will develop a portable, phone-based imaging tool that will allow grape quality grading improvement at the bunch and fruit level in the field. It will measure berry skin colour evolution for white cultivars, and berry volume and number for red and white varieties.
The timing of grape harvest is often decided by viticulturists and winemakers using a range of objective measures of grape maturity (e.g. Brix, titratable acidity, phenolics and anthocyanins/colour). However these indices give no information about the grape aromatic potential or the resulting wine flavour profiles.
The proposed grape imaging and analysis tool will help to optimise harvesting decisions by helping growers and winemakers to target the best aromatic windows during fruit ripening as identified by berry colour. Measurements of berry volume will allow improvements to vineyard management such as irrigation scheduling to control vine water status, avoid berry water loss and berry shrivel, thereby enhancing fruit quality, volume consistency and bunch homogeneity.
The project leverages the now broad availability of high quality cameras in mobile phones, their well-developed software environments, image processing, and machine learning for data and image classification.
The project will develop an automatic vision-based colour rating system for accurately measuring grape colours. The system will be calibrated using a reference card (such as ColorChecker) and calibration curves generated for the varying light intensities, illuminations, background colours and viewing angles that would be expected in the field. Machine learning analysis of multiple grape images will be used to determine the colour of a grape sample under different measurement conditions.
Having accurately calibrated the measurements to determine colour tint angles and hue, berry counts and estimates of berry volume, size distributions will be determined using digital image processing techniques. Calibrated models that estimate berry volume and mass will be developed using classic regression algorithms. The suite of codes for image, colour and size analysis will be optimised and then cast into a compact form using ‘open source’ coding methods suitable for incorporation firstly into Android and later, iOS-based phones. The final output will be a smartphone App that incorporates features developed for berry assessment.
The portable, low entry cost phone based application arising from this research will provide growers and winemakers with a tool that will enable harvest decisions targeting preferred wine styles based upon objective measures in the field and provide real time feedback.