Grapevine bunch weight estimation using image-based features: comparing the predictive performance of number of visible berries and bunch area

Recent advances in machine vision technologies have provided a multitude of automatic tools for recognition and quantitative estimation grapevine bunch features 2D images. However, converting them into weight (BuW) is still big challenge. This paper aims to compare the explanatory power number visible berries (#vBe) area (BuA) images, order predict BuW. A set 300 bunches from four cultivars were picked at harvest imaged using digital RGB camera. Then each was manually assessed several morphological attributes and, image, #vBe visually while BuA segmented manual labelling combined with an image processing software. Single multiple regression analysis between BuW image-based variables performed obtained models subsequently validated two independent datasets.The high goodness fit all linear indicates that either one can be used as accurate proxy actual general model also suitable. The comparison predicting showed based on predictor had slightly lower coefficient determination (R2) than BuA. combination produced similar or noticeably higher R2 those single-predictor models. adding second variable more generalised gain accuracy simple Our results recommend use variables, they generally robust single When gains by feature are small, option only chosen; such case, our indicate would less cultivar-dependent #vBe.