Assessment of grape cluster yield components based on 3D descriptors using stereo vision

Abstract

Wine quality depends mostly on the features of the grapes it is made from. Cluster and berry morphology are key factors in determining grape and wine quality. However, current practices for grapevine quality estimation require time-consuming destructive analysis or largely subjective judgment by experts.

The purpose of this paper is to propose a three-dimensional computer vision approach to assessing grape yield components based on new 3D descriptors. To achieve this, firstly a partial three-dimensional model of the grapevine cluster is extracted using stereo vision. After that a number of grapevine quality components are predicted using SVM models based on new 3D descriptors. Experiments confirm that this approach is capable of predicting the main cluster yield components, which are related to quality, such as cluster compactness and berry size (R2 > 0.80, p < 0.05). In addition, other yield components: cluster volume, total berry weight and number of berries, were also estimated using SVM models, obtaining prediction R2 of 0.82, 0.83 and 0.71, respectively.