Premium
Three‐dimensional reconstruction of Vitis vinifera (L.) cvs Pinot Noir and Merlot grape bunch frameworks using a restricted reconstruction grammar based on the stochastic L‐system
Author(s) -
Xin B.,
Liu S.,
Whitty M.
Publication year - 2020
Publication title -
australian journal of grape and wine research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.65
H-Index - 77
eISSN - 1755-0238
pISSN - 1322-7130
DOI - 10.1111/ajgw.12444
Subject(s) - computer science , vitis vinifera , algorithm , mathematics , artificial intelligence , horticulture , biology
Background and Aims Phenotypic traits of grape bunches are known to be related with grapevine yield, wine flavour and sensitivity to disease. Aiming to solve a phenotypic bottleneck in current breeding studies as well as to improve the performance of phenotypic tools, we put forward a combination of grammar‐based reconstruction and vision‐based reconstruction, and propose an empirical reconstruction grammar restricted by an outline hull, which can model parameters of the entire bunch framework. Methods and Results Statistical analysis of manual measurements of bunches was undertaken to empirically build a reconstruction grammar for a specific grape cultivar. During the reconstruction procedure, the grammar takes account of the estimation of the topological architecture and the geometrical parameters of bunch elements, while the outline hull formed from the input two‐dimensional (2D) image is used to constrain the volume and the overall shape of the bunch model. The reconstruction results indicated that the average percentage error of quantity estimation for various internode types ranged from 19.1 to 41.1% , and the average percentage error for individual lengths of respective internode types ranged from −0.4 to 10.4% . Conclusions The proposed three‐dimensional grape bunch reconstruction method achieves the parameter modelling of bunch components by using 2D images as input, and the performance has been shown to be an improvement over existing work. Significance of the Study The proposed method enables a more accurate reconstruction of grape bunch framework, which facilitates the automatic extraction of phenotypic traits and the improvement of breeding programs along with vineyard management. Due to its simple sensor input requirements, it is able to be applied under field conditions.