Digital Agriculture Food and Wine Group, School of Agriculture and Food, Faculty of Veterinary and Agricultural Sciences, University of Melbourne, Parkville, VIC 3010, Australia.
Sensors (Basel). 2021 Nov 3;21(21):7312. doi: 10.3390/s21217312.
Berry cell death assessment can become one of the most objective parameters to assess important berry quality traits, such as aroma profiles that can be passed to the wine in the winemaking process. At the moment, the only practical tool to assess berry cell death in the field is using portable near-infrared spectroscopy (NIR) and machine learning (ML) models. This research tested the NIR and ML approach and developed supervised regression ML models using Shiraz and Chardonnay berries and wines from a vineyard located in Yarra Valley, Victoria, Australia. An ML model was developed using NIR measurements from intact berries as inputs to estimate berry cell death (BCD), living tissue (LT) (Model 1). Furthermore, canopy architecture parameters obtained from cover photography of grapevine canopies and computer vision analysis were also tested as inputs to develop ML models to assess BCD and LT (Model 2) and the intensity of sensory descriptors based on visual and aroma profiles of wines for Chardonnay (Model 3) and Shiraz (Model 4). The results showed high accuracy and performance of models developed based on correlation coefficient (R) and slope (b) (M1: R = 0.87; b = 0.82; M2: R = 0.98; b = 0.93; M3: R = 0.99; b = 0.99; M4: R = 0.99; b = 1.00). Models developed based on canopy architecture, and computer vision can be used to automatically estimate the vigor and berry and wine quality traits using proximal remote sensing and with visible cameras as the payload of unmanned aerial vehicles (UAV).
浆果细胞死亡评估可以成为评估重要浆果品质特性（如在酿酒过程中可传递到葡萄酒中的香气特征）的最客观参数之一。目前，评估田间浆果细胞死亡的唯一实用工具是使用便携式近红外光谱（NIR）和机器学习（ML）模型。本研究测试了 NIR 和 ML 方法，并使用来自澳大利亚维多利亚州雅拉谷葡萄园的设拉子和霞多丽浆果和葡萄酒开发了监督回归 ML 模型。使用 NIR 测量值作为输入，开发了一个 ML 模型，用于估计浆果细胞死亡（BCD）和活组织（LT）（模型 1）。此外，还测试了从葡萄树冠摄影和计算机视觉分析获得的冠层结构参数，作为输入来开发 ML 模型，以评估 BCD 和 LT（模型 2）以及基于霞多丽（模型 3）和设拉子（模型 4）葡萄酒的视觉和香气特征的感官描述符的强度。结果表明，基于相关系数（R）和斜率（b）（M1：R = 0.87；b = 0.82；M2：R = 0.98；b = 0.93；M3：R = 0.99；b = 0.99；M4：R = 0.99；b = 1.00）开发的模型具有较高的准确性和性能。基于冠层结构和计算机视觉开发的模型可以用于使用近程遥感和可见摄像机作为无人驾驶飞行器（UAV）的有效载荷自动估计活力以及浆果和葡萄酒的品质特性。
