摘要
西北地区具有独特的风土条件,适宜酿酒葡萄的种植,已形成“大产区大品牌”的产业规划。但西北地区葡萄酒的地域风格尚不明晰,导致产业的规范性不足,可持续发展存在隐患。该研究分析西北地区干红葡萄酒的23项重要的色泽-味感理化指标,旨在开发一种基于机器学习技术的葡萄酒判别方法,实现西北地区干红葡萄酒产地和酒龄的精准判别。首先,以西北地区200款干红葡萄酒为研究对象,通过理化试验测定了总酚、总花色苷、滴定酸等23项色泽-味感理化指标;然后,采用Pearson相关系数分析了西北产区葡萄酒质量特征的一致性,并耦合随机森林(Random Forest,RF)分析理化指标对干红葡萄酒产地和酒龄表征的贡献度;最后,基于人工神经网络(Artificial Neural Networks,ANN)构建了西北地区干红葡萄酒产地和酒龄的精准判别模型。结果表明,色泽相关指标的模型贡献率合计31%,花色苷相关指标的模型贡献率合计26%,酚类物质相关指标的模型贡献率合计21.1%。模型对宁夏产区酒样判别的灵敏度(Sensitivity,SEN)为98.72%,准确率(Accuracy,CCR)98.72%;对新疆产区酒样判别的SEN为95.45%,CCR为100%;对甘肃产区酒样判别的SEN为100%,CCR为95.45%。该方法可以实现西北产区干红葡萄酒产地和酒龄的精准判别,可为中国西北地区优质葡萄酒的生产和产品的市场监管提供科学依据。
Northwest China has formed into a benign industrial area named “big appellation and big brand”. The unique black soil can also be a benefit for wine grape planting. However, the regional flavors of dry red wine vary greatly in the different regions in Northwest China. Potential risk can remain for the sustainable development of the wine industry. In this study, a wine discrimination system was developed to accurately identify the origin and age of dry red wine using machine learning. 23physical and chemical parameters were determined about the wine color, taste, and aroma. Firstly, 600 samples of dry red wine were collected from the different wineries in Ningxia, Gansu, and Xinjiang areas. The total phenols, total anthocyanin, and titrated acids were then measured during the experiment. Secondly, the Pearson correlation coefficient formula was selected to evaluate the consistency of 23 dry red wine parameters. Subsequently, random forest(RF) was used to calculate the percentage contributions of each parameter for the identification of the origin and age of dry red wines. Finally, an accurate discrimination system was developed to identify the origin and vintage of dry red wines using an artificial neural network(ANN) classifier.The results showed that the eight color-related parameters provided with 31% reference to distinguish the origin and year of dry red wine. Specifically, the wine samples from Ningxia’s wineries showed higher yellowness and brick red. The wine samples from Xinjiang’s wineries showed a high degree of redness, like the peach red. The wine samples from Gansu’s wineries showed higher blueness and purple. Five anthocyanin and seven phenolic-related parameters were provided with 26%and 21% reference, respectively. In terms of taste, Ningxia’s wine samples had less astringency, while Gansu’s wine samples showed a stronger astringency, and Xinjiang’s wine samples tasted the most astringent. The loading analysis demonstrated that the Ningxia wine shared a certain aging potential because the old wine from the Ningxia showed more flavor characteristics.By contrast, the new wine from Xinjiang presented strong flavor characteristics, but the old wine showed weak flavor characteristics, indicating the quick dissipation of flavor as time increased. Similar flavor characteristics were achieved in the old and new Gansu wine, indicating the slow quality loss of Gansu wine during aging. The sensitivity(SEN) and accuracy(CCR) of the wine’s origin discriminant model were 98.72% and 98.72% for the wine samples from Ningxia’s wineries,respectively, 95.45% and 100% for the wine samples from Xinjiang’s wineries, respectively, while 100% and 95.45% for the wine samples from Gansu’s wineries, respectively. The correct rate of the wine’s age discriminant model for each region was 98.7% for the Ningxia wine, and 100% for the Xinjiang wine and Gansu wine. The discrimination model can be expected to make accurate discrimination of wine origin and age from Northwest China. The finding can provide scientific support for the production of premium wine in the market supervision in these regions.
作者
白雪冰
杨佳宁
姜醒睿
赵宇
武运
陶永胜
Bai Xuebing;Yang Jianing;Jiang Xingrui;Zhao Yu;Wu Yun;Tao Yongsheng(College of Enology,Northwest A&F University,Yangling 712100,China;Ningxia Helan Mountain's East Foothill Wine Experiment and Demonstration Station,Northwest A&F University,Yongning 750104,China;College of Food Science and Pharmacy,Xinjiang Agricultural University,Urumqi 830052,China)
出处
《农业工程学报》
EI
CAS
CSCD
北大核心
2022年第13期319-326,共8页
Transactions of the Chinese Society of Agricultural Engineering
基金
国家重点研发计划项目(2019YFD1002504)
宁夏自治区科技重点研发计划项目(2018BBF02001)
新疆维吾尔自治区科技支疆项目计划(2022E02011)。
关键词
主成分分析
模型
干红葡萄酒
产地判别
酒龄判别
理化指标
随机森林
人工神经网络
principal component analysis
models
dry red wine
origin discrimination
age discrimination
physico-chemical index
random forest
artificial neural networks
