Recognition of tea plant variety and grade is essential for tea germplasm resources protection. The rapid and accurate acquisition of phenotype of tea leaves is a crucial step in exploring the variety type, nutrition status, and yield prediction. Monitoring the phenotypic characteristics of tea leaves is necessary for intelligent tea germplasm management. This study analyzed phenotypic features of tea leaves based on multispectral imaging technology. Tea leaf images of 12242 sets from 25 different types, along with 61 groups of chemical characteristics of fresh tea leaves were obtained. A total of 92 indicators were extracted, and 38 indicators were screened using the successive projection algorithm and the shuffled frog leaping algorithm, which showed significant differences among different tea varieties. The phenotypic indexes of different tea varieties were analyzed, and a tea variety recognition model was established based on these indexes combined with gray wolf optimization-support vector machine algorithm. The average accuracy of the training, test, and validation sets were 99.74 %, 92.17 %, and 91.56 %, respectively. Additionally, quantitative evaluation for tea plant germplasm resources was explored. Stepwise Fisher discriminant analysis was used to identify the 61 tea plant germplasm resources, achieving an accuracy of 93.44 % with the discrimination accuracy of each grade is above 90 %.

中文翻译：

---

基于多光谱成像技术的茶树品种叶片表型差异分析及品种识别


茶树品种和等级的识别对于茶种质资源保护至关重要。快速、准确地获取茶叶表型是探索品种类型、营养状况和产量预测的关键一步。监测茶叶表型特征是茶种质资源智能化管理的必要条件。本研究基于多光谱成像技术分析了茶叶的表型特征。获得了25个不同类型的12242组茶叶图像，以及61组新鲜茶叶的化学特征。共提取92个指标，利用逐次投影算法和混合蛙跳算法筛选出38个指标，不同茶品种间差异显着。分析不同茶树品种的表型指标，并基于这些指标结合灰狼优化-支持向量机算法建立茶树品种识别模型。训练集、测试集和验证集的平均准确率分别为 99.74%、92.17% 和 91.56%。此外，还对茶树种质资源的定量评价进行了探索。采用逐步Fisher判别分析法对61份茶树种质资源进行鉴定，准确率达到93.44%，各等级判别准确率均在90%以上。