Li He, Wang Yu, Fan Kai, Mao Yilin, Shen Yaozong, Ding Zhaotang
Tea Research Institute, Qingdao Agricultural University, Qingdao, China.
Tea Research Institute, Shandong Academy of Agricultural Sciences, Jinan, China.
Front Plant Sci. 2022 Jul 22;13:898962. doi: 10.3389/fpls.2022.898962. eCollection 2022.
Tea height, leaf area index, canopy water content, leaf chlorophyll, and nitrogen concentrations are important phenotypic parameters to reflect the status of tea growth and guide the management of tea plantation. UAV multi-source remote sensing is an emerging technology, which can obtain more abundant multi-source information and enhance dynamic monitoring ability of crops. To monitor the phenotypic parameters of tea canopy more efficiently, we first deploy UAVs equipped with multispectral, thermal infrared, RGB, LiDAR, and tilt photography sensors to acquire phenotypic remote sensing data of tea canopy, and then, we utilize four machine learning algorithms to model the single-source and multi-source data, respectively. The results show that, on the one hand, using multi-source data sets to evaluate H, LAI, W, and LCC can greatly improve the accuracy and robustness of the model. LiDAR + TC data sets are suggested for assessing H, and the SVM model delivers the best estimation (Rp = 0.82 and RMSEP = 0.078). LiDAR + TC + MS data sets are suggested for LAI assessment, and the SVM model delivers the best estimation (Rp = 0.90 and RMSEP = 0.40). RGB + TM data sets are recommended for evaluating W, and the SVM model delivers the best estimation (Rp = 0.62 and RMSEP = 1.80). The MS +RGB data set is suggested for studying LCC, and the RF model offers the best estimation (Rp = 0.87 and RMSEP = 1.80). On the other hand, using single-source data sets to evaluate LNC can greatly improve the accuracy and robustness of the model. MS data set is suggested for assessing LNC, and the RF model delivers the best estimation (Rp = 0.65 and RMSEP = 0.85). The work revealed an effective technique for obtaining high-throughput tea crown phenotypic information and the best model for the joint analysis of diverse phenotypes, and it has significant importance as a guiding principle for the future use of artificial intelligence in the management of tea plantations.
茶树高度、叶面积指数、冠层含水量、叶片叶绿素和氮浓度是反映茶树生长状况及指导茶园管理的重要表型参数。无人机多源遥感是一项新兴技术，它能获取更丰富的多源信息并增强作物动态监测能力。为更高效地监测茶树冠层表型参数，我们首先部署配备多光谱、热红外、RGB、激光雷达和倾斜摄影传感器的无人机来获取茶树冠层表型遥感数据，然后利用四种机器学习算法分别对单源和多源数据进行建模。结果表明，一方面，使用多源数据集评估茶树高度（H）、叶面积指数（LAI）、冠层含水量（W）和叶片叶绿素含量（LCC）能大幅提高模型的准确性和稳健性。建议使用激光雷达 + 热红外（LiDAR + TC）数据集评估茶树高度，支持向量机（SVM）模型给出最佳估计（决定系数Rp = 0.82，均方根误差RMSEP = 0.078）。建议使用激光雷达 + 热红外 + 多光谱（LiDAR + TC + MS）数据集评估叶面积指数，SVM模型给出最佳估计（Rp = 0.90，RMSEP = 0.40）。建议使用RGB + 热红外（TM）数据集评估冠层含水量，SVM模型给出最佳估计（Rp = 0.62，RMSEP = 1.80）。建议使用多光谱 + RGB（MS + RGB）数据集研究叶片叶绿素含量，随机森林（RF）模型提供最佳估计（Rp = 0.87，RMSEP = 1.80）。另一方面，使用单源数据集评估叶片氮浓度（LNC）能极大提高模型的准确性和稳健性。建议使用多光谱数据集评估叶片氮浓度，RF模型给出最佳估计（Rp = 0.65，RMSEP = 0.85）。这项工作揭示了一种获取高通量茶冠表型信息的有效技术以及用于多种表型联合分析的最佳模型，并且作为未来人工智能在茶园管理中应用的指导原则具有重要意义。
