基于树叶样本的柑橘黄龙病近红外快速诊断技术已经被证明可行, 但目前的研究尚局限于以树叶为光谱采集部位。 树皮韧皮部作为病菌及特异性营养组分运送的主干道, 在黄龙病的病理机制、 病程发展中占据重要地位, 能够在疾病的早期阶段提供特异信息, 有助于疾病的早期诊断。 为了探索以树皮为样本建立黄龙病近红外检测技术的可行性, 分析不同采样部位对黄龙病近红外预测模型的影响, 设计了树叶、 树皮和综合（树叶+树皮）三种采样方案。 通过与标准正态分布法（standard normal distribution, SNV）、 多元散射校正法（multiple scattering correction method, MSC）、 一阶导数法（first derivative）和二阶导数法（second derivative）对比, 发现归一化法（normalization）对树皮光谱数据的处理效果最好。 分别采用偏最小二乘回归法（partial least squares regression, PLSR）和主成分回归法（principal component regression method, PCR）建立柑橘黄龙病预测模型, 发现预测集均方根误差（root mean square error of prediction, RMSEP）都在10-5量级, 并且树叶预测集均方根误差最小（RMSEP of leaves, 1.690 9×10-5）, 树皮均方根误差其次（RMSEP of barks, 1.889 0×10-5）, 综合均方根误差（RMSEP of composite samples, 2.567 6×10-5）最大; 预测集决定系数（the determination coefficient, r2）都在0.9以上, 并且树叶样本所建模型的决定系数最小（the determination coefficient of leaves, r2L, 0.939 6）, 树皮其次（the determination coefficient of barks; r2B, 0.941 5）, 综合样本所建模型的决定系数最大（the determination coefficient of composite samples; r2C, 0.960 3）, 说明三种采样方案所建立的模型都有很好的精度和预测能力, 以树叶为样本所得模型精度虽然最高, 但预测能力最弱, 而综合采样方案所得模型预测能力虽然最强, 但模型精度最低, 只有以树皮为样本所得模型的精度（RMSEPB=1.889 0×10-5）、 预测能力（r2B=0.941 5）都能保持在良好水平。 通过对比分析树叶、 树皮的原始光谱、 模型效果, 探讨了以树皮为样本建立柑橘黄龙病近红外快速检测技术的可行性, 为近红外光谱技术在黄龙病诊断方面的应用提供新的思路。

The near-infrared models have been applied in huanglongbing detection and it has been proved to be feasible, but the present studies are limited to taking leaves as samples. The phloem of bark is a channel to transport pathogens and nutriment, it has been shown to play an important role in pathological initiation, progression and maintenance, so that we can detect huanglongbing in the early stages with the specific information of barks. In order to explore the feasibility of infrared spectroscopy based on the bark samples and analyze the influence of sampling position variability on near infrared models for huanglongbing, three kinds of sampling plan were designed in this paper: navel orange leaves, navel orange barks and composite samples (navel orange leaves and navel orange barks). Then, we established the prediction model of HLB (huanglongbing) with PLSR (partial least square regression) and PCR (principal component regression), when the normalization was turned out to be the optimal data preprocessing method. We found that the RMSEP (root mean squared error of prediction) are all at the level of 10-5: RMSEPL (RMSEP of leaves, 1.690 9×10-5)