基于可见与近红外光谱联用的柑桔黄龙病快速无损检测研究

黄龙病是柑桔果树的毁灭性病害，对柑桔产业危害巨大。基于模型平均理论，探讨联用可见与近红外光谱技术，提高柑桔黄龙病快速无损检测精度的可行性。采集记录柑桔叶片的可见与近红外光谱，经实时荧光定量PCR鉴别黄龙病叶片为轻度、中度和重度三类，缺素和正常样品也经PCR鉴定，共五类叶片。基于光谱直接拼接、光谱归一化拼接和模型平均三种不同策略，结合偏最小二乘判别分析(PLS-DA)和多元线性回归(MLR)方法，分别建立了柑桔黄龙病可见与近红外光谱联用无损检测模型。经比较发现，光谱联用模型的检测精度均高于可见或近红外单一检测模型，且经导数处理后的光谱直接拼接PLS-DA模型检测精度最高，模型预测相关系数为0.97，预测均方根误差为0.67，模型总误判率为3%，其原因是导数消除了光谱的基线漂移。光谱归一化拼接的PLS-DA模型检测精度次之，模型总误判率为7%。可见与近红外模型平均的检测精度最低，模型总误判率为7.2%。实验结果表明，联用可见与近红外光谱，结合光谱拼接方法，提高了柑桔黄龙病无损检测模型的检测精度，研究可为其他领域的光谱联用提供参考依据。

Study on the Quick Non-Destructive Detection of Citrus Huanglongbing Based on the Spectrometry of VIS and NIR

Huanglongbing (HLB) is a devastating disease of citrus fruit trees, which is very harmful to citrus industry. Based on the theory of model averaging, the feasibility of improving the accuracy of quick and non-destructive detection of citrus HLB was studied using the visible and near infrared spectrum technique. The visible and near infrared spectra of citrus leaves were collected and record, and three kinds of leaves with slight HLB, moderate HLB, serious HLB was identied using real time fluorescent quantitative PCR and nutrient deficiency and normal was identified by PCR also, in all five types. Three kinds of different strategies, including the spectra directly stitched, the normalized spectral stitched and model averaging were as basis, combined with partial least squares discriminant analysis (PLS-DA) and multiple linear regression (MLR) method, and the non-destructive detection Spectrometry model of citrus HLB were developed using visible and near infrared spectroscopy respectively. By comparison, it can be found that the detection accuracy of Spectrometry model was higher than visible or near infrared spectroscopy single detection model while the detection accuracy of spectra directly stitched PLS-DA model was highest after derivative preprocessing. The correlation coefficient (R_P) of the model was 0.97, the root mean square error (RMSEP) 0.67, and the total midjudgement rate 3%. The reason was to eliminate baseline drift of spectra. The detection accuracy of spectra directly stitched PLS-DA model after normalized was second, and the total midjudgement rate was 7%. The detection accuracy of visible and near infrared average model was lowest, and the total midjudgement rate was 7.2%. The experimental resultsshowed that using visible and near infrared spectroscopy combined with spectra stitched method can improved the detection accuracy of non-destructive detection model of Citrus HLB, and the research can provide a reference for other areas of the Spectrometry.