“赤霞珠”葡萄叶片缺磷胁迫的VIS/NIR光谱快速无损诊断方法  

Rapid and Nndestructive Dagnosis Mthod for Posphate Dficiency in"Cabernet Sauvignon"Gape Laves by Vis/NIR Sectroscopy

作　　者：白雪冰 宋昌泽 张倩玮 代斌秀 靳国杰[1,2] 刘文政 陶永胜 BAI Xue-bing;SONG Chang-ze;ZHANG Qian-wei;DAI Bin-xiu;JIN Guo-jie;LIU Wen-zheng;TAO Yong-sheng(College of Enology,Northwest A&F University,Yangling 712100,China;Ningxia Helan Mountain's East Foothill Wine Experiment and Demonstration Station of Northwest A&F University,Yongning 750104,China)

机构地区：[1]西北农林科技大学葡萄酒学院,陕西杨凌712100 [2]西北农林科技大学宁夏贺兰山东麓葡萄酒试验示范站,宁夏永宁750104

出　　处：《光谱学与光谱分析》2023年第12期3719-3725,共7页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金青年科学基金项目(32202213);陕西省自然科学基金项目(2022JQ-222);新疆维吾尔自治区重点研发计划项目(2020B01005)资助。

摘　　要：研究旨在明确“赤霞珠(Cabernet Sauvignon, Vitis Vinifera L.)”葡萄健康叶片和缺磷胁迫不同时期下的光谱信号特征变化,构建基于光谱技术的“赤霞珠”葡萄叶片缺磷胁迫快速无损诊断模型,为葡萄园病害防治与管理提供理论参考和技术支持。以酿酒葡萄“赤霞珠”葡萄叶片为研究对象,分别采集了正常、缺磷胁迫初期和末期葡萄叶的VIS/NIR反射率信息。对比Savitzky-Golay卷积平滑(S-G Smoothing)、移动平均平滑(MAS)、标准正态变换(SNV)和多元散射校正(MSC)4种预处理及组合方法对于去除葡萄叶光谱信号中随机噪声的能力,确定最佳预处理方法。采用连续投影法(SPA)筛选与“赤霞珠”葡萄叶缺磷胁迫相关的光谱特征变量,分别构建基于线性核函数(Linear)、多项式核函数(Poly)、径向基核函数(RBF)和二层神经网络核函数(Sigmoid)的支持向量机(SVM)模型,以灵敏度(SEN)和准确率(CCR)为依据评估模型诊断性能,形成基于VIS/NIR光谱的“赤霞珠”葡萄叶片缺磷胁迫快速无损诊断方法。S-G Smoothing预处理后的光谱信号的信噪比为110.58,以其为校正集构建的缺磷胁迫诊断模型最佳,因此确定其为最佳的预处理方法。采用主成分分析(PCA)计算样本光谱贡献率,以95%置信空间为依据检测数据集中的异常样本,最终发现并剔除了22的离群点。通过SPA筛选出402.6、 404.6、 409、 411.5、 539.4、 691.9、 729.9、 838.7、 1 011.9、 1 017.5和1 020.5 nm等11个反映“赤霞珠”葡萄叶缺磷胁迫的光谱特征波段,作为缺磷胁迫快速无损诊断模型的输入变量。通过对比分析上述4种核函数SVM的诊断结果,以Linear为核函数构建的“赤霞珠”葡萄叶缺磷胁迫诊断模型能力最佳,对正常叶片诊断的SEN为81.08%, CCR为100%;对缺磷胁迫早期叶片诊断的SEN为100%, CCR为84.78%;对缺磷胁迫末期叶片诊断的SEN为100%, CCR为100%。该研究建立了基于VIS/NIR光谱�The study aimed to clarify the VIS/NIR spectral characteristics of Cabernet Sauvignon leaves with phosphorus deficiency,then to construct a rapid and nondestructive diagnosis model,which is expected to help the vineyard management and disease control.Firstly,the grape leaves in healthy,early and later stress by phosphate deficiency were analyzed by VIS/NIR micro fiber spectrometer.In order to remove noise interference,four preprocessing methods,including Savitzky-Golay convolution smoothing(S-G Smoothing),moving average smoothing(MAS),standard normal variate(SNV)and multiple scattering corrections(MSC),were used to optimize spectral signals.Then,the successive projections algorithm(SPA)was used to select the feature wavebands of leaf phosphate deficiency.Finally,the support vector machine models were constructed based on four different kernel functions,including linear kernel function(Linear),polynomial kernel function(Poly),radial basis function(RBF)and Sigmoid tanh function(Sigmoid),to diagnose the phosphate deficiency of leaves.The sensitivity(SEN)and accuracy(CCR)were cited to assess the availability and effectiveness of those models.Experimental results proved that S-G Smoothing was the best preprocessing method because of the better signal-to-noise ratio of spectrum processed by it and the good availability of the model based on it.Principal component analysis(PCA)was used to find outliers with a confidence interval of 95%.22 samples were identified with outliers and removed.Eleven wavebands(402.6,404.6,409.0,411.5,539.4,691.9,729.9,838.7,1011.9,1017.5 and 1020.5 nm)were selected by SPA to consider as reflecting the information of phosphate deficiency and be the input variables of the diagnosis model.After the contrast of four models with different kernel functions,it can be known that the SVM model with Linear showed better sensitivity and accuracy than others.Its SEN was 81.08%,and CCR was 100%for healthy leaves,its SEN was 100%,and CCR was 84.78%for early-stage diseased leaves,and its SEN and CCR were 1o

关 键 词：酿酒葡萄 缺磷胁迫 可见光/近红外光谱 无损诊断 支持向量机 

分 类 号：O433.4[机械工程—光学工程]