基于高光谱成像技术的番茄叶片含水量检测研究  

Detection of water content in tomato leaves based on hyperspectral imaging technology

作　　者：马玲[1] 李亚娇 张祎洋 王静[1] 马燕[1] 马思艳 吴龙国[1,2] MA Ling;LI Yajiao;ZHANG Yiyang;WANG Jing;MA Yan;MA Siyan;WU Longguo(College of Enology and Horticulture,Ningxia University,Yinchuan 750021,China;Ningxia Modern Protected Horticulture Engineering Technology Research Center,Yinchuan 750021,China)

机构地区：[1]宁夏大学葡萄酒与园艺学院,宁夏银川750021 [2]宁夏现代设施园艺工程技术研究中心,宁夏银川750021

出　　处：《南京农业大学学报》2024年第6期1221-1229,共9页Journal of Nanjing Agricultural University

基　　金：宁夏高等学校自然科学项目(NYG-2024-019);宁夏重点研发计划项目(2021BEB04077);国家重点研发计划项目子课题(2021YFD1600302-3)。

摘　　要：[目的]为快速检测叶片含水量,本研究探索及时监测番茄植株生长状况的在线监测模型。[方法]利用高光谱成像技术,提取195个叶片样本的平均光谱反射率。通过异常值剔除、样本集划分、5种预处理方法对原始光谱进行预处理和优化,采用连续投影算法(successive projections algorithm,SPA)、无信息变量消除变换法(uninformation variable elimination,UVE)、迭代保留信息变量法(iterative retained information variable,IRIV)和遗传偏最小二乘算法(genetic partial-least-squares algorithm,GAPLS)提取特征波长,并建立偏最小二乘回归(partial-least-squares regression,PLSR)模型。基于优选的特征波长,建立PLSR、多元线性回归(multiple linear regression,MLR)以及主成分回归(principal component regression,PCR)模型和卷积神经网络模型(convolutional neural network,CNN)。[结果]优选基线校准-正交信号校正法(baseline-orthogonal signal correction,Baseline-OSC)对叶片含水量进行预处理;IRIV法提取的特征波长建立的叶片含水量定量预测模型效果最优,R_(c)^(2)为0.489,R_(p)^(2)为0.466;基于IRIV-CNN建立的叶片含水量模型效果好(R^(2)c=0.668,RMSEC=0.019;R_(p)^(2)=0.424,RMSEP=0.033)。[结论]利用高光谱成像技术结合Baseline-OSC-IRIV-CNN模型预测番茄叶片含水量是可行的。[Objectives]In order to quickly detect leaf water content,an online monitoring model was explored to monitor the growth of tomato plants in a timely manner.[Methods]In this experiment,the average spectral reflectance of 195 leaf samples was extracted using hyperspectral imaging.The raw spectra were preprocessed and optimised by outlier removal,sample set division,five preprocessing methods,successive projections algorithm(SPA),uninformation variable elimination transformation(UVE),iterative retained information variable(IRIV)and genetic partial-least-squares algorithm(GAPLS)were used to extract the feature wavelengths,and the partial-least-squares regression(PLSR)model was developed.PLSR,multiple linear regression(MLR),principal component regression(PCR)and convolutional neural network(CNN)models were built based on the preferred feature wavelengths.[Results]The results showed that the Baseline-orthogonal signal correction(Baseline-OSC)method was preferred for pre-processing of leaf water content;the quantitative leaf water content prediction model established by the feature wavelengths extracted from the IRIV method had an optimal R_(c)^(2) of 0.489 and R_(p)^(2) of 0.466;the leaf water content model based on IRIV-PCR was good(R^(2)c=0.668,RMSEC=0.019;R_(p)^(2)=0.424,RMSEP=0.033).[Conclusions]It is feasible to predict tomato leaf water content using hyperspectral imaging combined with Baseline-OSC-IRIV-CNN model.

关 键 词：番茄 叶片 含水量 高光谱 检测 

分 类 号：S641.2[农业科学—蔬菜学]