基于柑橘叶、花近红外高光谱信息的营养诊断与成花能力预测研究

发布时间：2018-04-23 08:56

本文选题：甜橙 + 叶片　；参考：《西南大学》2015年硕士论文

【摘要】：本文以15年生枳橙(Citrus. sinensis × Poncirus. trifoliata, "Carrizo Citrange")砧哈姆林甜橙(C.sinensis (L.), "Hamlin Sweet Orange")为对象,研究基于高光谱图像信息的叶、花营养和成花能力的光谱预测方法,旨在为建立甜橙叶片和花器营养水平实时监测与高效合理施肥管理技术提供依据。主要研究结果如下：1.哈姆林甜橙叶片营养的近红外高光谱特征与预测研究(1)哈姆林甜橙植株花芽分化期当年生春梢、夏梢和盛花期上一年春梢叶片450-1000nnm光谱平均反射率曲线呈现规律为：叶片成熟度越高,光谱曲线波动越趋于平稳,即光谱曲线绿峰、红谷、近红外平台区域光谱反射率间差值越小。(2)为消除或减小试验材料本身、仪器和外界环境因素噪声对光谱信息的干扰,本研究选用五点移动平均平滑(SG-平滑)、多元散射校正(MSC)、标准正态变量变换(SNV)、一阶导数(1-Der)和二阶导数(2-Der)等五种方法对甜橙叶片原始光谱(RAW)进行预处理,并在此基础上分析基于PLS模型的氮(N)、磷(P)、钾(K)和可溶性总糖(TC)含量预测精度。结果显示,花芽分化期当年生春梢叶片光谱经过五种方法预处理后对叶片N、P、K和TC含量的PLS模型预测,以SG-平滑算法预处理光谱预测效果优于其他预处理方法；花芽分化期当年生夏梢叶片光谱2-Der预处理后PLS建模预测N、P、K和TC的效果更优；盛花期上一年春梢营养枝叶片光谱数据经过SG-平滑预处理后对N、P、K和TC含量的预测效果优于其他预处理方法。(3)基于哈姆林甜橙叶片的近红外高光谱信息可以较好实现对叶片N、P、K和全C含量的预测。其中,对叶片N含量的预测以花芽分化期当年生春梢叶片近红外光谱SG-平滑算法预处理后的PLS建模预测精度较高,预测精度可达0.735;对P含量的光谱预测以花芽分化期当年生春梢叶片原始光谱PLS建模最好,预测精度达0.733；对K含量的近红外高光谱预测以花芽分化期当年生春梢叶片原始光谱的PLS建模预测较好,预测精度达0.728；对全C含量的光谱预测以采用盛花期上一年春梢叶片光谱,SG-平滑算法预处理后PLS建模预测,对全C含量预测精度达0.688。2.哈姆林甜橙花营养的近红外高光谱特征与预测研究(1)研究建立以0.1最优阈值间隔和0.5～0.6灰度值的甜橙花朵高光谱图像有效光谱信息提取技术,筛选出哈姆林甜橙花的N营养含量特征光谱波段为692.146～735.3nm,以及440.07～481.07nm,692.146～735.3nm波段组合。以该特征波段组合反射光谱的siPLS预测模型,可以较好的预测花的N含量,其预测精度达0.762；通过特征波段组合筛选使得预测模型运算所需实际波点数减少为108个,比全光谱预测减少了85.8%,预测效率较大提高。(2)经过筛选,得到哈姆林甜橙花P含量预测特征光谱波段为573.757～633.437nm,特征波段组合为573.757～633.437nm和693.74～752.653nm;以该特征波段组合反射光谱的siPLS预测模型,可以较好的实现对花的P含量预测,其预测精度达0.885,且实际仅用152个波点数参与模型预测运算,比全光谱预测的波点数减少了80.0%(3)筛选出哈姆林甜橙花K含量预测的特征高光谱波段为593.313～643.711nm,表明以特征波段筛建立的iPLS预测模型,对于花K含量的预测效果最优,其预测精度达到0.916,实际采用64个波点数参与预测运算,比全光谱波点数减少了91.6%。3.叶片碳氮比和树体花量的高光谱预测(1)哈姆林甜橙叶片C/N估测的特征波长为507.117,507.886,523.294,527.928,534.114,543.41nm,iPLS建模预测精度达0.914。(2)秋季上一年春梢叶片对次年植株花量预测的特征波长为544.186,552.726,567.516,572.196,575.319,582.352,588.613和593.313nm,以PLS建模的预测精度达0.893,可用于次年植株花量的预测。
[Abstract]:In this paper, the 15 year old orange orange (Citrus. sinensis x Poncirus. trifoliata, "Carrizo Citrange") anvil (C.sinensis (L.), Hamlin Sweet Orange) was used as the object to study the spectral pretest method based on hyperspectral image information of leaf, flower nutrition and flower forming ability, aiming at establishing the real-time monitoring of the nutrition level of orange leaves and flower organs. The main research results are as follows. The main results are as follows: 1. the near infrared hyperspectral characteristics and prediction of the leaf nutrition of Hamlin sweet orange (1) the annual spring shoot of the flower bud differentiation period of the Hamlin sweet orange plant, the average reflectance curve of the 450-1000nnm spectrum of the leaves of the spring shoot of the summer shoots and the flowering stage is: leaves The higher the maturity of the film, the more stable the spectral curves fluctuate, that is, the difference between the spectral reflectance of the spectral curve, the Red Valley and the near infrared platform is smaller. (2) to eliminate or reduce the experimental material itself, the interference of the instrument and the ambient noise to the spectral information, the research selects five points moving average smoothness (SG- smooth) and multiple scattering correction. Positive (MSC), standard normal variable transformation (SNV), first derivative (1-Der) and two order derivative (2-Der) were used to pretreat the original spectrum of sweet orange leaves (RAW), and on this basis, the prediction accuracy of nitrogen (N), phosphorus (P), potassium (K) and soluble total sugar (TC) based on PLS model was analyzed. The results showed that the spring shoot Ye Pianguang of the flower bud differentiation period was in the spring shoot. After preprocessing the PLS model of the content of N, P, K and TC, the spectral prediction effect of the SG- smoothing algorithm is superior to the other pretreatment methods after five methods. The effect of PLS modeling on the PLS modeling of the leaves of summer shoot leaves in the flower bud differentiation period is better than that of the PLS modeling, and the effect of P, K and TC is better. The prediction effect on the content of N, P, K and TC after SG- smoothing is superior to other pretreatment methods. (3) near infrared hyperspectral information based on Hamlin sweet orange leaves can be used to predict the content of N, P, K and all C in leaves. The prediction of the content of N of leaf blade N is the SG- flat near infrared spectrum of the spring shoot leaves of the flower bud differentiation period. The prediction accuracy of PLS modeling is high and the prediction accuracy can reach 0.735. The prediction of P content by spectral prediction is best and the prediction accuracy is 0.733. Near infrared hyperspectral prediction of K content is predicted by PLS modeling prediction at the flower bud differentiation stage as the original spectrum of the annual spring shoot leaf. The prediction accuracy is 0.728, and the spectrum prediction of the total C content is used for the spring shoot leaf spectrum of the first year of the flowering period, the prediction of PLS modeling after the SG- smoothing algorithm, the near infrared hyperspectral characteristics and prediction of the 0.688.2. Hamlin sweet orange flower nutrition for the total C content prediction accuracy (1), the 0.1 optimal threshold interval and 0.5 ~ 0. are established. 6 gray value of orange flower hyperspectral image effective spectral information extraction technology, the spectral band of N nutrient content of Hamlin sweet orange flower is 692.146 ~ 735.3nm, and 440.07 ~ 481.07nm, 692.146 to 735.3nm band combination. The siPLS prediction model of the combined reflectance spectrum of the characteristic band can be used to predict the N content of the flower better. The prediction accuracy is 0.762, and the number of actual wave points required for the prediction model operation is reduced to 108 through the combination of feature bands. The prediction efficiency is reduced by 85.8% and the prediction efficiency is higher than that of the full spectrum prediction. (2) after screening, the spectral wave segment of the P content prediction feature of Hamlin orange flower is 573.757 to 633.437nm, and the characteristic band combination is 573.757. To 633.437nm and 693.74 ~ 752.653nm, the siPLS prediction model of the combined reflectance spectrum of the characteristic band can be used to predict the P content of the flower better. The prediction accuracy is 0.885, and the actual number of 152 wave points is only involved in the model prediction operation, and the K content of Hamlin sweet orange flower is screened out by 80% (3) than the total spectrum predicted by the total spectrum. The characteristic hyper spectral band of 593.313 ~ 643.711nm shows that the iPLS prediction model established by the feature band screen has the best prediction effect for the flower K content, and its prediction precision is 0.916. The actual use of 64 wave points is involved in the prediction operation, and the high spectral prediction (1) of the carbon and nitrogen ratio of 91.6%.3. leaves and the flower volume of the tree body is reduced than the total spectral wave points. The characteristic wavelength of the Hamlin orange leaf C/N estimation is 507.117507.886523.294527.928534.114543.41nm, and the prediction accuracy of iPLS modeling is 0.914. (2) the characteristic wavelengths of the spring shoot of the spring shoot of the last year in the autumn are 544.186552.726567.516572.196575.319582.352588.613 and 593.313nm, and the predictive precision of PLS modeling is obtained. Up to 0.893, it can be used to predict the plant flower volume of the next year.

【学位授予单位】：西南大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S666.4;S127

【参考文献】