基于化学计量学方法的冬小麦长势光谱信息提取及监测研究

发布时间：2018-04-29 18:04

本文选题：冬小麦 + 长势　；参考：《山西农业大学》2016年博士论文

【摘要】：冬小麦是我国的第三大粮食作物,实时、准确、快速掌握冬小麦长势状况对于施肥、灌溉等田间管理和实现产量与品质的调控具有重要作用。基于高光谱遥感技术已经广泛的应用于作物研究。作物光谱信息提取不准确、监测模型精度和稳健度不高是限制光谱遥感技术进一步应用的主要原因。本研究以不同年份、不同冬小麦品种、不同氮肥水平的田间小区试验为基础,综合利用光谱技术、多元统计分析和化学计量学方法,基于高光谱遥感技术对冬小麦长势监测展开了研究,得出以下主要结论：1本研究所获取的冬小麦长势数据符合冬小麦一般生长规律,且差异性较大,能够表征不同生长条件下的冬小麦长势状况,本文基于因子分析构建的综合长势指标(Comprehensive growth indicators, CGI)能够表征冬小麦长势状况;2归一化植被指数(Normalized difference vegetation index, NDVI)与叶面积指数(Leaf area index, LAI)、地上生物量(Above ground biomass, AGB)存在“饱和”现象,研究发现该现象是由构成NDVI的波段先发生“饱和”造成的,证明发生“光谱饱和”现象的阈值为：LAI2.5或AGB 1 kg m-2。同时发现某些植被指数在高LAI和AGB条件下可以克服“饱和”现象,通过分析这些植被指数的构成总结出克服“光谱饱和”现象的方法：(1)构成该植被指数的波段应尽可能多的集中在“红边”区域；(2)该植被指数的取值范围应不受限制;(3)准确挖掘光谱特征信息,并利用多元统计和化学计量学方法构建稳健性的监测模型;3平滑处理中的9点平滑处理(SM9)、变化处理中的平方根(T4)处理和校正处理中的噪音校正(Noise)为最佳的光谱数据预处理方法；4光谱波段：400、512、536、555、680、700、735、760、816、890、920、1130、2040和2430 nm与冬小麦长势密切相关;5非线性模型决策支持机(Support vector machine, SVM)要优于偏最小二乘(Partial least square regression, PLSR)线性模型,主成分回归(Principle component regression, PCR)线性模型最差；在各化学计量学模型中基于CGI的表现最好,表明利用CGI表征冬小麦长势是有意义的；研究证实基于SVM非线性模型具有一定的稳健性和普适性,但是在监测冬小麦综合长势指标时,基于T4处理的PLSR模型(Rc2=0.768,RPDc=1.973;Rv2=0.724, RPDv=1.693)要优于SVM模型(Rc2=0.813, RPDC=1.945; Rv2=0.715, RPDV=1.554);基于特征波段建立的冬小麦长势指标的PLSR-SMLR模型具有一定的应用潜力;除PCR模型和PLR-SMLR模型中的PWC的原始光谱(SM0)模型表现较好外,所有基于原始光谱的冬小麦长势指标的监测模型表现最差,表明光谱数据经预处理后可以提高冬小麦长势指标监测模型的表现。其中,原始光谱的平方根处理(T4)的适用性要优于其它预处理方法6利用冬小麦高光谱技术监测冬小麦长势的光谱处理流程为：光谱数据的平方根处理→冬小麦长势指标和预处理光谱的相关性分析→结合PLSR和SMLR提取冬小麦长势的光谱特征信息,并利用相关性分析结果和多元统计分析结果进行多方面、多角度验证→利用PLSR-SMLR、PLSR线性方法或SVM方法构建冬小麦长势的预测模型→综合对比分析,选择预测精度高、表现稳定的模型。
[Abstract]:Winter wheat is the third largest grain crop in China. Real-time, accurate, fast mastery of winter wheat growth condition plays an important role in fertilization, irrigation, field management and control of yield and quality. Based on hyperspectral remote sensing technology, it has been widely used in crop research. Low health is the main reason for limiting the further application of spectral remote sensing technology. Based on the field plot experiments of different winter wheat varieties and different nitrogen fertilizer levels, this study is based on the comprehensive utilization of spectral, multivariate statistical analysis and chemometrics methods, based on hyperspectral remote sensing technology to study the monitoring of winter wheat growth. The main conclusions are as follows: 1 the winter wheat growth data obtained by the research institute are in accordance with the general growth pattern of winter wheat, and the difference is large, which can characterize the growth condition of Winter Wheat under different growth conditions. The comprehensive growth index (Comprehensive growth indicators, CGI) based on factor analysis can characterize the long winter wheat length. The 2 normalized vegetation index (Normalized difference vegetation index, NDVI) and the leaf area index (Leaf area index, LAI). The aboveground biomass (Above ground biomass) has a "saturation" phenomenon. It is found that the phenomenon is first "saturated", which proves that the phenomenon of "spectral saturation" occurs. The threshold is: LAI2.5 or AGB 1 kg m-2. also found that some vegetation indices can overcome the "saturation" phenomenon under the conditions of high LAI and AGB. By analyzing the composition of these vegetation indices, the method of overcoming "spectral saturation" is summed up: (1) the band of the vegetation index should be concentrated in the red edge area as much as possible; (2) The range of vegetation index should be unrestricted; (3) accurate mining of spectral feature information, and using multivariate statistical and chemometrics methods to build a robust monitoring model; 9 point smoothing treatment (SM9) in 3 smoothing treatment, noise correction (Noise) in the square root (T4) treatment and correction processing (Noise) as the best spectral data preconditioning The 4 spectral bands, 40051253655568070073576081689092011302040 and 2430 nm, are closely related to the winter wheat growth, and the 5 nonlinear model decision support machine (Support vector machine, SVM) is superior to the linear model of partial least squares (Partial least square regression, PLSR), principal component regression (Principle) The linear model of gression, PCR) is the worst, and the performance of CGI based on the chemometrics model is the best, indicating that it is meaningful to use CGI to characterize the winter wheat growth. The research confirms that the SVM nonlinear model has certain robustness and universality, but the PLSR model based on T4 treatment (Rc2=0.76) is based on the monitoring of the comprehensive growth index of winter wheat. 8, RPDc=1.973, Rv2=0.724, RPDv=1.693) are superior to the SVM model (Rc2=0.813, RPDC=1.945; Rv2=0.715, RPDV=1.554), and the PLSR-SMLR model based on the characteristics of the winter wheat growth index based on characteristic bands has a certain application potential; the original spectrum of the PWC is better than the PCR model and PLR-SMLR model, all based on the original spectrum. The monitoring model of winter wheat growth index is the worst, which indicates that the spectral data can improve the performance of the monitoring model of winter wheat growth index after pretreatment. The applicability of the square root treatment (T4) of the original spectrum is better than that of other pretreatment methods 6 using the winter wheat high light spectrum technology to monitor the winter wheat growth. The square root processing of the spectral data, the correlation analysis of the winter wheat growth index and the preprocessing spectrum, and the extraction of the spectral characteristics of the winter wheat growth with PLSR and SMLR, and using the correlation analysis results and multivariate statistical analysis results to carry out multiple aspects, multi angle verification, PLSR-SMLR, PLSR linear method or SVM method construction Prediction model of winter wheat growth > comprehensive comparative analysis, choose a model with high prediction accuracy and stable performance.

【学位授予单位】：山西农业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：S512.11

【参考文献】