基于色彩参数和高光谱特征的针叶树种色素含量预估

Prediction of Conifer Pigment Content Based on Color Parameters and Hyperspectral Characteristics

光谱成像技术广泛应用于植物理化参数无损伤测定等领域研究,而色素与色彩参数相关性研究也有学者探索。但比较并优选分别以色彩参数值、光谱参数值作为自变量与色素含量拟合出的模型,还未见报道。本实验以5种针叶树种为研究对象,筛选蓝边幅值Db、黄边幅值Dy、红边幅值Dr、绿峰幅值Rg、红谷幅值Rr、蓝边面积SDb、黄边面积SDy、红边面积SDr、比值植被指数RVI、差值植被指数DVI、归一化植被指数NDVI 11种光谱植被参数作为该光谱分析的基础,将实测针叶色彩参数值、光谱参数值分别作为自变量,采用多元线性逐步回归方法(SMLR)预估色素含量建立模型,以R^(2)、RMSE为评价标准,对比选出模型精度最高的参数组合应用于实践。研究结果表明:(1)树种间针叶色素含量、色相参数值、光谱反射率均存在一定差异(p<0.05)。(2)树种间针叶光谱反射率红松显著低于北美短叶松、樟子松、赤松(p<0.05),针叶树种原始光谱在可见光波段500和680 nm附近呈现“蓝谷现象”和“红谷现象”,在550和760 nm波段附近呈现“绿峰现象”和“红边现象”;一阶微分光谱反射率在700 nm附近产生剧烈变化。(3)色素含量与色彩参数、光谱反射率、光谱特征参数存在显著线性关系。(4)花青素和叶绿素分别以L,a*和L,a*,b*,S色彩参数组合为自变量时,拟合模型R^(2)最高,分别为0.588和0.638;而类胡萝卜素、叶绿素a、叶绿素b都是以FD_(652),FD_(700),SDb,SDy,RVI,DVI和NDVI光谱参数组合为自变量时,拟合模型R^(2)最高,分别为0.779,0.786,0.774。该研究运用高光谱相机、色彩色差仪、紫外-可见分光光度仪实现了快速预估针叶色素含量,在色彩参数值与光谱值都与色素含量存在显著相关性的基础上,成功选出建立模型精度最高的参数组合,在针叶树种色素预估时可以根据精度需求及研究条件选择不同方法和参数值。

Spectral imaging technology is widely used in the field of non-invasive determination of physical and chemical parameters of plants,and scholars have also explored the correlation between pigments and color parameters.However,it has not been reported that the models fitted using color parameter values and spectral parameter values as independent variables and pigment content,respectively,are compared and optimized.In this experiment,five conifer species were used as research objects,and 11 spectral vegetation parameters,including blue edge amplitude Db,yellow edge amplitude Dy,red edge amplitude Dr,green peak amplitude Rg,red valley amplitude Rr,blue edge area SDb,yellow edge area SDy,red edge area SDr,ratio vegetation index RVI,difference vegetation index DVI,and normalized vegetation index NDVI,were screened as the basis of spectral analysis in this paper.The measured conifer color parameter values and spectral parameter values were used as independent variables,respectively.Stepwise multiple linear regression(SMLR)was used to estimate the pigment content to establish a model,with R^(2) and RMSE as evaluation criteria,and the parameter combinations with the highest model accuracy were compared and selected for practice.The results of the study indicate that:(1)There are differences in leaf pigment content,color phase parameter values,and reflectance spectral between tree species(p<0.05).(2)The leaf spectral reflectance of Pinus koraiensis Sieb.et Zucc.was significantly lower in Pinus sylvestris var.mongolicaLitv.,Pinus banksiana Lamb and Pinus densifloraSieb.et Zucc.(p<0.05).The original spectrum of conifer species shows“blue valley phenomenon”and“red valley phenomenon”near 500 and 680 nm in the visible band,and“green peak phenomenon”and“red edge phenomenon”near 550 and 760 nm bands;the first-order differential spectral reflectance produces dramatic changes near 700 nm.(3)Pigment content was significantly correlated with color parameters,spectral reflectance,and spectral characteristic parameters,and there was a significant linear relationship.(4)When anthocyanins and chlorophyll were combined with L,a*and L,a*,b*,and S color parameters as independent variables,respectively,the fitted model R^(2) was the highest,0.588 and 0.638,respectively.In contrast,carotenoids,chlorophyll a,and chlorophyll b were all combined with FD652,FD700,SDb,SDy,RVI,DVI,and NDVI spectral parameters as independent variables.The fitted model R^(2) was the highest,0.779,0.786,and 0.774,respectively.In this study,a hyperspectral camera,color difference instrument and UV-Vis spectrophotometer were used to realize rapid prediction of needle pigment content.Based on a significant correlation between color parameter value and spectral value and pigment content,the parameter combination with the highest accuracy of the established model was successfully selected.Different methods and parameter values could be selected according to the accuracy requirements and research conditions in predicting of needle pigment.