李恒凯, 王英浩. 甜柚叶绿素含量高光谱无损检测模型[J]. 华南农业大学学报, 2019, 40(2): 126-132. DOI: 10.7671/j.issn.1001-411X.201806001
    引用本文: 李恒凯, 王英浩. 甜柚叶绿素含量高光谱无损检测模型[J]. 华南农业大学学报, 2019, 40(2): 126-132. DOI: 10.7671/j.issn.1001-411X.201806001
    LI Hengkai, WANG Yinghao. Hyperspectral nondestructive detection model of chlorophyll content of Citrus maxima[J]. Journal of South China Agricultural University, 2019, 40(2): 126-132. DOI: 10.7671/j.issn.1001-411X.201806001
    Citation: LI Hengkai, WANG Yinghao. Hyperspectral nondestructive detection model of chlorophyll content of Citrus maxima[J]. Journal of South China Agricultural University, 2019, 40(2): 126-132. DOI: 10.7671/j.issn.1001-411X.201806001

    甜柚叶绿素含量高光谱无损检测模型

    Hyperspectral nondestructive detection model of chlorophyll content of Citrus maxima

    • 摘要:
      目的  为监测甜柚Citrus maxima果树生长健康状况及预测甜柚产量,以赣州南康地区一片甜柚果园为研究对象,建立甜柚叶片叶绿素含量检测模型。
      方法  使用Field Spec4便携式地物光谱仪和SPAD-502叶绿素仪测定甜柚叶片光谱及SPAD值,分别采用单变量回归、逐步回归及偏最小二乘法(PLS)构建其叶绿素含量高光谱无损检测模型并进行精度检验。
      结果  原始光谱在553 nm处、一阶光谱在692和752 nm处的反射率与叶绿素含量相关性最高,这3个波段为甜柚叶片光谱反射率敏感波段;当主成分个数为4时,PLS具有最高的精度,且基于PLS技术所建立的模型较单变量、逐步回归模型精度更好,模型拟合度较高,其决定系数(r2)最高,为0.869,均方根误差(RMSE)和相对误差(RE)最小,分别为3.013和6.82%。对原始光谱、一阶导数光谱及PLS拟合的估测模型进行对比分析显示,PLS模型无论是从建模样本精度还是模型预测能力方面均优于前2种传统模型。
      结论  PLS模型适合于利用高光谱数据进行叶绿素含量的估测,可作为甜柚叶绿素含量的最佳无损检测模型。

       

      Abstract:
      Objective  To monitor the health status of fruit trees and predict the yield of sweet pomelo (Citrus maxima), a detection model of chlorophyll content in C. maxima leaves was established in a C. maxima orchard of Nankang, Ganzhou.
      Method  The leaf spectrum and SPAD value of C. maxima were measured using Field Spec4 portable earth spectrometer and SPAD-502 chlorophyll meter. The hyperspectral nondestructive detection model of chlorophyll content was constructed by single variable regression, stepwise regression and partial least squares (PLS) method, and the accuracy was examined.
      Result  The reflectances of original spectrum at 553 nm and the first order spectra at 692 and 752 nm had the highest correlation with chlorophyll content. These three bands were sensitive bands of spectral reflectance of C. maxima leaves. When the number of principal components was four, PLS had the highest level of precision. PLS model had higher accuracy, fitting degree and determination coefficient (r2=0.869) compared with the single variable and stepwise regression models, and PLS model had the lowest root mean square error (RMSE) being 3.013 and the lowest relative error (RE) being 6.82%. Comparing and analyzing the estimation models of original spectrum, first derivative spectrum and PLS fitting, PLS model was superior to the two traditional models in terms of sample precision and prediction ability.
      Conclusion  PLS model is suitable for the estimation of chlorophyll content by hyperspectral data, and the best nondestructive detection model for the chlorophyll content of C. maxima.

       

    /

    返回文章
    返回