利用叶片高光谱反射率预测棉花叶绿素含量

Predicting the content of chlorophyll in cotton using hyperspectral reflectance of leaves

为提高棉花叶绿素含量预测的准确性,利用连续小波分析和传统光谱变换对棉花叶片原始光谱进行分解和变换,以特征小波系数和光谱特征波段为自变量,并利用单变量、逐步回归和偏最小二乘法建立反演棉花叶片叶绿素含量的数学模型。结果显示,不同的光谱处理方法使得棉花叶片叶绿素和光谱反射率的相关性都有不同程度的提升,对于传统光谱变换,倒数对数一阶微分lg(1/R′)对棉花叶片叶绿素相关性提高了0.41。结果表明,连续小波分析在信息降噪和挖掘特征信息方面优于传统光谱模型,建立的模型RPD>2,具有很好的稳定性,对样本数据都具很好的预测能力。

With the development of hyperspectral remote sensing technology,hyperspectral prediction of crop growth can provide scientific management for agricultural production,which can improve crop yields and quality while avoiding excessive use of nitrogenous fertilizers.A mathematical model to invert the content of chlorophyll in cotton leaves was developed using continuous wavelet analysis and conventional spectral transformation to decompose and transform the raw leaf spectra of cotton.The characteristic wavelet coefficients and spectral characteristic bands were used as independent variables.Methods including univariate,stepwise regression and partial least squares were used.The results showed that different spectral treatments improved the correlation between the content of chlorophyll and spectral reflectance of cotton leaves.For the conventional spectral transformation,the inverse logarithmic first order differential lg(1/R′) improved the chlorophyll correlation of cotton leaves by 0.41.It is indicated that the continuous wavelet analysis is superior to traditional spectral models in terms of information noise reduction and mining of feature information.The model established has good stability with RPD2 and good prediction ability for data sampled.