A Study on the Estimation Model of Hyperspectral Reflectivity and Leaf Nitrogen Content of Cotton Leaves

Abstract

Modern agriculture requires more accurate field management capability compared with traditional agriculture. The development of hyperspectral remote sensing technology embodies rapid and non-destructive features in agricultural information monitoring, providing a technical guarantee for the scientific management of agricultural production. The mathematical model of inverse cotton leaf total nitrogen was established by decomposing and transforming the original cotton leaf spectrum using continuous wavelet analysis and traditional spectral transformation, taking the characteristic wavelet coefficients and spectral characteristic bands as independent variables, and using univariate, stepwise regression, and partial least squares methods. The correlation between the total nitrogen content of cotton leaves and the spectral reflectance, through different methods of spectral treatment, was improved to different degrees. For the conventional spectral transformation, the inverse logarithmic first-order differential lg $^{\prime} $ (1/R) improved the correlation of cotton leaf total nitrogen by 0.26. The continuous wavelet analysis outperformed the conventional spectral model regarding information noise reduction and mining of feature information. The established model with RPD>2 had good stability and predictive power for all sample data.

Keywords

• Hyperspectral
• non destructive testing
• continuous wavelet analysis
• spectral transformation
• total nitrogen
• cotton