Comparative study on estimation models of chlorophyll content in Phyllostachys pubescens leaves

Objective In order to realize hyperspectral inversion of leaf chlorophyll content of Phyllostachys pubesculus, the theoretical basis for growth state analysis and scientific management of P. pubesculus forest was provided.

Method In this study, based on the average reflectance data of leaves at different leaf positions and the measured average chlorophyll content data of sample plots, the average reflectance spectral curves were preprocessed by combining Savitzky-Golay smoothing, standard normalvariable transformation and first-order differentiation. The correlation coefficient method and successive projection algorithm (SPA) were used to extract the feature wavelength, respectively. Based on the characteristic wavelength results, SPXY algorithm, KS algorithm and random method (RS) were used to partition the data set, and six spectral training sets were constructed. Four machine learning algorithms including random forest (RF), extreme gradient boosting algorithm, support vector machine regression (SVR) and back propagation neural network were selected to construct chlorophyllin content estimation model of moso bamboo, and the most suitable model for chlorophyll content estimation was screened according to the evaluation results of the model.

Result In terms of the performance of data set partitioning algorithm, the SPXY data set partitioning method could significantly improve the estimation accuracy of chlorophyll content compared with KS and RS data partitioning method. Compared with the correlation coefficient method, SPA algorithm can effectively eliminate the influence of collinearity and improve the model accuracy in the feature wavelength screening method. In terms of machine learning model construction, the support vector machine regression model established by SPA algorithm combined with SPXY dataset partitioning method had the highest estimation accuracy for chlorophyll content of moso bamboo leaves, with R² values of 0.78 and 0.76 for training and validation, respectively.

Conclusion The SVR model established by the SPA algorithm combined with the SPXY data set partitioning method could accurately estimate the chlorophyll content of P. pubescens leaves, and could be used for rapid and non-destructive acquisition of chlorophyll content information of P. pubescens leaves.