Estimation of chlorophyll content of Cinnamomum camphora leaves based on hyperspectral and fractional order differentiation

Abstract

ABSTRACT Hyperspectral remote sensing combined with data preprocessing techniques and machine learning algorithms is a new way to efficiently estimate plant SPAD. In this study, the raw hyperspectral reflectance data of the dwarf forest of Cinnamomum camphora (Linn.) Presl (C. camphora) were subjected to 0-2nd order differential transformations (a step size of 0.25) using SPAD during the harvesting period of C. camphora. The spectral indices with the highest correlation with SPAD in each order were also selected as input variables for the model, and the measured values of SPAD from simultaneous observations were used as the output variable. Support vector machine (SVM), random forest (RF), and backpropagation neural network (BPNN) were used to construct the SPAD inversion model. The results showed that the correlation between the spectral index and the SPAD of the dwarf forest of C. camphora and the accuracy of the constructed prediction model tended to increase and then decrease with increasing differential order. After a comprehensive comparison, the RF-based SPAD prediction model achieved the highest accuracy with a coefficient of determination (R2) of 0.824, root mean square error (RMSE) of 1.619, and mean relative error (MRE) of 3.207% for the validation set, using spectral indices computed by the 1.5-order differential treatment as the input variables. This study provides a theoretical basis for improving the accuracy of plant leaf SPAD inversion based on hyperspectral remote sensing.