Estimation of Nitrogen Content on Apple Tree Canopy through Red-Edge Parameters from Fractional-Order Differential Operators using Hyperspectral Reflectance

Timely, nondestructive and effective determination of canopy nitrogen content provides an important reference value for real-time monitoring of total nitrogen status of apple trees. Different processing methods are used to mine hyperspectral information to estimate nitrogen content. However, the overprocessing or underprocessing of hyperspectral data leads to the underutilization of spectral information. The primary objective of this study was to establish a model for estimating the nitrogen content of apple tree canopy by red-edge parameters based on fractional differential. The Grunwald–Letnikov fractional difference algorithm was used to extract the red-edge parameters from the hyperspectral canopy data, so as to develop the support vector machine (SVM) and random forest (RF) models. The results showed that the correlation with nitrogen content can be enhanced by differential spectroscopy compared with the original spectrum. The spectral parameters such as red-edge peak area (Sr(α)) obtained by fractional differential and logarithmic transformation processing and the correlation coefficient with nitrogen content can reach 0.6 or greater. The R2 of SVM and RF models constructed with red-edge parameters reached 0.56 (RMSE was 1.51 for SVM) and 0.94 (RMSE was 0.84 for RF), respectively. The RPD greater than 2 indicates that both models could be used for nitrogen estimation, and the RF model has a better predictive effect (RPD was 2.17 for SVM, RPD was 2.43 for RF). It provides an effective method for real-time monitoring of apple canopy nitrogen status and provides theoretical and technical support for hyperspectral information mining and data processing.