Estimation of Anthocyanins in Apple Leaves Based on Ground Hyperspectral Imaging and Machine Learning Models

Abstract

The anthocyanins in apple leaves can indicate their growth status, and the health of apple leaves not only reveals the nutritional supply of the apple tree but also reflects the quality of the fruit. Therefore, real-time monitoring of anthocyanins in apple leaves can monitor apple growth, thereby promoting the development of the apple industry. This study utilizes ground hyperspectral imaging to estimate anthocyanins in Fuji apple leaves in the Loess Plateau through spectral transformation, feature extraction (including band selection and spectral indices construction), and regression algorithm selection, establishing models for three growth stages. The results indicate: (1) The average anthocyanins in apple leaves decrease from the Final Flowering stage to the Fruit Enlargement stage. The original hyperspectral imaging at wavelengths before 720 nm shows a decrease in reflectance as the growth stages progress, while the spectral curves after 720 nm remain largely consistent across stages; (2) Compared to single original spectral variables, multivariate estimation models using original spectra and second-order derivative transformed spectra show improved accuracy for anthocyanins estimation across different growth stages, with the most significant improvement during the Fruit Enlargement stage; (3) Although the computation of the three-band spectral indices is resource-intensive and time-consuming, it can enhance anthocyanins estimation accuracy; (4) Among all models, the CatBoost model based on original spectra and second-order derivative transformed spectra indices for the entire growth period achieved the highest accuracy, with a validation set R2 of 0.934 and a RPD of 3.888, and produced effective leaf anthocyanins inversion maps. In summary, this study achieves accurate estimation and visualization of anthocyanins in apple leaves across different growth stages, enabling rapid, accurate, and real-time monitoring of apple growth. It provides theoretical guidance and technical support for apple production and fertilization management.