Dados hiperspectrais e aprendizado de máquina na avaliação do crescimento e qualidade de mudas de eucalipto

Abstract

ABSTRACT: The combination of machine learning with hyperspectral data can represent a significant advancement in obtaining morphological and quality variables of forest seedlings, providing an innovative and efficient approach that surpasses the limitations of traditional methods. Therefore, the objective of this study was to evaluate the effectiveness of machine learning algorithms combined with hyperspectral data in predicting morphological variables and the quality of Eucalyptus deglupta Blume seedlings. In 90 seedlings, the following were measured: spectral variables (350 to 2500 nm), using a hyperspectral spectroradiometer sensor; and morphological variables such as height, diameter, dry masses, and relationships between these variables. To estimate the morphological variables, machine learning algorithms were trained: artificial neural networks (ANN), decision tree (DT), linear regression (LR), M5P algorithm, and random forest (RF). Two input combinations were evaluated in the training: the entire spectral range provided by the sensor (WL) and grouped into 24 spectral bands (SB). A cross-validation procedure was adopted, with k-fold equal to 10. To assess the performance of the tested prediction algorithms, metrics such as correlation coefficient, mean absolute error, and root mean square error were used. The combination of machine learning algorithms with hyperspectral data proved to be efficient in predicting morphological variables and the quality of Eucalyptus deglupta seedlings. The performance of machine learning algorithms depends on the input used and vice-versa. The SVM algorithm, using WL as input variables, was the most efficient in predicting the variables.