IDENTIFICAÇÃO DE MARCADORES EM ESPÉCIES DE MYRTACEAE POR MEIO DE ESPECTROSCOPIA, ANÁLISE DE IMAGENS E QUIMIOMETRIA

Abstract

Because of the importance of Chemometrics and the difficulty in identifying species of Myrtaceae, a family with several fructiferous species, with high ornamental and economic potential, the present study aimed to classify five species of Myrtaceae by chemometric tools and machine learning. The photosynthetic pigment contents of species of the Myrtaceae family were estimated by images, and modeling by the algorithms: Partial Least Squares (PLS), Random Forest (RF), Artificial Neural Network (ANN), and Support Vector Machine (SVM). The species analyzed in this study were: Campomanesia guazumifolia (Cambess.) O. Berg, Campomanesia xanthocarpa (Mart.) O. Berg, Eugenia involucrata DC., Eugenia uniflora L. e Psidium cattleyanum Sabine, belonging to the Alluvial Araucaria Forest (FOM) fragment, located at the UNICENTRO CEDETEG campi, Guarapuava, PR. The fluorescence analyses of chlorophyll a by induction of rapid or transient kinetics (OJIP curve) occurred in situ. The leaves of the five species of Myrtaceae were collected and their photosynthetic pigments were extracted and analyzed by UV-VIS spectrophotometry. The images of the extracts of each sample were obtained using a smartphone and the grayscales were analyzed using the ChemoStat® software, and RGB color histogram. The contents of photosynthetic pigments were modeled using the Weka 3.9 software, and regression models based on the PLS and RF algorithms. Species classifications were also performed using the PLS, RF, ANN, and SVM. Chlorophyll a (Chl a) content ranged between 0.99 - 1.38 μg/mL, chlorophyll b (Chl b) between 0.34 - 0.67 μg/mL, total chlorophyll between 1.38 - 2.11 μg/mL and carotenoids between 0.14 - 0.21 μg/mL, for the all species analysed. The highest content of Chl a, Chl b and total chlorophyll were observed in the species E. involucrata and the lowest content in the species E. uniflora. The contents of carotenoids were higher in E. uniflora and lower in C. guazumifolia. The validation of the models has always occurred through Cross- Validation and external validation. In Cross Validation, the hyperparameters of each model were adjusted and it was always observed that the RF algorithm was better than the PLS algorithm for modeling the pigment contents of the Myrtaceae species and their RGB histograms. The classification by spectrophotometric scanning was evaluated by the ANN and SVM algorithms, which always presented the best results for this type of modeling. E. uniflora showed the highest fluorescence emission by the OJIP curve. The RF impurity analysis selected 15 fluorescence parameters that allowed the classification of the species evaluated. The methods used for modeling were efficient to estimate the concentrations of photosynthetic pigments, by RGB color histograms and spectroscopic scanning by UV-VIS. The chemometric tools, as well as the machine learning tools: PLS, RF, ANN, and SVM were important to classify Myrtaceae species, evaluated in this research. The fluorescence chlorophyll-a parameters and the contents of photosynthetic pigments present in leaves of the species evaluated can be used as markers for taxon-chemometric differentiation.