Sexagem de Aves Utilizando Espectroscopia no Infravermelho por Transformada de Fourier e Machine Learning

Abstract

Early identification of sex in birds plays a crucial role in reproduction, breeding, commercialization, and scientific knowledge production. DNA-based molecular techniques have been the primary approach for sex determination, distinguished by their accuracy and risk-free nature for the bird, but they face challenges related to time consumption, costs, and the need for specialized laboratories. This study investigates the application of Fourier Transform Infrared Spectroscopy (FTIR) combined with machine learning algorithms as a non-invasive and precise technique for early sex determination in birds, through measurements made in a feather region known as the vexil. The research focuses on four species: Oryzoborus maximiliani (Bicudo), Nymphicus hollandicus (Cockatiel), Oryzoborus angolensis (Curió), and Psittacula krameri (Ring Neck). The methodology involved collecting feather FTIR spectra from young birds (sex confirmed by DNA analysis) using the Attenuated Total Reflectance (ATR) accessory. Spectra obtained in the range of 3800 to 800 cm⁻¹ were processed using Standard Normal Variate (SNV) and subjected to Principal Component Analysis (PCA) to reduce dimensionality and identify significant variables. The combination of FTIR with the Random Forest classifier demonstrated promising results. Significant accuracies were obtained in the identification of bird sex, with emphasis on Bicudo and Curió (94.4%), Cockatiel (77,8%) and Ring Neck (72,2%). In this way, the study shows great potential for application in avian sexing and indicates the need for additional studies to allow its generalized application in certain species.