Análise de imagens e biometria na identificação de maternidade de ovos de codorna

Abstract

Quail are animal models for many fields of life sciences, as well as an important species for meat and egg production worldwide. Egg production, both as food or as for breeding purposes is often based on multiple-hen cages, hindering individual identification for control of production and in-breeding programs. The aim of the present study was to test algorithms of statistical learning and housing schemes for quail that optimize individual laying control based on quail egg external features. 90 birds were used, with a minimum of ten eggs each, four statistical learning algorithms with cross-validation were tested, as well as verifying the influence of number of quail per cage and methods to assign the birds to each cage. Model with better performance consist in the use of ten variables per egg: weight, height, width, eggshell ratio of patterned area, hue, saturation, lightness, intensity of red, green, and blue of egg background color. The classification accuracy increases when cages have less quail (maximum of three birds) and aimed to increase inside-cage variance. The present method shows feasibility for real-world data and with possibility of improvements with new features and more advanced methods.