Noninvasive in ovo sexing in Korat chicken by pattern recognition of its embryologic vasculature

Abstract

SUMMARY: Identifying chick's sex as early and as accurately as possible is a vital task in chicken farming. The prominent methods are feather, color, and vent sexing, in 1-day old chicks, but these tasks require a skilled examiner. Alternatively, gonad inspection is also adopted, but it is more invasive and usually involves anesthetic and surgical procedures. To resolve their shortcomings, in ovo sex determination was recently proposed. Some techniques relied on analyzing biochemical contents (hormone) or spectroscopic patterns (feather pigment) in a developing embryo. The others need to precisely make a small hole on an eggshell, while keeping the inner membrane intact. Furthermore, advanced and costly equipment involved has impeded wider adoption in most smaller farm settings. To address both issues, this paper proposes a noninvasive in ovo sexing technique that focuses on the phenotypic differences. Particularly, chorioallantoic maturation and vascular anastomosis at days 10 and 12 of incubation were characterized and hence discerned between sexes. The appearance of embryologic vasculature was represented by the entropy of its multiband visual features, extracted from a candled egg. These features were then used to identify their sexes by a supervised classifier. The experiments showed that, as early as day 12, the method could predict embryo's sex from its candled image with accuracy, precision, recall, and balanced accuracy of 83.33, 84.62, 73.33, and 81.90%, respectively. Moreover, false discovery and false omission rates, which reflect practical errors, were 15.38 and 17.39%, respectively.

Keywords

• angiogenesis
• computer vision
• machine learning
• phenotypic differences
• poultry farming