Computer Vision for Digital Phenotype Measures In Livestock

The United Nations Food and Agriculture Organization (FAO) Global Plan for Action for Animal Genetic Resources (AnGR), affirmed that accurate, consistent phenotypic data is integral to understanding and effectively conserving AnGR. They published guidelines for phenotypic characterization for “systematically documenting” phenotype, and to promote “the use of standard practices and formats for describing (animal) characteristics” so credible comparisons between livestock populations are possible. However, the published guidelines do not provide standardized or exact data. The purpose of this research is to integrate digital image analysis for a reliable enumeration and capture of traditional livestock (goat) phenotypic data (size, weight, shape, horns, coat color and texture) over varied situations and locations for more meaningful research comparisons. Digital images of 77 goats were acquired under highly controlled conditions, and physical measures (height, length, front and rear width, barrel, etc.) were recorded for comparing digital image analysis results. Software was developed to isolate the animal to extract digital measures and for calibration to real-world measure. Accuracy of prediction varied with the trait considered, but most had a coefficient of determination of 70-80% in the initial predictions.