NON-CONTACT SENSING OF HOG WEIGHTS BY MACHINE VISION

Abstract

Non-contact weighing of animals can, as compared to the conventional direct weighing, minimize the stress onboth the animal and operator, reduce price dockage at sale, and enhance health monitoring. A literature review on the studiesof direct and indirect animal weighing was conducted, and a machine vision system was developed in this study to measurehog weights without physical contact. This method used image processing techniques to extract the characteristic featuresof the animal and correlate them to liveweight. A total of 187 hogs were imaged from the top, and physical and morphologicalfeatures including rear area and center width were extracted from the obtained images using Matrox imaging software. Resultsindicated that the best correlation was found to exist between the projected area and the hog weight (r = 0.96); a goodcorrelation also existed with the width (r = 0.95). Some features, such as compactness, elongation, and roughness, hadnegligible correlation to hog weight. Mathematical models were developed to estimate the hog weight based on the mostoutstanding feature (i.e., the rear area) identified through image processing. The suitability of the mathematical models forweight approximation was examined through regression analyses.