Walk-through weighing of pigs using machine vision and an artificial neural network

Abstract

Machine vision-based weighing of pigs is a non-intrusive, fast and relatively accurate approach that could reduce stress on both the animal and the stockman during the weighing process. An image-based walk-through system was developed in this study for pig liveweight approximation without having to restrain the pig to a certain area for stationary imaging. A protocol was developed to automatically screen and select the images captured for image processing. The artificial neural network technique was used in this study to correlate a multitude of physical features extracted from the walk-through images to pig liveweight in an attempt to improve the accuracy of liveweight approximation. The results showed that the average relative error of the walk-through weighing system was around 3%. The walk-through system has made it even easier for stockmen to obtain the liveweight of pigs using a machine vision-based weighing system.