Application of machine vision to swine environmental control

Summary form only given. Use of computer vision to enhance animal well-being and overall production efficiency is expected to be characteristic of the future electronic stockmanship. This study explores the feasibility of the technique with growing pigs and identifies the areas that need further investigation. Specifically, nursery pigs were subjected, in groups of 10 pigs, to cold, comfortable, and warm environments. Postural behaviors of the pigs were captured at 40-minute intervals with programmable cameras overseeing the pig pen area. The raw behavioral images were processed by thresholding, edge detection, and morphological filtering to separate the pigs from their background. Feature extraction of Fourier coefficients, moments, perimeter and area, and combination of perimeter, area and moments was performed on the processed behavioral images. The selected features were used as the inputs to a 3-layer neural network which then classifies the pig behavior into cold, comfortable, or warm category.