Comparison of image feature extraction for classification of swine thermal comfort behavior

Abstract Interactive, behavior-based environmental control for swine production has inherent advantages over the conventional, temperature-based control methods. Correctly distinguishing the thermal comfort behavior of swine is critical for the success of interactive environmental control and appropriate feature selections are the basis of the correct classification. This paper compares four feature selection methods using postural images of young pigs subjected to a range of environmental conditions. Programmable cameras were used to capture the behavioral pictures which were then processed into binary images. Fourier coefficients (8×8); moments (first eight); perimeter and area; and combination of perimeter, area and moments of the binary behavioral images were used as the input patterns to a neural network. With the respective feature selection, the neural net correctly classified 96, 92, 96 and 99% training images and 78, 73, 86 and 90% testing images. Thus, the combination of perimeter, area and moments of the binary images as neural network features gave the best performance in the behavioral classification.