Detecting Canine Mastication: A Wearable Approach

Abstract

Mastication is considered a coping mechanism in dogs, therefore, providing chew toys as an enrichment technique may be particularly important in stressful environments, such as the kennel. However, the relationship between chewing and welfare in kennel-housed dogs has not been systematically examined. The purpose of this study was to develop a sensor to quantify chewing with the intention that this technology could be used to understand the relationship between chewing and welfare in kennel-housed dogs. We show that a collar-based microphone can sense canine bites on a Nylabone chew toy. Four human raters annotated bites on audio samples collected from twelve dogs with five minutes of continuous access to the chew toy. A high degree of reliability was found with an average intraclass correlation coefficient (ICC) of 0.994. Using consensus labeling for training, we created bite detection algorithms using random forest, logistic regression, and convolutional neural network (CNN) based techniques. The CNN-based system achieved the highest performance recognition with an accuracy of 88% with a 91% F1 score. This technology will allow us to analyze a larger data set to uncover relationships between chewing styles and stress, cognition, and other characteristics associated with dog welfare.