Apparatus and methods for mouse behavior recognition on foot contact features

Abstract

Abstract Behavior recognition of model animals such as mice, rats, and monkeys, is of great significance in medical research and vital in the development in medicines or drugs. Studies of the animal’s responses to specific stimulations consisted of visual imaging methods that was easily affected by cage lighting conditions and shooting angle. In this study, we designed a behavior classification system that focused on object-floor contact features extracted from touch images of the frustrated total internal reflection injected with infrared lights. A mouse behavioral dataset (IMBD) was then established, and a support vector machine was applied to conduct this classification. To improve its performance, an improved particle swarm optimization was proposed for the optimization of parameters. Based on IMBD and the support vector machine, we tested different feature extraction methods and classifiers. The system performance of the current study showed more accuracy and efficacy compared with performance of other prevalent particle swarm optimizations. Our study resulted in a recognition rate of up to 94.37% for individual behavior. The average rate of behavioral recognition for all tested mice reached 83.09%. The results suggested that foot contact features of mice were more effective than regular video features in behavior recognition.