Rat Grooming Behavior Detection with Two-stream Convolutional Networks

Rat grooming behavior can be used to reflect its states of physiology and psychology. Behavioral studies in rats are often based on human observations involving the viewing of segments of long video recordings. In the case of grooming, the number of subjectively identified grooming movements is manually counted, typically over long video sessions lasting for days. Therefore, an intelligent approach is needed to help analyze such datasets automatically with high precision. Here, we develop a grooming detection method using deep learning algorithms that combine a Convolutional Neural Network (ConvNets) and a Long Sort-Term Memory network (LSTM). Experimental results demonstrate that the proposed method produces a satisfactory and higher detection rate for grooming behavior of rats.