Real-time Animal Behavior Recognition And Quantitative Analysis Based On Convolutional Neural Network

Traditionally,ethology has been defined purely as a branch of biology,but now it is deeply integrated with other disciplines to use its research tools to solve behavioral problems.Among them,machine learning and machine vision have made a series of achievements in the processing and analysis of animal behavior pictures,especially the convolutional neural network can achieve the behavior classification with high accuracy.To some extent,this enables the automation and quantification of behavior capture and analysis.However,most of the current technical achievements focus on video analysis of animal behavior after experiments.There is no solution about how to combine other non-video ways to realize real-time recognition and stimulate animal according to its results,which is what we want to discuss in this paper.We design a universal behavioral analysis device for mice,including recognition procedures.After testing,the program we built can meet the requirements of high judgment accuracy and fast recognition,and integrate and collect other kinds of information.It can realize closed-loop control:behavior capture,species determination,animal stimulation and behavior capture.Moreover,we used this experimental platform to study the behavioral response of the hungry mice to the video stimulation of the simulated sky predator in the upper field when they were feeding.The resulting data revealed behavioral differences between the experimental group and the control group.Together,we demonstrate the reliability of the device and its potential for other ethological experiments.