Study On Behavior Analysis Method Of Multiple Mice Based On Key Point Detection

As one of the typical model animals,rodents are widely used in various behavioral experimental studies because their external manifestations of body language are sufficient to express their individual functions as behavioral states.With the continuous update of animal behavior analysis technology,from traditional human identification to sensor recognition,and now artificial intelligence recognition,unsupervised behavior analysis technology has gradually been widely used in behavioral experimental research.With the help of artificial intelligence-related technologies,this paper designs and realizes the automatic identification and tracking of key points of multiple rodents and the accurate analysis of their related social behaviors in an open field experimental environment.The main work and results of this paper are as follows:(1)Research on key point recognition and tracking algorithms for multiple rodents in the same scene.In response to the limited space in the experimental environment of mice and the occlusion problem caused by behavioral interaction between multiple mice,this paper adopts the Mice Key Points Network(MKPNet)based on a cascaded pyramid network to detect key points in multiple mice.Firstly,the key points are divided into "difficult" and "easy" points,and pooling operations are improved to enhance the detection ability of small targets using non rounding methods.Finally,the overall accuracy of the algorithm is improved by combining the "plug and play" attention mechanism.The experimental results show that the optimized model has improved average accuracy and recall by 1.8 and 0.1 on public datasets.(2)Study of behavioral analysis algorithms for multiple mice.This paper uses Bayesian classifier based on "causation" relationship for behavior analysis.First,rodent behavior is divided into two categories,single and multiple.Standing,crawling,scanning,drinking,eating,and aggression,sniffing,imitation,reproduction,and domain in multiple behaviors are defined respectively.Then,the corresponding behavior model is established to realize the estimation of model parameters.Finally,complete the validation of the model.In the early stage of preparation,a large amount of behavioral data,such as moving speed,moving distance,movement direction,etc.,were obtained,and experimental verification was carried out on public datasets and self-made datasets,and the results showed that the accuracy of the model reached more than 96.5%.(3)A classical behavioral experiment of open field social behavior was designed to explore the correlation between animal behavior and pathology.According to the specific needs of animal behavior experiments,the open field experiment box is designed,from the selection of appearance and hardware,to the acquisition and production of data sets,to the combination of key point detection algorithms and automatic behavior analysis algorithms,and finally the behavior analysis results are associated with "diseases".Through experiments,130,000 frames of image data were obtained,and the accuracy of the key point detection network reached 98% on the self-made dataset,and the average accuracy of the three social behaviors imitation,sniffing and attack was classified by 99.6%,93.75% and 99.05%,respectively.According to the analysis of trajectory plot and frequency map,it was concluded that there was a positive correlation between social behavior and "disease" in mice.