Preliminary Research Of Mouse Behavior Recognition Based On Foot Contact Features

Experiments on the behaviors of mice and other laboratory animals are essential components of aerospace medicine.Environmental changes may lead to variations in the physiological states of laboratory animals such as mice,which in turn affect animal behaviors.By detecting small changes in animal behaviors,key behavioral characteristics of the animals can be identified,providing an important reference for aerospace medical diagnostic research.However,experiments involving laboratory animals in aerospace medicine are featured with complicated procedures,long cycle,and large amount of data analysis.Besides,the recording of animal behaviors through automatic behavior recognition software is easily affected by light,covering,and other factors,resulting in low accuracy of mouse behavior recognition.The paper constructed a mouse behavior recognition system based on foot contact features,aiming to achieve intelligent monitoring of mouse behaviors and make mouse behavior recognition more accurate and efficient.The system consists of four parts,including the information collection module for mouse behaviors,the video processing module,the feature extraction module,and the behavior classification module.According to experimental findings,the system effectively raised the accuracy of mouse behavior recognition and was expected to provide valuable references for the studies in aerospace medicine.The information collection module for mouse behaviors was designed based on the frustrated reflection principle to collect behavioral samples containing a mouse’s foot contact features under natural illumination and non-illumination,respectively.The video processing module was designed to accurately abstract a mouse’s footprint images and body images from collected videos.The feature extraction module was used to abstract a mouse’s foot contact features and other features accurately.The behavior classification module included improved particle swarm optimization and optimized support vector machine.Specifically,improved particle swarm optimization optimized the parameters of support vector machine fast and efficiently,raising the ability of support vector machine in classifying samples.In addition,three data sets of mouse behaviors were first generated based on the mouse behavior recognition system.Next,data sets of edited videos were labeled,and corresponding feature sets were provided,making it convenient for scholars to use in further research.The paper experimented with mouse behavior recognition based on the data sets of edited videos,designed several methods and experiments for feature abstraction,and analyzed the impacts of different features and integration mechanisms on the constructed behavior recognition system.According to experimental findings,the constructed mouse behavior recognition system identified mouse behaviors effectively.