基于柔性传感器的鱼类摆尾行为识别量化OA

Traditional non-contact methods for monitoring fish behavior,including acoustic and optical approaches,have limitations in scenarios involving dense targets,obstructions,or concealment.To address this,this study proposes a fish behavior analysis and monitoring method utilizing a wearable flexible sensor.Based on fish behavioral characteristics and signal occurrence frequencies,a flexible strain sensor was designed and developed.It integrates thermoplastic polyurethane resistive strain-sensitive units,non-invasive adhesion,wireless signal transmission,and flexible circuitry.Visual synchronous acquisition was employed to analyze the time-frequency domain features of tail-flapping strain signals under different movement speeds.Data preprocessing methods,including smoothing and baseline correction,were investigated and determined.A data sampling strategy with a window size of 4 and a stride of 1 was adopted.To extract effective features,16 typical waveform signal characteristics,such as frequency center and waveform factor,were initially selected.To extract effective features,16 typical waveform characteristics,including frequency center and waveform factor,were initially selected.Based on random forest feature importance analysis,six core features were ultimately retained.Six analysis methods,including linear regression,random forest,XGBoost,LSTM,LightGBM,and SVM,were compared.Experimental results demonstrate that the wearable flexible strain sensor can effectively collect swimming data signals from crucian carp.Among the regression models evaluated,the random forest regression model demonstrated optimal performance,achieving a test set root mean square error(RMSE)of 0.0039 m/s and a coefficient of determination(R2)of 0.998,with excellent generalization stability(ΔRMSE=0.0010).The proposed method effectively overcomes the limitations of non-contact approaches,providing robust technical support for fish behavior research.