Study On Human-Machine Interface Based On Hybrid A-Mode Ultrasound And Surface Electromyogram Signals

Traditional human-machine interfaces(HMI)often adopt single-mode sensing,which is difficult to break through their inherent limitations,seriously affecting the clinical application of prosthetic devices.In order to further improve the performance of HMIs,multi-source sensing has gradually become a research trend.Ultrasound and electromyography(EMG)both are the mainstream sensing signals in HMI field.Their sensing fusion can exert the complementary advantages of muscle shape information and electrophysiological information,and break through the performance bottleneck.Therefore,this paper proposes a novel HMI based on the hybrid A-mode ultrasound and surface EMG signal(HUE-HMI),and develops the corresponding software and hardware system to explore its performance in gesture recognition and finger angle prediction.Firstly,in view of the lack of equipment that can simultaneously acquire ultrasound and EMG signals in the same muscle position,this paper completed the hardware system development of the HUE-HMI,including sensor selection,wearing device structure design and hardware circuit design.Based on these,this paper achieves the portable wearing of the sensor and the miniaturization and integration of the hardware circuit system.Then,for the lack of a unified operation interface for processing ultrasound and EMG signals,this paper completed the software design of the HUE-HMI,including embedded program design and human-computer interaction interface design.This paper achieves the orderly acquisition and stable transmission of signals,and provides a more convenient signal acquisition and experimental platform.In addition,compared with the commercial equipment,this paper verifies the quality of the signals collected by HUE-HMI.In terms of signal-to-noise ratio,the ultrasound(32.31dB)and EMG(55.02dB)of HUE-HMI have better performance.Finally,in order to explore the potential of hybrid ultrasound and EMG features in practical applications,this paper carried out gesture decoding experiments and finger angle prediction experiments.The results of gesture decoding experiment show that the accuracy of hybrid features(91.6%)is significantly(p<0.001)higher than that of ultrasound(82.2%)and EMG(72.1%)based on SVM classifier.Further analysis finds that ultrasound and EMG have complementary advantages in resting and non-rest gesture recognition.The results of finger angle prediction experiment show that,based on neural network regression analysis,the fit goodness,root mean square error and correlation coefficient(0.699,0.176,0.830)of the hybrid feature prediction results are superior to ultrasound and EMG.In addition,both ultrasound and EMG features have their own advantages on predictive performance of different bending moments and different fingers.The above experimental results prove that the ultrasound and EMG features have good complementarity,and the hybrid feature can stably and effectively improve the gesture recognition rate and finger angle prediction performance.