The poor effect of current practical gesture recognition methods based on surface electromyo-graphic signals,a signal feature sample construction method and attention residual network(ARN)were proposed to identify 23 types of functional grasp gestures in NinaproDB1.In the course of sample con-struction,the surface electromyography signal was low-pass filtered by the Butterworth filter to remove noise interference and meanwhile retain the signals of pass band and transition band.The filtering signal was truncated to generate surface electromyography by using the time window.The mean absolute value,variance and wavelength features of signal were simultaneously calculated and fused to generate the fea-ture electromyogram.After feeding the feature electromyogram samples into ARN,the accuracy of ges-ture recognition on the test set reached 86.87%,proving the effectiveness in the constructed sample of sig-nal feature combined with ARN to recognize practical gestures.
维普
万方数据
