Research On High Density SEMG Feature Classification Of Upper Limbs

The development of human-machine interface technology has been a noteworthy research in the medical rehabilitation field.Two-dimensional matrix style high density surface EMG(HD-sEMG)signals acquisition technology can measure the muscle electrical activity that the limited skin area,capture the entire muscle activity area EMG signals in the time and spatial domain distribution information,which is beneficial to the EMG signals analysis and optimization of complicated motions.Because it’s difficult to analyze the EMG signals of the human upper arm motion,in this thesis,a method that analyze the human-machine interface of HD-sEMG signals for upper arm movement is proposed.The subject upper arm’s flexion and extension continuous motion are divided into forward flexion and horizontal flexion eight different motion degrees,which is simulate to the upper arm continuous motions.Firstly,64 channels high-density sEMG signals acquisition electrode is used to record HDsEMG signals.In the data preprocessing,three spatial filtering algorithms that principle component analysis,fast independent component analysis and multiclass common spatial pattern are compared and analyzed in detail based on the structure of myoelectric control interface technology,HD-sEMG signals are respectively preprocessed by spatial filtering algorithm to obtain separation matrix.It is used to reconstruct the original signal data to reduce the dimension of data processing.A channel selection method for maximizing the mutual information between different types of channels is proposed based on the multiclass common spatial pattern.The channels are arranged in descending order by the mutual information,and the original channel with the best muscle source signal is selected to analyze,the original 64 channels are substitute with a less number of channels.The purpose of separating the signal channel with muscular source strong from the original signal channels is achieved.Secondly,in the feature extraction,the pre-processed HD-sEMG signals are analyzed respectively in the time,frequency and time-frequency domains.It is concluded that timefrequency domain features decomposed by wavelet packet transform can effectively reflect the upper arm motion HD-sEMG signals features.The spatial characteristics of HD-sEMG signals are extracted by maximizing the mutual information channel selection algorithm,which is combined with time-frequency domain characteristics to better reflect HD-sEMG signals characteristics.Finally,artificial neural network,linear discriminant analysis,k-nearest neighbor analysis and support vector machine pattern classifiers are respectively adopted to analyze the characteristics combine spatial and time-frequency domain in different spatial filtering data.When five original signal channels are selected,the pattern recognition results show that support vector machine classifier based on the combination time-frequency domain and space domain features can achieve the expected recognition accuracy of 95%.In addition,the recognition robustness is analyzed in the case of electrode shift,it is concluded that this method can good to electrode shift,and it provide a powerful solution to the human-machine interface system analysis technology for HD-sEMG signals.