Decomposition Of Surface Electromyography Signal Based On ANN And SVM

Electromyography (EMG) signal is one that is generated by the temporal and spatial summation of motor unit action potential trains (MUAPTs) propagating along muscle fiber on the detection electrode when a lot of motor units are excited. It is the electrical manifestation of neuromuscular activation associated with a contracting muscle. The decomposition of the EMG signal is the procedure by which the EMG signal can be seperated into its constituent MUAPTs. The information obtained from EMG decomposition, such as the MUAP characteristics and the statistic of motor unit firing and recruitment, can be used to study control mechanisms of neuromuscular system and assist clinical diagnosis of neuromuscular diseases. Therefore, there are important theoretical significance and application value for the research of EMG decomposition.Compared with indwelling electromyography (iEMG), surface electromyography (sEMG) can be used for many more fields as its non-invasive characteristic. However, sEMG is much more difficult to decompose due to relative low signal-to-noise ratio, similarity in the MUAP waveform of different motor units and the significant superimposition of MUAPs. Nowadays, most of the studys are restricted to contraction levels of <10% maximum voluntary contraction (MVC) and rarely are concerned with the decomposition of superposed waveform.In this paper, the main work is to study on preprocessing techniques for noise removal and pattern recognition algorithms for sEMG decomposition. The main research work and results are as follows:(1)A method which is named three-level filtering technology based empirical mode decomposition is proposed for sEMG preprocessing. In the method, three filtering algorithms are adopted according to the noise characteristics of sEMG, that is: spectrum interpolation for power line noise, morphological filter for the baseline drift and empirical mode decomposition for white noise. The experimental results demonstrate that the proposed three-level filtering technique can not only improve the SNR of sEMG but also effectively reserve the main features of MUAP's waveform. This will facilitate the identification of the MUAP and sequentially to improve the accuracy of sEMG decomposition.(2)The research on sEMG signal decomposition algorithm. On the basis of extensive investigation and deep analysis about EMG decomposition algorithms, a new automatic decomposition method is proposed. First, the Self-Organizing Feature Map (SOFM) neural network and the Learning Vector Quantization (LVQ) network are combined together to accomplish the MUAP waveform cluster analysis. Then support vector machine is used to classify MUAP waveform. Finally the superposition is decomposed based on the recursive template alignment technique .Experimental results show that high accuracies of the decomposition can be achieved using proposed method in this paper, especially for signals recorded at lower or moderate level of contraction force.(3)Building a platform of decomposition system. A two-channel sEMG decomposition system is set up under the development platform of MATLAB. It not only includes advanced algorithms as mentioned above, but also provides a convenient user interface for manually operating to improve the decomposition accuracy.