Combined Influence of Classifiers, Window Lengths and Number of Channels on EMG Pattern Recognition for Upper Limb Movement Classification

Electromyogram pattern recognition (EMG-PR) is a very important method for upper limb movement classification. However, there are many factors such as classifiers, window lengths and number of channels which can make an influence on EMG-PR efficiency. Previous studies examined the effects of three different factors on EMG-PR separately. However, the combinations of three different factors (classifiers, window lengths and number of channels) may also affect the classification accuracy of EMG-PR. In present study, we discussed the effects of combinations of three different factors including classifiers, window lengths and number of channels on EMG-PR. We analyzed the different combinations of three factors. Four healthy subjects participated in this study, and they played five motions of hand and wrist in this experiment. We found that these three factors had a significant effect on EMG-PRe The performance of linear discriminant analysis (LDA) of EMG-PR outperformed the performance of back propagation neural network (BPNN) (p < 10−3). The classification accuracy of LDA is higher than support vector machine (SVM) (p < 10−3). In addition, 200 ms window length had enough data to classify the different motions. Furthermore, we also found that five channels has a significant increase when compared to three channels (p < 0.05). The proposed method can increase the performance of EMG-PR.