An Improved Feature Extraction Method for Surface Electromyography Based on Muscle Activity Regions

In the analysis of surface electromyography signals(sEMG), the extraction of suitable features is one of the key factors affecting pattern recognition. The aim of this paper is to propose an improved sEMG feature extraction algorithm based on muscle activity regions. The fusion of muscle activity intensity on the basis of muscle activity regions compensates for the low accuracy of the original features for the recognition of similar movements. In this paper, the sEMG signals of five leg movements were collected, including two similar movements: upstairs and downstairs, standing and sitting. The classification performance of the features before and after the improvement was tested with six classifiers. It proves that the new characteristic, active muscle position and intensity (AMPI), greatly improves the classification accuracy of similar movements. The paper also compares the new features with the traditional eight classification features. The results show that the new features are at the forefront of the classification performance, with a very small difference in classification accuracy of 4.1% compared to the best performing features. This confirms the high practical value of the new features. New features are still based on the mapping relationship between movement patterns and active muscle regions. This provides new ideas for the feature extraction method of sEMG signals. In addition, compared with the traditional features, the new feature still have the ability to reduce the dimension, which provides a more applicable feature extraction method for the application of multi-channel electromyogram(EMG) signals acquisition devices and high-density electrodes.