A novel feature extraction method for SEMG signals using image processing

Several methods have been used for the feature extraction of surface electromyography (sEMG) signals, such as the integrated EMG (EEMG) in time domain and wavelet transform method in time and frequency domain. Because the EMG signals contain some inherent noise, the features extracted by these methods contain some incorrect values inevitably. According to the human vision, the image processing can be used to obtain the shape of sEMG signals. This paper focuses on the feature extraction of sEMG signals by calculating the geometric feature of sEMG signal using the pixel count method (PCM) in the image and calculating the textural feature using angular second moment (ASM) of gray level co-occurrence matrix (GLCM). The raw sEMG signals were recorded from three healthy subjects' biceps muscles. At the same time, the gray-scale image can be generated from the raw EMG signals. And then the proposed methods are used to calculate sEMG signals features from the gray-scale images. In the experiment, we classified the upper-limb voluntary movement into three motions and these motions are recognized by Back-propagation neural network (BPNN). For comparison, IEMG and wavelet packet transform (WPT) are also used to extract the sEMG features for motion recognition. The experimental results show that the proposed method is superior to the IEMG and WPT method.