Novel Features for EMG Pattern Recognition Based on Higher Order Crossings

In this work, we present a novel set of higher order time domain features for surface electromyographic (EMG) pattern recognition. The proposed methods employ simple measures of frequency information extracted from EMG time series when a sequence of differencing filters is applied. Multiple EMG datasets consisting of 48 able-bodied and transradial amputee subjects performing a large variety of hand and fingers movements are used to evaluate the performance and robustness of the proposed features. The results show that these novel higher order-based features provide significantly better performance than their traditional counterparts by 3–15 % $(p < 0.05)$. The best proposed feature, higher-order myopulse percentage rate, also significantly outperformed other frequency information-based EMG features in the time and frequency domains: histogram, mean frequency, and median frequency, by 8-14%, 8-25%, and 14-35% $(p < 0.05)$, respectively. With relatively less computational complexity, the proposed features could potentially be used as new features for extracting frequency information for EMG- based pattern recognition systems.