High-density surface electromyography (HD-sEMG) is to record muscles’ electrical activity from a restricted area of the skin by using two dimensional arrays of closely spaced electrodes. This technique allows the analysis and modelling of sEMG signals in both the temporal and spatial domains, leading to new possibilities for studying next-generation muscle-computer interfaces (MCIs). sEMG-based gesture recognition has usually been investigated in an intra-session scenario, and the absence of a standard benchmark database limits the use of HD-sEMG in real-world MCI. To address these problems, we present a benchmark database of HD-sEMG recordings of hand gestures performed by 23 participants, based on an 8 × 16 electrode array, and propose a deep-learning-based domain adaptation framework to enhance sEMG-based inter-session gesture recognition. Experiments on NinaPro, CSL-HDEMG and our CapgMyo dataset validate that our approach outperforms state-of-the-arts methods on intra-session and effectively improved inter-session gesture recognition.

Hand movement classification using Surface Electromyogram (sEMG) is of utmost importance for myoelectric controlled prosthetic hand. The accurate myoelectric movement classification depends on careful experimental design in relation to the configuration of each processing step. As a machine learning algorithm, Support Vector Machine (SVM) has demonstrated outstanding performance in myoelectric pattern recognition. However, there is little effort to comprehensively evaluate the performance of SVM in myoelectric hand movement classification when varying experimental configurations. We therefore aim to carry out a large-scale evaluation so as to provide a guideline for the practice in this area. We evaluated the SVM performance when recognizing three subsets of hand movements extracted from the Non-Invasive Adaptive Hand Prosthetic (NinaPro) database for various configurations of window lengths, feature sets, feature normalization strategies, feature selection algorithms and SVM kernels. Both intra-subject and inter-subject cross-validation accuracies were used as evaluation metrics to simulate different application scenarios. Results showed that recognition performance using features extracted in longer sliding windows was better than features extracted in shorter sliding windows. An optimized feature set containing best single features on NinaPro delivered strong discriminative power in intra-subject cross-validation. However when combined with the Phinyomark feature set it demonstrated the best performance in inter-subject cross-validation. We have also found that the subject-specific feature normalization approach was more effective than the standard feature normalization approach for inter-subject cross-validation. Importantly, the Minimal-Redundancy-Maximal-Relevance (mRMR) proves a better feature selection criterion for myoelectric hand movement classification. Moreover, the best performance occurs when the Optimal Relaxation Factor (ORF) kernel is used for constructing SVM models. We conclude from this study that a careful experimental design is crucial for achieving high performance of myoelectric hand movement classification. Based on our evaluation, we believe that the ORF kernel together with a longer window size can be considered as the best configuration for applying SVM to myoelectric hand gesture recognition. As the best performing feature sets were originally formed based on an evaluation of various sEMG features on NinaPro, we conclude that a feature evaluation process is necessary for myoelectric hand movement classification.