Online myoelectric pattern recognition based on hybrid spatial features

• Published in: Biomedical signal processing and control Vol. 66; p. 102482
• Main Authors: Jaber, Hanadi Abbas, Rashid, Mofeed Turky, Fortuna, Luigi
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.04.2021
• Subjects: Classifier; Gesture recognition; HD-sEMG; Inter-session evaluation; Myoelectric pattern recognition; Real time classification; Spatial features extraction; SVM; HD-sEMG; Gesture recognition; Real time classification; Spatial features extraction; SVM; Classifier; Inter-session evaluation; Myoelectric pattern recognition; Map
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2021.102482

•Three types of spatial feature sets are proposed named as H, HI and AIH features. H features correspond to extracting HOG features by Histogram Oriented Gradient method from HD-sEMG map. HI features are obtained by concatenating H features and scalar intensity feature. Finally, The hybrid spatial features AIH are obtained by combining H features and the intensity features matrix (AI). AI features are proposed as a modification of I feature.•The integration of HD-sEMG electrodes technology and reliable spatial features can guarantee an efficient classification performance.•The feature sets are evaluated in both intra-session and inter-session evaluation. The online classification reports the robustness of these features to overcome the variability of EMG signals across time and between sessions with accurate classification performance.•Reducing the sampling rate to a certain extent without degrading the classifier’s performance shows the reliability of the proposed features. Although Electromyography (EMG) signals are sources of neural information that are essential in controlling the prosthetic hand, many confounding factors caused the variation of EMG signals properties over time. These factors degraded the performance of myoelectric prosthesis and made it unstable over time, across subjects and sessions such as stress, fatigue, muscular dystrophy, shifting electrodes locations, etc. The spatial information of muscle activity can be augmented using high-density surface electromyography (HD-sEMG) electrodes technology. In this study, HD-sEMG electrodes technology are integrated with the robust hybrid spatial features to improve the performance of myoelectric prostheses towards the non-stationary characteristics of EMG signals over time and across sessions. Three types of spatial feature sets are proposed using histogram oriented gradient (HOG) algorithm and intensity features. Three sub databases are used for evaluating the SVM classifier based on the proposed features. Intra-session and inter-session evaluation in offline manner show the potential of the proposed feature sets to improve the classification performance. The classification performance based on hybrid spatial features achieved precision of 97.9 %, sensitivity of 97.5 % for intra-session evaluation and a classification accuracy about 92.18 % for inter-session evaluation. Online classification results exhibit the robustness of hybrid spatial features (i.e. it achieved a classification accuracy based on hybrid spatial features of 92 % for intra-session evaluation and 89.9 % between sessions). Further, reducing the sampling rate to a certain extent without affecting the classification accuracy indicates the robustness and reliability of the proposed features. The results confirm that the robust spatial features have a significant effect on the classification accuracy more than that of the classifier algorithm.