Deep Cross-User Models Reduce the Training Burden in Myoelectric Control

Abstract

The effort, focus, and time to collect data and train EMG pattern recognition systems is one of the largest barriers to their widespread adoption in commercial applications. In addition to multiple repetitions of motions, including exemplars of confounding factors during the training protocol has been shown to be critical for robust machine learning models. This added training burden is prohibitive for most regular use cases, so cross-user models have been proposed that could leverage inter-repetition variability supplied by other users. Existing cross-user models have not yet achieved performance levels sufficient for commercialization and require users to closely adhere to a training protocol that is impractical without expert guidance. In this work, we extend a previously reported adaptive domain adversarial neural network (ADANN) to a cross-subject framework that requires very little training data from the end-user. We compare its performance to single-repetition within-user training and the previous state-of-the-art cross-subject technique, canonical correlation analysis (CCA). ADANN significantly outperformed CCA for both intact-limb (86.8–96.2%) and amputee (64.1–84.2%) populations. Moreover, the ADANN adaptation computation time was substantially lower than the time otherwise devoted to conducting a full within-subject training protocol. This study shows that cross-user models, enabled by deep-learned adaptations, may be a viable option for improved generalized pattern recognition-based myoelectric control.

Day-to-Day Stability of Wrist EMG for Wearable-Based Hand Gesture Recognition

“…Therefore, CCA limits the amount of data from previous days to only one repetition leading to transformation matrices that don't cover the full range of variability from previous days. In a previous study [27], Campbell et al reported similar limitations of the CCA technique with unsatisfactory performance. In our preliminary work, adding more recalibration repetitions improved the classification accuracy of CCA, corroborating results reported in [52], however, this defeats the intended purposes in this context.…”

Toward Robust, Adaptiveand Reliable Upper-Limb Motion Estimation Using Machine Learning and Deep Learning–A Survey in Myoelectric Control

“…Using a self-calibration strategy, the effectiveness of ADANN was then validated in the presence of confounding factors including inter-session and across-day variations [112]. In another following work, Campbell et al [113] further tested ADANN in the cross-subject classification by requiring minimal training data from an end-user. Different from those efforts, another investigation of feature-based deep TL was presented by Bao et al [114] based on a two-stream CNN with shared weights.…”

“…By contrast, sufficient but unlabelled D T data are available in UTL. For instance, FT approaches [67,[103][104][105][106][107][108] typically belong to STL, whilst the DANNbased approaches [34,112,113] exploit UTL. According to investigations on both conventional TL [96,115] and deep TL [104], the supervised versions usually perform significantly better than unsupervised ones.…”

Multi-day dataset of forearm and wrist electromyogram for hand gesture recognition and biometrics

“…However, extensive investigation has demonstrated sEMG-based HGR has poor cross-user transference performance 9 , suggesting a calibration-free and one-size-fits-all model for all users is still elusive, which suggests that sEMG signals inherently contains individual differences, i.e., biometric information. This has provided motivation for investigating the potential of sEMG as a biometric trait.…”

“…Extensive research on EMG has been performed on gesture recognition with application in rehabilitation using prosthetic and orthotic devices, home application for assisting daily activities, virtual environment control, and sign language recognition 3,4,34 . Recent studies have suggested deep learning techniques for cross-user calibration-free which trains generalized models using the population data, and hence reduces the training burden of the user 9,35,36 . The presented large-sample dataset can provide resources for such calibration-free models.…”