Deep Learning for Electromyographic Hand Gesture Signal Classification Using Transfer Learning (1801.07756v5)

Abstract: In recent years, deep learning algorithms have become increasingly more prominent for their unparalleled ability to automatically learn discriminant features from large amounts of data. However, within the field of electromyography-based gesture recognition, deep learning algorithms are seldom employed as they require an unreasonable amount of effort from a single person, to generate tens of thousands of examples. This work's hypothesis is that general, informative features can be learned from the large amounts of data generated by aggregating the signals of multiple users, thus reducing the recording burden while enhancing gesture recognition. Consequently, this paper proposes applying transfer learning on aggregated data from multiple users, while leveraging the capacity of deep learning algorithms to learn discriminant features from large datasets. Two datasets comprised of 19 and 17 able-bodied participants respectively (the first one is employed for pre-training) were recorded for this work, using the Myo Armband. A third Myo Armband dataset was taken from the NinaPro database and is comprised of 10 able-bodied participants. Three different deep learning networks employing three different modalities as input (raw EMG, Spectrograms and Continuous Wavelet Transform (CWT)) are tested on the second and third dataset. The proposed transfer learning scheme is shown to systematically and significantly enhance the performance for all three networks on the two datasets, achieving an offline accuracy of 98.31% for 7 gestures over 17 participants for the CWT-based ConvNet and 68.98% for 18 gestures over 10 participants for the raw EMG-based ConvNet. Finally, a use-case study employing eight able-bodied participants suggests that real-time feedback allows users to adapt their muscle activation strategy which reduces the degradation in accuracy normally experienced over time.

• The paper demonstrates that transfer learning enhances EMG gesture classification by leveraging diverse datasets for pre-training and evaluation.
• It explores three deep learning networks using raw signals, STFT spectrograms, and CWT, highlighting the superior performance of the CWT-based ConvNet.
• The study provides accessible datasets and models, paving the way for improved real-time adaptation in assistive and rehabilitation technologies.

Deep Learning for Electromyographic Hand Gesture Signal Classification Using Transfer Learning

The paper presented investigates the application of deep learning and transfer learning to electromyographic (EMG) hand gesture recognition. The primary objective is to leverage large aggregated datasets from different users to improve classification accuracy while reducing individual training effort. Utilizing the Myo Armband, the paper explores innovative approaches to classifier design through deep learning techniques, significantly enhancing the performance of traditional models in this domain.

Methodology

This research utilizes transfer learning to improve gesture recognition from EMG signals. Two newly recorded datasets, comprising 19 and 17 participants respectively, facilitate pre-training and evaluation. An additional dataset from the NinaPro database encompasses 10 additional participants, broadening the testing landscape.

Three deep learning networks are designed: one based on raw EMG signals, another on Spectrograms using Short-Time Fourier Transform (STFT), and a third employing Continuous Wavelet Transform (CWT). These configurations explore various input modalities to evaluate performance improvements brought about by the proposed transfer learning scheme.

Results

The paper reveals that transfer learning consistently enhances classifier performance. Notably, the CWT-based ConvNet achieves remarkable accuracy for both limited (7 gestures over 17 participants with an accuracy of 98.31%) and expansive gesture sets (18 gestures over 10 participants reaching 68.98% accuracy). These outcomes underscore the considerable potential of using aggregated datasets to uncover common signal patterns among varying subjects.

Key Contributions

1. Transfer Learning Scheme: The transfer learning framework designed is both robust and flexible, permitting knowledge transfer across different user datasets without sacrificing accuracy.
2. Real-Time Adaptation: An interesting finding is that with real-time feedback, users can adjust muscle strategies reducing accuracy degradation over time. This practical insight is immensely valuable for real-world applications of muscle-controlled prosthetics.
3. Data and Models: By developing and releasing a new dataset and models, the paper offers the sEMG research community accessible data and methodologies to further advance this domain.

Theoretical and Practical Implications

Theoretically, the paper propels the understanding of leveraging common neural patterns across different users, opening pathways for not only gesture recognition but broader applications in bio-signal processing. Practically, the work implies more efficient and effective training paradigms for AI systems applied to human-computer interaction technologies, notably in assistive and rehabilitation devices.

Future Directions

Moving forward, research can explore adapting the proposed methodologies to datasets with clinical end products, such as prosthetics for amputees. Furthermore, investigating the framework for out-of-the-box inter-session adaptability will provide insights into the longevity and robustness of trained models without continual recalibration.

In summary, this paper offers a comprehensive approach to improving sEMG-based gesture recognition using deep learning and transfer learning, contributing significantly to both the practical and theoretical landscapes in AI and artificial intelligence-assisted rehabilitation technologies.