Researchers develop a self-supervised artificial intelligence adaptation framework.

2023-05-26

Researchers have developed a self-supervised artificial intelligence adaptation framework to improve the sensing accuracy of EMG devices. Surface electromyography (EMG) has been widely used to measure the electrical activity of muscles. However, the variability of EMG sensing signals due to biological differences among different users greatly reduces the performance and potential of EMG systems. Recently, researchers from City University of Hong Kong have developed a deep learning-based framework called EMGSense, which uses self-training techniques of artificial intelligence to achieve high sensing performance for new users. This opens up a new path for the development of more advanced and accurate wearable EMG devices in fields such as neural rehabilitation and virtual reality. This latest invention was awarded at the 21st International Conference on Pervasive Computing and Communications (PerCom 2023) held in Atlanta, USA. It helps overcome the bottlenecks of existing methods and supports the popularization of EMG-based applications. EMG measures the electrical activity of muscles using surface electrodes on the skin. EMG sensing has attracted considerable attention in recent years and has created many intelligent applications such as neural rehabilitation, activity recognition, gesture recognition, and virtual reality. However, a fundamental challenge of existing EMG systems is how to deal with inter-user scenarios. Muscle electrical signals are heavily influenced by various biological factors such as body fat, skin condition, age, and fatigue. Therefore, the temporal biological heterogeneity when different users use EMG systems leads to significant performance degradation. To address this challenge, researchers from the Department of Computer Science at City University have recently proposed a low-cost, AI-empowered domain adaptation framework called EMGSense, which utilizes AI training techniques to provide new users with high-precision EMG sensing. EMGSense is a self-supervised system with self-intelligent strategies. It can cope with the performance degradation caused by inter-user biological heterogeneity. The new framework integrates advanced self-supervised techniques into a well-designed deep neural network (DNN) structure. It uses small-scale unlabeled data from new users and pre-collected data from multiple existing users to train discriminative models for intelligent applications for new users. The pre-collected data is stored in the cloud and can serve all new users, reducing the burden of data collection and annotation. The key principle of this approach is to share a common feature extractor to ensure the transferability of features. The combination of domain-specific feature extractors and classifiers is responsible for independently exploring the differences between deep features from different source domains. EMGSense's DNN structure consists of two complementary training stages. It first eliminates domain-specific features for ease of transfer, and then uses AI techniques to relearn user-specific biological features of the new target to achieve high-performance EMG sensing. This allows EMGSense to adapt to new users with satisfactory performance in a low-workload, self-supervised manner without wasting a large amount of deployment overhead. In addition, researchers utilize unlabeled data collected during the usage process to achieve long-term robust performance, which can handle the temporal variability of EMG signals. A comprehensive evaluation of two large datasets collected from 13 participants shows that EMGSense achieves average accuracies of 91.9% and 81.2% in gesture recognition and activity recognition, respectively. EMGSense outperforms existing techniques in EMG domain adaptation methods by 12.5%-17.4% and performs comparably to frameworks trained in a supervised learning manner. This novel EMGSense framework achieves high accuracy in a cost-effective manner by reducing the burden of data collection and annotation, and has the potential to completely change the field of EMG sensing. It fills the research gap in the heterogeneity problem of EMG sensing and enables various new cross-user applications based on EMG, such as clinical practice, neural rehabilitation, and human-computer interaction. It also takes a small step towards the popularization of intelligent EMG wearable devices with higher performance in real-world scenarios. Source: https://techxplore.com/news/2023-05-self-supervised-ai-framework-accuracy-emg.html

RECENT AI NEWS

RECENT AI TOOLS