Compressed reconstruction of electromyographic signals based on deep learning

In recent years, the rapid advancement of wearable technology has propelled significant progress in vital sign monitoring and remote diagnostic assistance. This evolution has steered bio-electrical signal monitoring towards portable and real-time solutions. Among these signals, surface electromyography (sEMG) plays a pivotal role in observing human activity. However, the long-term multi-channel continuous monitoring of sEMG signals imposes a substantial burden on transmission systems, reducing the portability of devices and increasing power consumption. These limitations fail to meet the stringent requirements of wireless body area networks (WBANs). Compressed sensing-based reconstruction of sEMG signals emerges as an effective solution to this challenge. Traditional compression methods that rely on the Nyquist sampling theorem are not only inefficient but also wasteful of sampling resources. In contrast, conventional compressed sensing (CS) methods, which depend on prior knowledge, exhibit low reconstruction efficiency. Fortunately, deep learning based on CS frameworks has demonstrated superior adaptability and performance. This study proposes a temporal convolutional network (TCN) based on the CS reconstruction method, which achieves high-quality signal recovery with significantly fewer sampling points. Experimental results demonstrate that the proposed method significantly outperforms traditional approaches in terms of reconstruction accuracy. Moreover, it enables efficient signal compression without compromising signal quality, thereby addressing the key challenges faced by WBANs. This research paves the way for more efficient and portable bio-electrical signal monitoring systems in the future.