Abstract-Wearable technology has gained increasing popularity in the applications of healthcare, sports science and biomedical engineering in recent years. Because of its convenient nature, the wearable technology is particularly useful in the acquisition of the physiological signals. Specifically, the sEMG systems, which measure the muscle activation potentials, greatly benefit from this technology in both clinical and industrial applications. However, the current wearable sEMG systems have several drawbacks including inefficient noise cancellation, insufficient measurement quality and difficult integration to customized applications. Additionally, none of these sEMG data acquisition systems can detect sEMG signals (i.e., contractions), which provides a valuable environment for further studies such as human machine interaction, gesture recognition and fatigue tracking. To this end, we introduce an embedded, 8-channel, noise canceling, wireless, wearable sEMG data acquisition system with adaptive muscle contraction detection. Our design consists of two stages, which are the sEMG sensors and the multi-channel data acquisition unit. For the first stage, we propose a low cost, dry and active sEMG sensor that captures the muscle activation potentials, a data acquisition unit that evaluates these captured multi-channel sEMG signals and transmits them to a user interface. In the data acquisition unit, the sEMG signals are processed through embedded, adaptive methods in order to reject the power line noise and detect the muscle contractions. Through extensive experiments, we demonstrate that our sEMG sensor outperforms a widely used, commercially available product and our data acquisition system achieves 4.583 dB SNR gain with 98.9784% accuracy in the detection of the contractions.

We investigate the nonlinear regression problem under L2 loss (square loss) functions. Traditional nonlinear regression models often result in non-convex optimization problems with respect to the parameter set. We show that a convex nonlinear regression model exists for the traditional least squares problem, which can be a promising towards designing more complex systems with easier to train models.

No abstract

Muscle fatigue detection and tracking has gained significant attention as the sports science and rehabilitation technologies developed. It is known that muscle fatigue can be evaluated through surface Electromyography (sEMG) sensors, which are portable, non-invasive and applicable for real-time systems. There are plenty of fatigue tracking algorithms, many of which uses frequency, time and time-frequency behaviors of sEMG signals. An example to most commonly used sEMG-based fatigue detection methods can be mean frequency (MNF), median frequency (MDF), zero-crossing rate (ZCR) and continuous wavelet transform (CWT). However, all of these muscle fatigue calculation methods are adversely affected by the dynamically changing sEMG contraction amplitude, since EMG spectrum also demonstrates a shift with the changing signal RMS; powerful contractions lead a shift to high frequency bounds and the opposite happens for the weak. To overcome that, we propose an adaptive algorithm, which learns the effect of contraction power on sEMG power spectral density (PSD) and subtracts that amount of frequency shift from the PSD.

<p><strong>Motion classification with surface electromyog-</strong><br> <strong>raphy (sEMG) has been studied for practical applications</strong><br> <strong>in prosthesis limb control and human-machine interaction. Recent studies have shown that feature learning with deep neural networks (DNN) reaches considerable accuracy in motion classification tasks. However, DNNs require large datasets for acceptable performance and fail for tasks with few data samples available for training. Professional athlete training includes hundreds of exercises, and coupled with privacy and confidentiality issues acquiring a large dataset for all the exercises is not feasible. As a result, state-of-the-art DNN architectures are unsuitable for real-life sports applications. We utilise few-shot learning (FSL) techniques to overcome the small dataset problem of sports-related motion classification tasks. The employed methodology uses the knowledge gathered from a large set of tasks to classify unseen tasks with a few data samples. The FSL approach with a siamese network and triplet loss reached the best performance with a median F1-score of</strong> 72.01%<strong>,</strong> 76%<strong>, and</strong> 79% <strong>for 1, 5 and 10 shot datasets that include an unseen set of tasks, respectively. In contrast, DNN with transfer learning (TF) reached</strong> 49.27%<strong>,</strong> 51.58%<strong>, and</strong> 67.66% <strong>for the same set of tasks, respectively.</strong></p>