Use of a wearable electromyography armband to detect lift-lower tasks and classify hand loads

Abstract

We used an armband with embedded surface electromyography (sEMG) electrodes, together with machine-learning (ML) models, to automatically detect lifting-lowering activities and classify hand loads. Nine healthy participants (4 male and 5 female) completed simulated lifting-lowering tasks in various conditions and with two different hand loads (2.3 and 6.8 kg). We compared three sEMG signal feature sets (i.e., time, frequency, and a combination of both domains) and three ML classifiers (i.e., Random Forest, Support Vector Machine, and Logistic Regression). Both Random Forest and Support Vector Machine models, using either time-domain or time- and frequency-domain features, yielded the best performance in detecting lifts, with respective accuracies of 79.2% (start) and 86.7% (end). Similarly, both ML models yielded the highest accuracy (80.9%) in classifying the two hand loads, regardless of the sEMG features used, emphasizing the potential of sEMG armbands for assessing exposure and risks in occupational lifting tasks.