Lifting activity assessment using surface electromyographic features and neural networks

The surface electromyographic (sEMG) data of 12 trunk muscles of 10 workers during the execution of lifting tasks using three lifting indices (LI) were recorded. The aims of this work were to: 1) identify the most sensitive trunk muscles with respect to changes in lifting conditions based on the selected sEMG features and 2) test whether machine-learning techniques (artificial neural networks) used for mapping time and frequency sEMG features on LI levels can improve the biomechanical risk assessment. The results show that the erector spinae longissimus is the trunk muscle for which every sEMG feature can significantly discriminate each pair of LI. Furthermore, only when using multi-domain features (time and frequency) a more complex artificial neural network can lead to an improved biomechanical risk classification related to lifting tasks. [Display omitted] •LI (Lifting Index) calculated from RNLE can be estimated from EMG features.•Combining time and frequency EMG features improves the classification of LI levels.•Erector Spinae Longissimus is the most LI-sensitive muscle during lifting tasks.•ANNs provide a feasible technical solution for an EMG-based LI estimation.