Machine learning methods for electromyography error detection in field research: An application in full-shift field assessment of shoulder muscle activity in apple harvesting workers

Abstract

This study presented an alternative technique for processing electromyography (EMG) data with sporadic errors due to challenges associated with the field collection of EMG data. The application of this technique was used to detect errors, clean and optimize EMG data in order characterize and compare shoulder muscular load in farmworkers during apple harvesting in a trellised orchard. Surface EMG was used to take measurements from twenty-four participants in an actual field work environment. Anomalies in the EMG data were detected and removed with a customized algorithm using principal component analysis, interquartile range cut-off and unsupervised cluster analysis. This study found significantly greater upper trapezius muscle activity in farmworkers who used a ladder as compared to the alternative platform-based method where a team of mobile platform workers harvested apples from the tree tops and a second separate team of ground workers harvested apples from the tree bottoms. By comparing the unprocessed and the processed, anomaly-free EMG data, the robustness of our proposed method was demonstrated.