A Survey on Deep Learning in Electromyographic Signal Analysis

In the recent past Deep Learning (DL) has been used to develop intelligent systems that perform surprisingly well in a large variety of tasks, e.g. image recognition, machine translation, and self-driving cars. The huge improvement of the elaboration hardware and the growing need of big data processing have boosted the DL research in several fields. Recently, physiological signal processing has taking advantage of deep learning as well. In particular, the number of studies concerning the analysis of electromyographic (EMG) signals with DL methods is exponentially raising. This phenomenon is mainly explained by both the existing limitation of the myoelectric controlled prostheses and the recent publication of big datasets of EMG recordings, e.g. Ninapro. Such increasing trend motivated us to search and review recent papers that focus on the processing of EMG signals with DL methods. A comprehensive literature search of papers published between January 2014 and March 2019 was performed referring to the Scopus database. After a full text analysis, 65 papers were selected for the review. The bibliometric research shows four distinct clusters focused on different applications: Hand Gesture Classification; Speech and Emotion Classification; Sleep Stage Classification; Other Applications. As expected, the review process revealed that most of the papers related to DL and EMG signal processing concerns the hand gesture classification, and the convolutional neural network is the most used technique.