Compact CNN for Rapid Inter-day Hand Gesture Recognition and Person Identification from sEMG

Surface electromyography (sEMG)-based hand gesture recognition (GR) is one alternative means for human-machine interaction. However, the real application of sEMG confronts many challenges. One challenge is recognizing GR on a daily basis which is difficult due to interferences, e.g., electrode shift. This study proposed a compact multi-view CNN model that can be used for the GR mentioned above and also generalized for person identification tasks (PI). Simultaneously capturing inter-channel and intra-channel features by combining separable and depthwise convolutions with specific data shapes in our model results in performance elevating even in inter-day-circumstance. Specifically, our model is approximately 56 times smaller in terms of the number of parameters, more than 16 times faster in terms of training time, and more than 1.5 times faster in terms of inference time for both GR and PI tasks compares to the baseline model by the same validation method and the dataset used.