Rapid gesture recognition method based on fusion of transient electromyographic signal spatial-temporal characteristics and force sensor signals

Abstract

The invention relates to a rapid gesture recognition method for transient high-density myoelectricity. Auxiliary supervision training is carried out by adopting spatial-temporal feature extraction and combining with a force sensor signal. In the training stage, electrode interference detection, damaged electrode repair, standardization and other preprocessing are performed on transient high-density electromyographic signals, then three-dimensional spatial-temporal characteristics of the electromyographic signals are constructed, phase-physiology joint time coding is performed, the two are combined and then sent into a joint model based on a selective attention mechanism, and a joint model is constructed; optimization is carried out by minimizing weighted total loss including cross classification loss, force signal regression loss and physiological dynamic loss. In the test stage, only transient electromyographic signals are input, and gesture classification is completed by utilizing a trained model after preprocessing and spatial-temporal feature extraction.

Images