Prediction of hand grip strength based on surface electromyographic signals

Abstract

The activity status of muscles in real-time should be obtained in hand rehabilitation training to adjust the auxiliary grip-strength value with a good hand grip-strength prediction model. Surface electromyography (sEMG) has great advantages in reflecting human hand motions noninvasively. Therefore, based on sEMG, the work proposed a hand grip-strength prediction method optimizing the support vector regression (SVR) model through Sparrow Search Algorithm (SSA). First, the time domain characteristics were extracted from sEMG signals, and grip-strength signals were interpolated by the cubic spline function to extract the average value of grip-strength signals. Then Pearson’s correlation coefficient was used to select appropriate feature combinations to enter the SSA-SVR model for grip-strength prediction. Finally, particle swarm optimization (PSO) and genetic algorithm (GA) were used to optimize the SVR model, and the effectiveness of the SSA-SVR model was verified by comparison. Random forest (RF) with SSA algorithm and BP neural network models was optimized to test the effect of SSA on improving the performance of the SVR model. The performance of the SSA-SVR model was better than that of other regression prediction models, with a root mean square error (RMSE) of 0.5054, an average absolute error (MAE) of 0.3242, an average absolute percentage error (MAPE) of 0.0864, and an accuracy (R2) of 93.2%. It can accurately predict the hand grip-strength value, which is suitable for hand rehabilitation training.