sEMG-based upper limb motion recognition using improved sparrow search algorithm

The study of motion intention recognition of patients has become one of the key directions in the current research on human-machine coordination control of rehabilitation robots. To improve the accuracy of motion intention recognition and shorten the recognition time, in this work, an improved sparrow search algorithm based on multi-strategy (MSISSA) is designed for improving the prediction performance of the classification algorithm for motion pattern recognition of human upper limbs based on 4-channel sEMG signals. For the poor quality of the initial solution of the population, elite initial solutions are defined by an opposition-based learning strategy to enhance the diversity and traversal of the population. Due to the lack of effective step size control and variation mechanism in the iterative process, a nonlinear exponential decreasing strategy is proposed to balance the global search and local exploitation ability of the algorithm, and a vertical and horizontal crossover strategy is introduced after the individual position update in the population to improve the ability of the algorithm to jump out of the local optimum. The cross-sectional and longitudinal experiments show that the proposed MSISSA algorithm has certain advantages in terms of convergence speed, solution accuracy and robustness, and the classifier optimized based on the MSISSA algorithm has a 2.835% improvement in the accuracy of sEMG signal recognition compared with the original classifier, which is of positive significance for the application in acquiring patient intention for robot-assisted rehabilitation motions.