Electromyographic movement pattern recognition based on random forest algorithm

Movement pattern recognition is the basis for flexible control of lower-limb rehabilitation aids. To improve the effect of lower-limb movement pattern recognition, an electromyographic recognition method to identify movement patterns without movement information was proposed. Firstly, the initial moment of feature extraction was detected by the integral of absolute value of surface electromyography (EMG) recorded from gluteus medius muscle. Secondly, the features were extracted from the surface EMG recorded from five main muscles of lower limb. Finally, Random Forest algorithm was applied to recognize the five movement modes (level-ground walking, stair ascent, ramp ascent, stair descent, and ramp descent), while the importance of every feature was estimated by the recognition precision and Gini index. The features with greater contribution were picked out and applied to recognize. The simulation result indicates that this method had achieved an average accuracy of 99.2% in five movement modes recognition, which is conductive to the further study of lower-limb rehabilitation aids.