High dimensional feature data reduction of multichannel sEMG for gesture recognition based on double phases PSO

Surface electromyography (sEMG) is a kind of valuable bioelectric signal and very potential in the field of human–machine interaction. Ideal interactions require sEMG based patterns recognition not only with high accuracy but also with good rapidity. However, too much real-time feature-related computation will greatly slow down the interaction, especially for multichannel sEMG. To decrease the feature-related time consumption, the paper formulates the feature reduction as an optimization problem, and develops a double-phases particle swarm optimization (PSO) with hybrid coding to solve the problem. In the research, the initial feature data set with 31 kinds of feature is built firstly based on eight subjects’ 16 channels forearm sEMG signals, then PSO is introduced to conduct the feature reduction of 31 × 16 dimensions through the feature and channel optimization in double phases. During the optimization, two improved k -nearest neighbor (KNN) methods such as weighted representation based KNN (WRKNN) and weighted local mean representation based KNN (WLMRKNN) are introduced to classify the gestures, and the classification accuracy is used to evaluate the particles of PSO. Experimental results and comparison analysis show that PSO based feature reduction methods outperform genetic algorithm (GA), ant colony optimization (ACO) and principal component analysis (PCA) based feature reduction methods. With the optimized feature data subset by PSO, WRKNN and WLMRKNN are superior to KNN, quadratic discriminant analysis (QDA), and naive bayes (NB) greatly. The proposed method can be applied in the pattern recognition of high dimensional sEMG with multichannel or high-density channels for the purpose of rapidity and without a decline of accuracy in real-time control. Further, it can be used to reduce the economic cost of the personalized customization equipment through the optimal channels for any subjects in the future.