An Improved Greedy Reduction Algorithm Based on Neighborhood Rough Set Model for Sensors Screening of Exoskeleton

Abstract

The reasonable selection of sensors is essential for sensor fusion for gait recognition, and this paper proposes a reduction method based on an improved greedy algorithm, which can screen a suitable sensor combination with strong distinguishing ability. This novel reduces the multi-sensor system containing 17 sensors by optimizing the neighborhood rough set model, and uses grey relational analysis (GRA) for post-processing to get the optimal reduction. In addition, these reductions under three different terrains are restructured and extended to form the optimal combination. In order to verify the feasibility of the algorithm, we detect the gait phases and compare the accuracy. Results reveal that the number of sensors is reduced from 17 to 8, and the accuracies of three different terrains are increased by 1.05%, 2.633% and 5.934% respectively compared with previous sensors, and increased by 0.533%, 5.950% and 3.834% respectively compared with sensors using principal component analysis (PCA). The Wilcoxon rank sum test is carried out and the results show the algorithm has good performance. The experiments show that this method can screen a few sensors while maintaining or improving the classification ability, and it has high engineering practical significance in wearable robotics field and many other sensor fields.