Multiple classifier system with sensitivity based dynamic weighting fusion for hand gesture recognition

Surface electromyogram (sEMG) signal has been applied to gesture recognition successfully. However, performance of gesture recognition degrades due to noise, which is unavoidable in practical environment. This study aims to propose a gesture recognition system using a multiple classifier system (MCS) based on random subspace. Our method makes a decision using a signal which is less sensitive to noise. Each base classifier is trained using the subset of channels selected randomly. The influence of a base classifier by a noisy sample is measured by sensitivity, defined as the perturbation of an output of a classifier. The final decision of the MCS is determined according to the output of the based stable classifiers fused by majority vote. Experimental results show that the proposed method achieves better classification performance in comparison with its base classifier and the random subspace method in terms of accuracy and robustness to noise.