关键词: 表面肌电信号（sEMG）, 模式识别, 误差反向传播（BP）神经网络

Abstract: For better recognizing hand gestures，this paper reports a new thought that has taken the single finger’s condition as recognizing target set.Six groups’ sEMG of commonly used hand gestures are gathered，which are planned reasonably taking the single finger’s condition as datum.Each channel’s sample means are used to constitute feature eigenvector.Three parallel BP neural networks are designed，which can study the single finger’s condition from the hand gesture sample.The method makes the classified cardinal number to be small，thus reduces the complexity of classified order，and overcomes the shortcomings，which need to gather the movement many enough in the traditional multi-taxonomic approach.The experimental result indicates that：the sEMG of 12 kinds of hand movements are gathered;the hand movement is simplified reasonably to the finger movement，and the neural network is trained using finger’s condition.All composite movements of finger’s three conditions can be distinguished，that is to say，all commonly used 18 kinds of hand gestures have been classified.

Key words: surface Electromyograms（sEMG）, pattern recognition, Back Propagation（BP） neural network