A Gesture Recognition Method Based on Recurrent Spatiotemporal Depth Neural Network
To solve the problem of weak robustness and low precision of existing hand gesture recognition models induced by lack of spa-tiotemporal information,a hand gesture recognition model based on recurrent spatial and temporal deep neural network is proposed to improve the characterization ability for surface EMG(sEMG)signals.Firstly,a multi-channel convolutional neural network is designed and integrated into the bidirectional recurrent neural network to extract the spatiotemporal characteristics information with strong dis-crimination.Secondly,channel attention mechanism is used to catch the channel importance information in spatiotemporal characteris-tics,then an attention module based on spatiotemporal characteristics is designed to further enhance the spatiotemporal characteristics information.Thirdly,based on the ideology of feature pyramid network,a multi-scale feature fusion module is designed to acquire multi-stage feature information based on multi-scale and multi-angle aspects to improve the decoding ability of the model to electromyography signals.Finally,the proposed hand gesture recognition model is tested based on a large hand gesture recognition database of Ninapro.The results show that the representation capability for sEMg signals is effectively improved by the proposed method.It provides reference for the deep learning modeling work of human body hand gesture recognition.
国家科技期刊平台
NSTL
万方数据
