Decoding finger movement patterns from microscopic neural drive information based on deep learning

•A novel method for decoding finger movement patterns from microscopic neural drive information is proposed, which is more in accordance with the physiological nature.•The proposed method is superior to other common methods based on motor units (MU) activities and macroscopic features.•Convolutional neural network was used for MU classification by better characterization of spatial information from MU.•A fuzzy weighted decision strategy was designed to solve the issue of MU shared between movements due to common muscle co-activation.