Motion direction prediction through spike timing based on micro Capsnet networks

Neural activity extraction and neural decoding from neural signals are an important part of critical components of brain-computer interface systems. With the development of brain-computer interface technology, the demand for precise external control and nervous activities in macaque monkey during unilateral hand grasp has increased the complexity of control and neural decoding, which puts forward higher requirements for the accuracy and stability of feature extraction and neural decoding. In this study, a micro Capsnet network architecture that consists of a few network layers, a vector feature structure, and optimization network parameters, is proposed to decrease the computing time and complexity, decrease artificial debugging, and improve the decoding accuracy. Compared with KNN, SVM, XGBOOST, CNN, SimpleRNN, and LSTM, the algorithm in this study improves the decoding accuracy by 98.03%, and achieves state-of-the-art accuracy and stronger robustness. Furthermore, the proposed algorithm can further enhance the control accuracy in the brain-computer interface.