A Decoding Model of Upper Limb Movement Intention Based on Data Augmentation

With the aging of the population, the application of brain-computer interfaces(BCIs) in the neural decoding of upper limb motion direction is becoming more extensive. However, how to improve the recognition accuracy of neural decoding is one of the key problems given limited training samples. In this paper, we proposed a neural decoding method of upper limb motion direction based on data augmentation. We used the deep convolutional generative adversarial networks(DCGANs), which is a data augmentation algorithm to generate more data to expand the training set to improve the accuracy of the model. We completed analysis on different numbers of real training data across eight subjects. The analysis results show that after using data augmentation, the average decoding accuracy given small amounts of training samples significantly increases, showing that the DCGANs algorithm can indeed improve the accuracy of the neural decoding model, and help to improve the practical application of BCIs in decoding upper limb motion.