Classification of P300 signals from P300 spelling system based on ASK-CNN model

How to detect P300 signal efficiently and accurately is great importance to improve the performance of P300-speller system. In this paper, a convolutional neural network (CNN) classification with average-scale kernel size (ASK-CNN model) is proposed to extract the temporal features of P300 signals for different subjects in P300 spelling system, and further optimize the classification accuracy compare with traditional CNN for P300 EEG signals detection. The experimental results show that compared with the traditional CNN model for detecting P300 signals, the ASK-CNN model can significantly improve the classification accuracy of P300 signals, especially when the superposing times of stimulus flashing is less. In addition, our model training speed is also faster than the traditional CNN model.