EEG recognition model based on wavelet packet decomposition and ConvNet for art design decisions

Abstract

Aiming at the subjective and unquantifiable problems in art design decisions, using EEG (electroencephalogram) to explain individual behavior is an effective solution. However, the accuracy of the traditional method is lower because of the complexity of EEG. Therefore, we propose the WPD-ConvNet recognition model based on wavelet packet decomposition and convolutional networks to enhance the recognition accuracy of EEG. First, subjects wear brain electrode caps to watch four art designs with different styles and need to make decisions. The EEG is collected during this process. Then, we use the WPD to obtain the features of EEG signals from the decomposed optimal subband nodes. Compared with FFT, WPD divides the frequency band through multiple levels to obtain more EEG time-frequency features related to art design decisions. Finally, we construct an convolutional neural network to identify these features. The experimental results show that our proposed EEG recognition model can achieve an accuracy of 88.7%, which is about 10% higher than others.