Abstract: Intelligent recognition of electroencephalogram (EEG) signals is an important means of epilepsy detection. Current deep learning methods, especially convolutional neural networks, have some problems in the classification of EEG signals. For example, the algorithm complexity is too high, the sample size is too small, and the classification effect is poor. In order to predict seizures more accurately, an epileptic EEG detection method based on Fourier-based synchro squeezing transform  and deep convolution generative adversarial network is proposed. Firstly, the synchrosqueezing transform compresses the time-frequency energy processed by the short-time Fourier transform to the real instantaneous position, so that the frequency curve is more concentrated. Secondly, a deep convolution generation confrontation network is constructed as a feature extractor. Finally, epileptic seizure prediction is achieved. The experiment was conducted on the CHB-MIT EEG data set. Experimental results show that this method has a high classification accuracy rate of up to 97.9%. The use of generative adversarial networks effectively solves the problem of insufficient sample size. Combined with the synchrosqueezing transform preprocessing method, it has high recognition accuracy. 

Key words: epilepsy, electroencephalogram, synchrosqueezing transform, deep convolutional generative adversarial network