Abstract: In order to solve the problem of low recognition rate of electroencephalogram (EEG) signal by deep belief network (DBN), multi-band frequency domain deep belief network (multi-band FDBN) algorithm is presented for feature extraction. Due to the individual differences of different bands, their contributions to the classification results aren't exactly the same. In this paper, the band-pass filter is used to divide the original EEG signal into multiple frequency bands, and then the FFT (fast Fourier transform) is used to convert the time domain signals into frequency domain signals which are normalized. Finally, the frequency domain data of each frequency band is input into DBN for training and identification. Offline experiments show that compared with frequency domain deep belief network (FDBN), the accuracy of multi-band FDBN increases by 3.25% in average, the variance is smaller, and the robustness is better. Also, the multi-band FDBN algorithm is used to control an intelligent wheelchair platform to perform "8" trajectory with motion imaginary EEG signals of the left and right hands, which proves the effectiveness of the algorithm in EEG feature extraction.