In order to solve the problem about the low classification accuracy of the motor imagery electroencephalography(EEG), this paper proposes an improved stacked denoising auto-encoders(ISDAE) network combined with the immune optimization algorithm and decision-making mechanism.The ISDAE network extractes the most effective feature vectors by multi-layer DAE, and then identifies these feature vectors by the neural network. Meanwhile, the decision-making mechanism is obtained, and the immune optimization algorithm is used to optimize parameters of ISDAE network, thus the improved ISDAE model with higher classification accuracy is obtained.The experimental results show that, the proposed ISDAE model has a strong ability for feature learning from raw EEG data and a higher recognition rate, which provides an efficient technique for the recognition of the motor imagery EEG.