In order to further improve the performance of motor imagery electroencephalogram (EEG) decoding, a new common spatial pattern (CSP) improvement method is proposed to address the problems of the CSP feature extraction method, that is, the logarithmic band power feature extraction method based on CSP transform and filter bank. First, the original EEG signals are preprocessed; then the preprocessed signals are spatially filtered using CSP transform; after that, the spatially filtered signals are decomposed into multiple sub-bands using filter bank, and the logarithmic band power of each sub-band signal is extracted as a feature; finally, the least absolute shrinkage and selection operator (LASSO) is used for feature selection, and the support vector machine (SVM) is used for classification. Experiments were conducted on the data set IIa of the brain-computer interface (BCI) competition IV, the proposed method achieved the highest average classification accuracy, and the result was 81.97%. The experimental results show that the classification performance of the proposed method is better than the existing improved CSP method, and the feature extraction time also has a greater advantage.