It is difficult to use a single EEG classifier to achieve an ideal recognition accuracy of control intention, when the operator is in different operating states, and both arm operation and EEG control are used in the underwater vehicle operation. An algorithm is proposed to improve the recognition accuracy by selecting the classification result by using the combined classifier and correcting the classification result according to the specific situation of the actual operation. Firstly, the Fisher discriminant method is used to train the data generated by the armless operation and the arm operation to get the classifier in two operation states. Then, the two classifiers are combined, and the curve fitting is used to determine the reference distance difference which is used to determine the classification result(the selection of the difference takes into account individual differences). Futhermore, the distance difference is corrected by the distance correction function according to the particularity of the actual operation situation. Finally, the final classification result is determined by comparing the difference between the reference distance difference and the corrected distance difference. In order to verify the effectiveness of the proposed algorithm, six subjects are invited to participate in the testing process. The experimental results show that the proposed algorithm improves the recognition accuracy by 13.42%, 5.55% and 5.55% respectively, compared with the other three methods in the designed online experiment, which demonstrates that the proposed algorithm is feasible and effective.