Athletes have to receive a large amount of complex and variable information at all times during training and competition, especially in high-level events, where the processing of clinical stimuli directly determines victory or defeat when athletes are of comparable abilities. Therefore, it is particularly important to improve the attention and executive function of high-level athletes, and the traditional way is to improve the attention of sports through the perspective of sports training, while neglecting the analysis of sports EEG characteristics. Therefore, this paper proposes a deep learning-based EEG signal classification method. In the data processing stage, z-score normalization is used for feature data and one-hot coding is used for label data. After that, the pre-processed EEG data are divided into three parts: training set, validation set and test set. The training set is mainly used for the training stage of the model, the validation set is mainly used for the setting of hyper-parameters, and the test set is used for the evaluation of the model performance. In the model training stage, two recursive structures, Long Short-term Memory (LSTM) and Gated Recurrent Unit (GRU), were used as the base predictor, and the AdaBoost algorithm was used to integrate the prediction of the results obtained after the training of the base predictor, and the four states of the athlete's brainwave were finally classified and identified. The four states of the athlete's brain waves were finally classified and identified. The experiments in this paper are carried out on the dataset provided by Physio Net, and the experimental results show that this scheme has better effectiveness and accuracy, and can improve the performance and training effect of athletes.