| Attention is the foundation and prerequisite for all cognitive activities.Distraction in students’ learning not only affects their learning efficiency and performance,but may further affect their mental health.Therefore,improving students’ attention levels through attention training has important social significance and value.However,current attention training methods are mainly targeted at children with attention deficit and hyperactivity disorders,and all of them have limitations.For example,sensory integration training has a long training cycle and strong professionalism;Biofeedback training requires expensive hardware systems support and cannot achieve personalized training for different individuals.Therefore,it is necessary to propose an attention training method that is simple to operate,easy to implement,and widely applicable.Based on the above considerations,this thesis carries out the research of attention training method from the perspective of EEG signal processing,and the main work of the thesis includes:(1)Aiming at the problem that healthy students are easily distracted during learning,α music is employed to train students’ attention.In this method,the music track and duration can be adjusted according to the training situation and the subjects’ preferences to achieve the effect of personalized training.Before and after the music training,subjects are required to complete the attention network test,content test and EEG signal acquisition.By analyzing the results of the attention network test and content test,the effect of training is preliminarily analyzed;the raw EEG signals are preprocessed,and the minority samples are synthesized by oversampling technique.In addition,the Schulte square training method is used as a control experiment,during which the number of squares and other parameters are adjusted according to the training situation of different subjects.Experimental results show that compared with Schulte square training,the α musical attention training can be done in a relaxed state,which makes it easier to stick to and achieve the expected effects.(2)Aiming at the problem that a single feature can only represent part of the information in EEG signal,the time-frequency domain features and the nonlinear dynamics features are normalized and fused,in order to comprehensively characterize the attention EEG signal.In the time-frequency domain,the energies of δ-band,θ-band,α-band and β-band,as well as the energy ratio between each of the two bands are extracted by wavelet packet transform and the sample entropy features are extracted using non-linear analysis.The 11 features of different dimensions are normalized and fused,which are then fed into several common classifiers to distinguish between attention and non-attention states.Based on the fused multi-dimensional feature fusion matrixs,an 88.7% classification accuracy can be achieved using a support vector machine(SVM)classifier,which is 4% to 13% higher than using a single feature.In addition,the effectiveness of the training method is further discussed by analyzing the feature differences before and after attention training.(3)Aiming at the problem that classification accuracy of attention EEG signals is not satisfactory,a double convolutional neural network fusion model is proposed by linear weighted fusion of the modified AlexNet network and VGG11 network.The multi-dimensional feature fusion matrixs before and after attention training are converted into the corresponding multi-dimensional feature color RGB images for input model to distinguish attention and non-attention states.The results show that the classification accuracy of the proposed model can reach 97.53%,which is better than using the single AlexNet network and VGG11 network,respectively.Compared with the recognition accuracy of the model before attention training,the accuracy of the model after attention training is 2%-4% higher.Meanwhile,compared with the performance of SVM,K-Nearest Neighbor and Naive Bayes algorithm,the proposed double convolutional neural network fusion model shows advantages in various evaluation indicators. |
