function net = train_EEGNet(X_train, Y_train, Fs, T, EEGNet_Params) % 创建EEGNet模型 layers = [ sequenceInputLayer([1 T*Fs 1],'Name','InputLayer') convolution2dLayer([1 EEGNet_Params.F1],'NumChannels',EEGNet_Params.F2,'Padding','same','Name','ConvLayer1') batchNormalizationLayer('Name','BatchNormLayer1') depthwiseConv2dLayer([EEGNet_Params.D EEGNet_Params.F2],'Padding','same','Name','DepthConvLayer') batchNormalizationLayer('Name','BatchNormLayer2') averagePooling2dLayer([1 EEGNet_Params.T],'Name','AvgPoolingLayer') dropoutLayer(EEGNet_Params.dropOutRate,'Name','DropoutLayer') convolution2dLayer([1 EEGNet_Params.F3],'NumChannels',EEGNet_Params.F4,'Padding','same','Name','ConvLayer2') batchNormalizationLayer('Name','BatchNormLayer3') flattenLayer('Name','FlattenLayer') fullyConnectedLayer(EEGNet_Params.numClasses,'Name','OutputLayer') softmaxLayer('Name','SoftmaxLayer') classificationLayer('Name','ClassificationLayer')]; % 设置训练选项 options = trainingOptions('adam', ... 'MaxEpochs', EEGNet_Params.numEpochs, ... 'MiniBatchSize', EEGNet_Params.miniBatchSize, ... 'InitialLearnRate', EEGNet_Params.initialLearnRate, ... 'LearnRateSchedule','piecewise', ... 'LearnRateDropFactor', EEGNet_Params.dropFactor, ... 'LearnRateDropPeriod', EEGNet_Params.dropPeriod, ... 'Shuffle','every-epoch', ... 'Plots','training-progress', ... 'ExecutionEnvironment','gpu'); % 训练EEGNet模型 net = trainNetwork(X_train, categorical(Y_train), layers, options); end