matlab使用deep network designer搭建网络进行图像增强，将搭建好的网络（没有在deep network designer导入数据进行训练）导出到工作区后，训练网络，编写一段程序，使batch 大小设置为 4，总计迭代 60 个 epoch，每完成 15 轮迭代学习率衰减为原先的 10%

好的，这是一个比较复杂的问题，需要分步骤来解决。

首先，导入深度学习工具箱和图像增强网络：

% 导入深度学习工具箱和图像增强网络
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\nnet\nnutils\customlayer');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\nnet\nnutils');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\nnet\nngui');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\nnet');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\deepapp');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\deep');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\cnn');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\vision');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\nnet');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\deep_learning');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\series');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\stats');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\rl');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\time_series');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\transfer_learning');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\gans');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\reinforcement_learning');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\training');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\timeseries');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\deeplearning_shared');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\deeplearning_shared\utils');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\deeplearning_shared\data');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\deeplearning_shared\training');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\examples\deeplearning_shared\visualization');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\deepLearning\supportPackages\cuDNN');
addpath('C:\Program Files\MATLAB\R2021a\toolbox\nnet\deepLearning\supportPackages\cuDNN\bin');

% 导入图像增强网络
net = importKerasNetwork('image_enhancer.h5');

然后，我们需要编写一个训练函数，该函数将负责训练网络并在每个 epoch 结束时调整学习率：

function trainNet(net, data, labels, batchSize, epochs)
% 训练函数
% net：网络模型
% data：训练数据
% labels：训练标签
% batchSize：批大小
% epochs：迭代次数

% 设置学习率
initialLearningRate = 0.01;
decay = 0.1;
numEpochsToDrop = 15;
learnRateSchedule = @(epoch) initialLearningRate * decay^(floor(epoch/numEpochsToDrop));

% 配置训练选项
options = trainingOptions('sgdm', ...
    'MiniBatchSize', batchSize, ...
    'MaxEpochs', epochs, ...
    'InitialLearnRate', initialLearningRate, ...
    'LearnRateSchedule', 'piecewise', ...
    'LearnRateDropFactor', decay, ...
    'LearnRateDropPeriod', numEpochsToDrop, ...
    'Shuffle', 'every-epoch', ...
    'Verbose', true, ...
    'Plots', 'training-progress');

% 训练网络
trainedNet = trainNetwork(data, labels, net.Layers, options);
end

最后，我们可以在主程序中调用 trainNet 函数来训练网络：

% 导入训练数据和标签
data = ...
labels = ...

% 训练网络
batchSize = 4;
epochs = 60;
trainNet(net, data, labels, batchSize, epochs);

这样就完成了训练过程，每 15 轮迭代学习率会衰减为原先的 10%。