Neural Network Toolbox User's Guide 2017

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

本帖隐藏的内容

Neural Network Toolbox User’s Guide 2017.pdf (4.65 MB)

2017-9-24 01:46:48 上传

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏1 回帖

关键词：toolbox network Neural Guide User

Get Started with Transfer Learning

1. Unzip and load the new images as an image datastore. Divide the data into training and validation data sets. Use 70% of the images for training and 30% for validation.
   
2. 
   
3. unzip('MerchData.zip');
   
4. images = imageDatastore('MerchData','IncludeSubfolders',true,'LabelSource','foldernames');
   
5. [trainingImages,validationImages] = splitEachLabel(images,0.7,'randomized');

复制代码

1. Load the pretrained AlexNet network. If Neural Network Toolbox™ Model for AlexNet Network is not installed, then the software provides a download link. AlexNet is trained on more than one million images and can classify images into 1000 object categories.
   
2. 
   
3. net = alexnet;

复制代码

1. To retrain AlexNet to classify new images, replace the last three layers of the network. Set the final fully connected layer to have the same size as the number of classes in the new data set (5, in this example). To learn faster in the new layers than in the transferred layers, increase the learning rate factors of the fully connected layer.
   
2. 
   
3. layersTransfer = net.Layers(1:end-3);
   
4. numClasses = numel(categories(trainingImages.Labels));
   
5. layers = [
   
6.     layersTransfer
   
7.     fullyConnectedLayer(numClasses,'WeightLearnRateFactor',20,'BiasLearnRateFactor',20)
   
8.     softmaxLayer
   
9.     classificationLayer];

复制代码

1. options = trainingOptions('sgdm',...
   
2.     'MiniBatchSize',10,...
   
3.     'MaxEpochs',4,...
   
4.     'InitialLearnRate',1e-4,...
   
5.     'Verbose',false,...
   
6.     'Plots','training-progress',...
   
7.     'ValidationData',validationImages,...
   
8.     'ValidationFrequency',5);

复制代码

1. Train the network using the training data.
   
2. 
   
3. netTransfer = trainNetwork(trainingImages,layers,options);

复制代码

Transfer Learning Using AlexNet

1. #Load Data
   
2. 
   
3. unzip('MerchData.zip');
   
4. images = imageDatastore('MerchData',...
   
5.     'IncludeSubfolders',true,...
   
6.     'LabelSource','foldernames');

复制代码

1. #Divide the data into training and validation data sets. Use 70% of the images for training and 30% for validation. splitEachLabel splits the images datastore into two new datastores.
   
2. 
   
3. [trainingImages,validationImages] = splitEachLabel(images,0.7,'randomized');

复制代码

1. #Display some sample images.
   
2. 
   
3. numTrainImages = numel(trainingImages.Labels);
   
4. idx = randperm(numTrainImages,16);
   
5. figure
   
6. for i = 1:16
   
7.     subplot(4,4,i)
   
8.     I = readimage(trainingImages,idx(i));
   
9.     imshow(I)
   
10. end

复制代码

1. #Load Pretrained Network
   
2. net = alexnet;
   
3. #Display the network architecture. The network has five convolutional layers and three fully connected layers.
   
4. 
   
5. net.Layers

复制代码

1. #Transfer Layers to New Network. The last three layers of the pretrained network net are configured for 1000 classes. These three layers must be fine-tuned for the new classification problem. Extract all layers, except the last three, from the pretrained network.
   
2. 
   
3. layersTransfer = net.Layers(1:end-3);
   
4. 
   
5. numClasses = numel(categories(trainingImages.Labels))
   
6. layers = [
   
7.     layersTransfer
   
8.     fullyConnectedLayer(numClasses,'WeightLearnRateFactor',20,'BiasLearnRateFactor',20)
   
9.     softmaxLayer
   
10.     classificationLayer];

复制代码

1. #Train Network
   
2. 
   
3. miniBatchSize = 10;
   
4. numIterationsPerEpoch = floor(numel(trainingImages.Labels)/miniBatchSize);
   
5. options = trainingOptions('sgdm',...
   
6.     'MiniBatchSize',miniBatchSize,...
   
7.     'MaxEpochs',4,...
   
8.     'InitialLearnRate',1e-4,...
   
9.     'Verbose',false,...
   
10.     'Plots','training-progress',...
    
11.     'ValidationData',validationImages,...
    
12.     'ValidationFrequency',numIterationsPerEpoch);

复制代码

1. netTransfer = trainNetwork(trainingImages,layers,options);

复制代码

谢谢楼主分享！

Transfer Learning Using GoogLeNet

1. unzip('MerchData.zip');
   
2. images = imageDatastore('MerchData','IncludeSubfolders',true,'LabelSource','foldernames');
   
3. images.ReadFcn = @(loc)imresize(imread(loc),[224,224]);
   
4. [trainImages,valImages] = splitEachLabel(images,0.7,'randomized');

复制代码

1. net = googlenet;

复制代码

1. #Extract the layer graph from the trained network and plot the layer graph.
   
2. lgraph = layerGraph(net);
   
3. figure('Units','normalized','Position',[0.1 0.1 0.8 0.8]);
   
4. plot(lgraph)

复制代码

1. #To retrain GoogLeNet to classify new images, replace the last three layers of the network. These three layers of the network, with the #names 'loss3-classifier', 'prob', and 'output', contain the information of how to combine the features that the network extracts into class #probabilities and labels. Add three new layers, a fully connected layer, a softmax layer, and a classification output layer, to the layer #graph. Set the final fully connected layer to have the same size as the number of classes in the new data set (5, in this example). To #learn faster in the new layers than in the transferred layers, increase the learning rate factors of the fully connected layer.
   
2. 
   
3. lgraph = removeLayers(lgraph, {'loss3-classifier','prob','output'});
   
4. 
   
5. numClasses = numel(categories(trainImages.Labels));
   
6. newLayers = [
   
7.     fullyConnectedLayer(numClasses,'Name','fc','WeightLearnRateFactor',20,'BiasLearnRateFactor', 20)
   
8.     softmaxLayer('Name','softmax')
   
9.     classificationLayer('Name','classoutput')];
   
10. lgraph = addLayers(lgraph,newLayers);

复制代码

1. #Connect the last of the transferred layers remaining in the network ('pool5-drop_7x7_s1') to the new layers. To check that the new layers #are correctly connected, plot the new layer graph and zoom in on the last layers of the network.
   
2. 
   
3. lgraph = connectLayers(lgraph,'pool5-drop_7x7_s1','fc');
   
4. 
   
5. figure('Units','normalized','Position',[0.3 0.3 0.4 0.4]);
   
6. plot(lgraph)
   
7. ylim([0,10])

复制代码

1. #Specify the training options, including learning rate, mini-batch size, and validation data.
   
2. options = trainingOptions('sgdm',...
   
3.     'MiniBatchSize',10,...
   
4.     'MaxEpochs',3,...
   
5.     'InitialLearnRate',1e-4,...
   
6.     'VerboseFrequency',1,...
   
7.     'ValidationData',valImages,...
   
8.     'ValidationFrequency',3);

复制代码

1. #Train the network using the training data.
   
2. net = trainNetwork(trainImages,lgraph,options);

复制代码

1. #Classify the validation images using the fine-tuned network, and calculate the classification accuracy.
   
2. predictedLabels = classify(net,valImages);
   
3. accuracy = mean(predictedLabels == valImages.Labels)

复制代码

1. #Load Data
   
2. unzip('MerchData.zip');
   
3. images = imageDatastore('MerchData',...
   
4.     'IncludeSubfolders',true,...
   
5.     'LabelSource','foldernames');
   
6. [trainingImages,testImages] = splitEachLabel(images,0.7,'randomized');

复制代码

1. #Display some sample images.
   
2. numTrainImages = numel(trainingImages.Labels);
   
3. idx = randperm(numTrainImages,16);
   
4. figure
   
5. for i = 1:16
   
6. subplot(4,4,i)
   
7. I = readimage(trainingImages,idx(i));
   
8. imshow(I)
   
9. end

复制代码

1. #Load Pretrained Network
   
2. net = alexnet;
   
3. #Display the network architecture.
   
4. net.Layers

复制代码

1. #Extract the class labels from the training and test data.
   
2. trainingLabels = trainingImages.Labels;
   
3. testLabels = testImages.Labels;

复制代码

1. #Use the features extracted from the training images as predictor variables and fit a multiclass support vector machine (SVM) using #fitcecoc (Statistics and Machine Learning Toolbox).
   
2. classifier = fitcecoc(trainingFeatures,trainingLabels);

复制代码

1. #Classify the test images using the trained SVM model the features extracted from the test images.
   
2. predictedLabels = predict(classifier,testFeatures);

复制代码

1. #Display four sample test images with their predicted labels.
   
2. idx = [1 5 10 15];
   
3. figure
   
4. for i = 1:numel(idx)
   
5.     subplot(2,2,i)
   
6.     I = readimage(testImages,idx(i));
   
7.     label = predictedLabels(idx(i));
   
8.     imshow(I)
   
9.     title(char(label))
   
10. end

复制代码

Neural Network Toolbox User's Guide 2017