SeriesNetwork
深度学习系列网络
描述
系列网络是一个用于深度学习的神经网络，其层面是一个又一次安排的。它具有单个输入层和一个输出层。
Creation
There are several ways to create aSeriesNetworkobject:
Load a pretrained network using
alexnet,darknet19,VGG16, orVGG19。例如，请参阅负载预告额的Alexnet卷积神经网络。Train or fine-tune a network using
火车网。例如，请参阅Train Network for Image Classification。从Tensorflow™-KERAS，CAFFE或ONNX™（开放神经网络交换）模型格式导入预计的网络。
For a Keras model, use
importKerasNetwork。例如，请参阅Import and Plot Keras Network。For a Caffe model, use
importCaffeNetwork。例如，请参阅Import Caffe Network。For an ONNX model, use
importONNXNetwork。例如，请参阅导入onnx网络作为dagnetwork。
使用验证的层中组装一个深度学习网络
assembleNetworkfunction.
Note
To learn about other pretrained networks, such asGooglenetandresnet50, seePretrained Deep Neural Networks。
Properties
对象功能
activations |
计算深度学习网络层激活 |
classify |
使用经过训练的深度学习神经网络对数据进行分类 |
predict |
Predict responses using trained deep learning neural network |
predictAndUpdateState |
使用训练有素的复发神经网络预测响应，并更新网络状态 |
classifyAndUpDateState |
使用训练有素的复发神经网络对数据进行分类，并更新网络状态 |
resetState |
Reset state parameters of neural network |
阴谋 |
Plot neural network layer graph |
例子
负载预告额的Alexnet卷积神经网络
Load a pretrained AlexNet convolutional neural network and examine the layers and classes.
Load the pretrained AlexNet network usingalexnet。The outputnet是一个SeriesNetworkobject.
Net = Alexnet
net = SeriesNetwork with properties: Layers: [25×1 nnet.cnn.layer.Layer]
Using theLayersproperty, view the network architecture. The network comprises of 25 layers. There are 8 layers with learnable weights: 5 convolutional layers, and 3 fully connected layers.
net.Layers
ans = 25x1层阵列带有图层：1'数据'图像输入227x227x3图像，带有“ zerecenter”标准化2'Conv1'卷积96 11x11x3卷积，步幅[4 4]和Padding [0 0 0 0 0 0] 3'relu1'relu relu 4 relu 4 relu 4“ NORM1”跨通道归一化横向通道归一化，每元素5通道5'pool1'最大boming 3x3 max boming plid [2 2]和填充[0 0 0 0 0] 6'conv2'分组卷积2组128 5x5x48卷积，步幅[1 1]和填充[2 2 2] 7'RERU2'RELU 2'relu 8'norm2'跨通道横向通道横向通道归一化，每个元素9'pool2'最大池2'最大boming 3x3 max bol [2 2]和填充[0 0 0 0] 10'Conv3'卷积384 3X3X256大步[1 1]和填充[1 1 1 1] 11'RELU3'RELU3'RELU RELU RELU 12'CORV4'分组的卷积2组192 3x3x192卷积，横步[1 1]和填充[1 1 1 1] 13'relu4'relu relu 14'conv5'分组2组128 3x3x192卷积[1 1]和填充[1 1 1 1] 15'relu5'relu relu 16'pool5'最大池池3x3最大池，步幅[2 2]和填充[0 0 0 0 0] 17'fc6'FC6'完全连接4096完全连接第18层'relu6'relu 19'drop6'辍学50％辍学20'fc7'完全连接4096完全连接的层21'relu7'relu 22'relu 22'drop7 drop7 drop7'辍学50％辍学23'fc8 fc8完全连接1000完全连接的层24'Prob'SoftMax SoftMax 25“输出”分类'带有“ Tench”和999个其他类
You can view the names of the classes learned by the network by viewing theClasses分类输出层的属性（最后一层）。通过选择前10个元素查看前10个类。
net.layers（end）.classes（1:10）
ans =10×1分类阵列Tench Goldfish Great White Shark Tiger Shark Hammerhead Electric Ray stingray公鸡母鸡鸵鸟
来自CAFFE网络的导入层
Specify the example file'digitsnet.prototxt'进口。
Protofile ='digitsnet.prototxt';
导入网络层。
layers = ImportCaffelayers（Protofile）
层=带有层的1x7层阵列：1'testdata'图像输入28x28x1图像2'conv1'卷积20 5x5x1卷积，步幅[1 1]和填充[0 0] 3'relu1'relu1'relu 4'relu 4'pool1 pool1 pool 1'max ploming 2x2 max max 2x2 max 2x2与步幅[2 2]合并[0 0] 5'IP1'完全连接10完全连接的第6层'损失'softmax softmax 7'输出'分类'分类'分类crossentropyex搭配'class1'，'class2'和其他8个类别
Train Network for Image Classification
Load the data as anImageDatastoreobject.
digitDatasetPath = fullfile(matlabroot,'toolbox','nnet',。。。'nndemos','nndatasets',“ digitdataset”）；imds = imageDatastore(digitDatasetPath,。。。'IncludeSubfolders',true,。。。'LabelSource','foldernames'）；
数据存储包含10,000个数字的合成图像，从0到9。图像是通过应用随机转换来创建不同字体的数字图像来生成的。每个数字图像为28乘28像素。数据存储包含每个类别的图像数量相等。
Display some of the images in the datastore.
figure numImages = 10000; perm = randperm(numImages,20);fori = 1:20 subplot(4,5,i); imshow(imds.Files{perm(i)}); drawnow;结尾
Divide the datastore so that each category in the training set has 750 images and the testing set has the remaining images from each label.
numTrainingFiles = 750; [imdsTrain,imdsTest] = splitEachLabel(imds,numTrainingFiles,'randomize'）；
SpliteachLabelsplits the image files indigitDatainto two new datastores,Imdsrainandimdstest。
定义卷积神经网络体系结构。
层= [。。。imageInputLayer([28 28 1]) convolution2dLayer(5,20) reluLayer maxPooling2dLayer(2,'Stride',2) fullyConnectedLayer(10) softmaxLayer classificationLayer];
Set the options to the default settings for the stochastic gradient descent with momentum. Set the maximum number of epochs at 20, and start the training with an initial learning rate of 0.0001.
options = trainingOptions('sgdm',。。。'MaxEpochs',20,。。。“初始删除”,1e-4,。。。“冗长”,false,。。。'Plots','training-progress'）；
Train the network.
net = trainNetwork(imdsTrain,layers,options);
Run the trained network on the test set, which was not used to train the network, and predict the image labels (digits).
YPred =分类(净,imdsTest);欧美= imdsTest.Labels;
计算准确性。精度是测试数据中真实标签数的比率，该标签与从classify对于测试数据中的图像数量。
accuracy = sum(YPred == YTest)/numel(YTest)
accuracy = 0.9404
Extended Capabilities
Version History
You can also select a web site from the following list:
美洲
- América Latina(Español)
- 加拿大（英语）
- 美国（英语）