Matlab官网Deep Learning Onramp课程学习

一、体验网络

deepnet = alexnet %利用库内pretrained alexnet

classify(deepnet,img) %使用alexnet来分类当前的图片

二、使用预训练的网络：使用预先创建和训练的网络进行图像分类

1.做出预测
 

deepnet = alexnet;
pred= classify(deepnet ,img);

2.CNN架构

​​​​​​​deepnet = alexnet;
ly=deepnet.Layers
inlayer=ly(1)
insz=inlayer.InputSize
outlayer=ly(end)
categorynames=outlayer.Classes

3.检查预测结果

img = imread('file01.jpg');
imshow(img)
net = alexnet;
categorynames = net.Layers(end).ClassNames;

%Classify an image
[pred,scores] = classify(net,img);

%Display scores
bar(scores)

%Threshold scores
highscores = scores > 0.01;

%Display thresholded scores
bar(scores(highscores))%提取出大于阈值的分数并显示

%Add tick labels
xticklabels(categorynames(highscores))


thresh = median(scores) + std(scores);%中值+ 标准差
highscores = scores > thresh;

xticks(1:length(scores(highscores)))%划出有多少个刻度值
xticklabels(categorynames(highscores))%标出该刻度值的标签
xtickangle(60)%标签显示角度

三、管理数据集：导入图像文件夹并使其可用于给定网络

1.图像数据库存储

ls *.jpg
net = alexnet;

%Create datastore
imds=imageDatastore("file*.jpg");

%Extract file names
fname=imds.Files

%Read an image
img=readimage(imds,7)

%Classify images
[preds,scors]=classify(net,imds)

%max(scors,[],2)
[C,labels]=max(scores,[],2)   %C为预测的最大值，labels 为预测的种类标签  长度= 文件数*1
bar(C)

2.准备要作为输入的图像

ls *.jpg
net = alexnet

%Create datastore
imds=imageDatastore('*.jpg')

%Create augmentedImageDatastore
auds = augmentedImageDatastore([227 227],imds)

%Classify datastore
preds=classify(net,auds)

3.处理数据存储中的图像

ls *.jpg
net = alexnet
imds=imageDatastore('*.jpg')

% Display images in imds
montage(imds)

% Create augmentedImageDatastore
auds = augmentedImageDatastore([227 227],imds,'ColorPreprocessing','gray2rgb')

% Classify datastore
preds=classify(net,auds)

4.使用子文件夹创建数据存储

net = alexnet;
% Create datastore
flwrds = imageDatastore('Flowers','IncludeSubfolders',true)

% Classify images
preds=classify(net,flwrds)

四、进行迁移学习：使用新数据修改预先训练的网络并将图像划分为新类

1.什么是迁移学习

2.迁移学习所需的组件

load pathToImages
flwrds = imageDatastore(pathToImages,'IncludeSubfolders',true);
flowernames = flwrds.Labels %获取标签

%Create datastore with labels
flwrds = imageDatastore(pathToImages,...
'IncludeSubfolders',true,...
'LabelSource','foldernames')

% Extract new labels
flowernames=flwrds.Labels

3.准备训练数据

load pathToImages
flwrds = imageDatastore(pathToImages,'IncludeSubfolders',true,'LabelSource','foldernames')

% Split datastore
[flwrTrain,flwrTest] = splitEachLabel(flwrds,0.6)

%Split datastore randomly
[flwrTrain,flwrTest] = splitEachLabel(flwrds,0.8,'randomized')

%Split datastore by number of images
[flwrTrain,flwrTest] = splitEachLabel(flwrds,50)

4.修改网络层次

anet = alexnet;
layers = anet.Layers

%Create new layer
fc = fullyConnectedLayer(12)

%Replace 23rd layer
layers(23) = fc

%Replace last layer
layers(end) = classificationLayer 
%You can use the classificationLayer function to create a new output layer for an image %classification network.
%例中我们要识别12种花，我们要将第23层有1000个单元的神经层替换成12个单元的全连接层。再将最后一
%层1000个类别的输出层换成我们的类别输出层

5.设置训练选项

%Set default options
opts = trainingOptions('sgdm')

%Set initial learning rate
opts = trainingOptions('sgdm','InitialLearnRate',0.001)

6.训练网络

7.评估性能

8.迁移学习小结

Get training images
flowerds = imageDatastore('Flowers','IncludeSubfolders',true,'LabelSource','foldernames');
Split into training and testing sets
[trainImgs,testImgs] = splitEachLabel(flowerds,0.6);
Determine the number of flower species
numClasses = numel(categories(flowerds.Labels));

Create a network by modifying AlexNet
Get the layers from AlexNet
net = alexnet;
layers = net.Layers;
Modify the classification and output layers
layers(end-2) = fullyConnectedLayer(numClasses);
layers(end) = classificationLayer;

Set training algorithm options
Lower the learning rate for transfer learning
options = trainingOptions('sgdm','InitialLearnRate', 0.001);

Perform training
[flowernet,info] = trainNetwork(trainImgs, layers, options);

Use the trained network to classify test images
testpreds = classify(flowernet,testImgs);

Evaluate the results
Calculate the accuracy
nnz(testpreds == testImgs.Labels)/numel(testpreds)
Visualize the confusion matrix
confusionchart(testImgs.Labels,testpreds);