1111111111111111111111

20191014初步深度學習網路
clear
folder=fullfile(pwd,'newg')%目前資料夾名稱
%檔案讀取
imgs=imageDatastore(folder,...
    "IncludeSubfolders",true,"LabelSource",...
    "foldernames","FileExtensions",{'.jpg','.png','.jfif','.svg','.jpeg'})

檔案讀取
N=length(imgs.Files)
for i=1:N
    img=imread(imgs.Files{i}); %讀取照片
    [m,n,k]=size(img);%取得照片長寬顏色
end
idx=randperm(N);
for i = 1:16
    subplot(4,4,i)
    img=imread(imgs.Files{idx(i)}); %讀取任意照片
    imshow(img)%顯示照片
end
labelCOunt = countEachLabel(imgs)

numTrainFIles=0.7;%70%的資料
[Train,Test] = splitEachLabel(imgs,numTrainFIles,'randomized');

labelTrainCount = countEachLabel(Train)%顯示統計Training類別數目
labelTestCount = countEachLabel(Test)%顯示統計TEST類別數目

網路模型建立
numofprediction=3;
layers = [
    imageInputLayer([m n k],'Name','inputs')

    convolution2dLayer(3,8,'Padding','same','Name','Conv1')
    batchNormalizationLayer('Name','batchNrom1')
    reluLayer('Name','relu')

    maxPooling2dLayer(2,'Stride',2,'Name','Maxpooling1')

    convolution2dLayer(3,16,'Padding','same','Name','Conv2')
    batchNormalizationLayer('Name','batchNrom2')
    reluLayer('Name','relu2')

    maxPooling2dLayer(2,'Stride',2,'Name','Maxpooling2')

    convolution2dLayer(3,32,'Padding','same','Name','Conv3')
    batchNormalizationLayer('Name','batchNrom3')
    reluLayer('Name','relu3')

    fullyConnectedLayer(256,'Name','Fullconnect')
    fullyConnectedLayer(128,'Name','Fullconnect2')
    fullyConnectedLayer(numofprediction,'Name','Fullconnectout')
    softmaxLayer('Name','softmaxLayer')
    classificationLayer("Name","classificationLayer")];

lgraph = layerGraph(layers);
plot(lgraph)

options = trainingOptions('adam', ...
    'InitialLearnRate',0.01, ...
    'MaxEpochs',5000, ...
    'Shuffle','every-epoch', ...
    'ExecutionEnvironment',"gpu",...
    'ValidationData',Test, ...
    'ValidationFrequency',20, ...
    'Verbose',false, ...
    'Plots','training-progress');

net = trainNetwork(Train,layers,options)