classification using CNN (trainNetwork)

2 次查看（过去 30 天）

Toqa Am 2019-11-20

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/492175-classification-using-cnn-trainnetwork

Why the validation accuracy of CNN using trainNetwork has been change in each rerun. It has been decreased from 89% to 39% in each re run the program.

Please could anyone answer me???

close all, clear all, clc;
output_folder=fullfile('datasets - Copy (2)','REMBRANDT') ;  %creat file path        
categories={'Grade_II - Copy','Grade_III - Copy','Grade_IV - Copy'};
 imds=imageDatastore((fullfile(output_folder,categories)),'FileExtensions','.dcm','ReadFcn',@(x) dicomread(x),'LabelSource','foldernames');
 [trainingset, testset]=splitEachLabel(imds, 0.8,0.2);
 labelCount = countEachLabel(imds);
 rng(1);
layers = [
    imageInputLayer([128 128 1])
    convolution2dLayer(3,8,'Padding','same')
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer(1,'Stride',1)
    
    convolution2dLayer(3,16,'Padding','same')
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer(1,'Stride',1)
    
    convolution2dLayer(3,32,'Padding','same')
    batchNormalizationLayer
    reluLayer
    
    fullyConnectedLayer(3)
    dropoutLayer
    softmaxLayer
    classificationLayer];
options = trainingOptions('sgdm', ...
    'InitialLearnRate',0.01, ...
    'MaxEpochs',40, ...
    'Shuffle','every-epoch', ...
    'ValidationData',testset, ...
    'ValidationFrequency',60, ...
    'Verbose',false, ...
    'Plots','training-progress');
 net = trainNetwork(trainingset,layers,options);