clear
close all

load('ArabicHC.mat');

imageAugmenter = imageDataAugmenter( ...
    'RandRotation',[-1.5,1.5] ,...
    'RandXTranslation',[-3,3] ,...
    'RandYTranslation',[-3,3] ,...
    'RandXScale',[0.9,1.1],...
    'RandYScale',[0.9,1.1]); 
imageSize = [32,32,1];
auimds = augmentedImageSource(imageSize,XTrain,TTrain,'DataAugmentation',imageAugmenter);
    
%% Define the Network Layers

layers = [
    imageInputLayer([32 32 1],'Normalization','none','Name','input')
    convolution2dLayer(5,64,'Name','conv1')
    batchNormalizationLayer('Name','BN1')
    reluLayer('Name','relu1')
% 28x28
    convolution2dLayer(5,128,'Name','conv2')
    batchNormalizationLayer('Name','BN2')
    reluLayer('Name','relu2')
% 24x24
    convolution2dLayer(5,176,'Name','conv3')
    batchNormalizationLayer('Name','BN3')
    reluLayer('Name','relu3')
% 20x20
    maxPooling2dLayer(2,'Stride',2,'Name','max1')
% 10x10 
    convolution2dLayer(5,208,'Name','conv4')
    batchNormalizationLayer('Name','BN4')
    reluLayer('Name','relu4')
% 6x6  
    convolution2dLayer(5,240,'Name','conv5')
    batchNormalizationLayer('Name','BN5')
    reluLayer('Name','relu5')
% 2x2
    maxPooling2dLayer(2,'Stride',2,'Name','max2')

    fullyConnectedLayer(28,'Name','fc')
    softmaxLayer('Name','softmax')
    classificationLayer('Name','classOutput')];

%% Specify the Training Options

options = trainingOptions('sgdm','LearnRateSchedule','piecewise',...
      'MiniBatchSize',192,'LearnRateDropFactor',0.998,... 
      'LearnRateDropPeriod',2,... 
      'MaxEpochs',3200, ...
	'InitialLearnRate',0.002462,...
    'Shuffle','every-epoch');  

   convnet = trainNetwork(auimds,layers,options);

YTest = classify(convnet,XTest);
accuracy = sum(YTest == TTest)/numel(TTest);
Error=(1-accuracy)*100