基于深度学习的数字识别GUI的设计

用matlab的deeplearning工具箱搭建了CNN来识别手写数字的GUI。

一.训练CNN

采用的是matlab自带的数字训练集和验证集，搭建的CNN的代码如下：

clc,clear;
digitalDatasetPath = fullfile(matlabroot,'toolbox','nnet','nndemos','nndatasets','DigitDataset');
imds = imageDatastore(digitalDatasetPath,'IncludeSubfolders',true,'LabelSource','foldernames');
figure;
perm = randperm(10000,20);
for i = 1:20
    subplot(4,5,i);
    imshow(imds.Files{perm(i)});
end
labelCount = countEachLabel(imds);
img = readimage(imds,1);
[m,n] = size(img);
numTrainFiles = 750;
[imdsTrain,imdsValidation] = splitEachLabel(imds,numTrainFiles,'randomize');
%定义CNN架构
layers = [
    imageInputLayer([m n 1])
    
    convolution2dLayer(3,8,'Padding','same')
    batchNormalizationLayer
    reluLayer
    
    maxPooling2dLayer(2,'Stride',2)
    
    convolution2dLayer(3,32,'Padding','same')
    batchNormalizationLayer
    reluLayer
    
    fullyConnectedLayer(10)%输出类别有10个类
    softmaxLayer           %对全连接层的输出归一化
    classificationLayer];  %分类层

options = trainingOptions('sgdm',... %求解器参数设置
    'InitialLearnRate',0.01,...
    'MaxEpochs',4,...
    'Shuffle','every-epoch',...%每一轮都需要验证
    'ValidationData',imdsValidation,...%验证的数据集
    'ValidationFrequency',30,...      
    'Verbose',false,...
    'Plots','training-progress');

net = trainNetwork(imdsTrain,layers,options);
save('CNNstructure','net');

%验证的图像分类并且计算准确度
YPred = classify(net,imdsValidation);
YValidation = imdsValidation.Labels;
accuracy = sum(YPred == YValidation)/numel(YValidation)

训练的过程如下图所示：

二.结果测试

将训练好的神经网络存储于当前文件夹：

并且用以下代码去测试网络的准确性，当然输入的图像必须是$28\times 28$像素的图(因为训练用的图是$28\times 28$像素的图)。

clear;
load CNNstructure.mat;
[filename pathname filterindex] = uigetfile(...
    {'*.png','图像文件(*.png)';...
    '*.jpg','图像文件(*.jpg)';...
    '*.*','所有文件(*.*)'},...
    '选择图像文件','MultiSelect','off',...
    pwd);
filePath = 0;
if isequal(filename,0)||isequal(pathname,0) %只要这里面表示返回或没有合适的文件时候
    return;
end
filePath = fullfile(pathname,filename);
im = imread(filePath);
figure(1)
imshow(im(:,:,1));
figure(2)
prediction = classify(net,imresize(im(:,:,1),[28 28]));
imshow(im(:,:,1));
title(char(prediction));

三.搭建GUI

搭建的框架如下图所示：

导入图像按钮的回调函数:

% --- Executes on button press in import.
function import_Callback(hObject, eventdata, handles)
% hObject    handle to import (see GCBO)
% eventdata  reserved - to be defined in a future version of MATLAB
% handles    structure with handles and user data (see GUIDATA)
[filename pathname filterindex] = uigetfile(...
    {'*.png','图像文件(*.png)';...
    '*.jpg','图像文件(*.jpg)';...
    '*.*','所有文件(*.*)'},...
    '选择图像文件','MultiSelect','off',...
    pwd);
if isequal(filename,0)||isequal(pathname,0) %只要这里面表示返回或没有合适的文件时候
    return;
end
filePath = fullfile(pathname,filename);
im = imread(filePath);
handles.figure = imresize(im(:,:,1));
guidata(hObject, handles);
imshow(im);

开始识别的回调函数

% --- Executes on button press in test.
function test_Callback(hObject, eventdata, handles)
% hObject    handle to test (see GCBO)
% eventdata  reserved - to be defined in a future version of MATLAB
% handles    structure with handles and user data (see GUIDATA)
load CNNstructure.mat;
prediction = classify(net,handles.figure);
imshow(handles.figure);
title(char(prediction));
set(handles.result,'String',char(prediction));

最后的结果显示: