自用

clc;

clear;

close all;

warning off;

addpath(genpath(pwd));

rng('default');

 

% 数据集加载

Dataset = imageDatastore('Photograph', 'IncludeSubfolders', true, 'LabelSource', 'foldernames');

%数据集划分

[Training_Dataset, Validation_Dataset, Testing_Dataset] = splitEachLabel(Dataset, 0.8, 0.1, 0.1);

 

% 加载预训练的 GoogLeNet

load googlenet.mat

analyzeNetwork(net);

 

% 调整输入层大小

Input_Layer_Size = net.Layers(1).InputSize(1:2);

 

% 调整数据集大小

Resized_Training_Dataset = augmentedImageDatastore(Input_Layer_Size, Training_Dataset);

Resized_Validation_Dataset = augmentedImageDatastore(Input_Layer_Size, Validation_Dataset);

Resized_Testing_Dataset = augmentedImageDatastore(Input_Layer_Size, Testing_Dataset);

 

% 修改网络架构

Feature_Learner = net.Layers(142).Name;

Output_Classifier = net.Layers(144).Name;

 

Number_of_Classes = numel(categories(Training_Dataset.Labels));

 

New_Feature_Learner = fullyConnectedLayer(Number_of_Classes, ...

'Name', 'Coal Feature Learner', ...

'WeightLearnRateFactor', 10, ...

'BiasLearnRateFactor', 10);

 

New_Classifier_Layer = classificationLayer('Name', 'Coal Classifier');

 

Network_Architecture = layerGraph(net);

 

New_Network = replaceLayer(Network_Architecture, Feature_Learner, New_Feature_Learner);

New_Network = replaceLayer(New_Network, Output_Classifier, New_Classifier_Layer);

 

%深度学习网络分析器界面

analyzeNetwork(New_Network);

 

% 训练深度学习网络

maxEpochs = 1;

Minibatch_Size = 8;

Validation_Frequency = floor(numel(Resized_Training_Dataset.Files) / Minibatch_Size);

%启用训练进程图

Training_Options = trainingOptions('sgdm', ...

'MiniBatchSize', Minibatch_Size, ...

'MaxEpochs', maxEpochs, ...

'InitialLearnRate', 1e-3, ...

'Shuffle', 'every-epoch', ...

'ValidationData', Resized_Validation_Dataset, ...

'ValidationFrequency', Validation_Frequency, ...

'Verbose', false, ...

'Plots', 'training-progress');

 

net = trainNetwork(Resized_Training_Dataset, New_Network, Training_Options);

 

% 保存训练好的深度学习网络

save gnet.mat net;

 

% 提取特征

featureLayer = 'pool5-7x7_s1'; % 选择一个合适的特征提取层

features = activations(net, Resized_Training_Dataset, featureLayer, 'MiniBatchSize', 16);

 

% 初始化标签数组

labels = Training_Dataset.Labels; % 从原始数据集中获取标签

 

% 将特征和标签转换为适合 SVM 的格式

% 压缩特征维度

features = squeeze(mean(features, [2, 3]));

% 将标签转换为 categorical 类型

Y = categorical(labels);

X = features; % 特征

 

% 训练 SVM 分类器

SVMModel = fitcecoc(X, Y, 'Learners', 'svm', 'Coding', 'onevsall');

 

% 保存 SVM 分类器

save svmModel.mat SVMModel;

disp('SVM 分类器训练完成并已保存到 svmModel.mat 文件中。');

------------------------------------------------------------------------------------------

clc; clear; close all; warning off;

addpath(genpath(pwd));

rng('default');

 

% 加载模型

load gnet.mat; % 确保网络输入层为 224x224

load svmModel.mat;

 

% 数据集加载与预处理

Dataset = imageDatastore('Photograph', 'IncludeSubfolders', true, 'LabelSource', 'foldernames');

[Training_Dataset, Validation_Dataset, Testing_Dataset] = splitEachLabel(Dataset, 0.8, 0.1, 0.1);

 

% ------------------ 关键修正：添加图像尺寸调整 ------------------

inputSize = net.Layers(1).InputSize(1:2); % 获取网络输入尺寸 [224 224]

 

% 创建调整后的验证集和测试集

Resized_Val = augmentedImageDatastore(inputSize, Validation_Dataset, 'ColorPreprocessing', 'gray2rgb');

Resized_Test = augmentedImageDatastore(inputSize, Testing_Dataset, 'ColorPreprocessing', 'gray2rgb');

% -----------------------------------------------------------

 

% 验证集测试

featureLayer = 'pool5-7x7_s1'; % 确保层名正确

 

% 使用调整后的数据集

% 提取特征

features = activations(net, Resized_Val, featureLayer, 'MiniBatchSize', 16, 'OutputAs', 'channels');

features = squeeze(mean(features, [1 2]))'; % 压缩特征维度

disp(size(features)); % 查看特征矩阵的尺寸

labels = Validation_Dataset.Labels;

 

% 如果尺寸不符合要求，进行重新调整

features = reshape(features, [], size(features, 3))'; % 重新调整为适合 SVM 输入的形状

 

% 预测与显示

Predicted_Label = predict(SVMModel, features);

disp(['验证集准确率: ', num2str(mean(Predicted_Label == labels) * 100), '%']);

 

 

showResults(Testing_Dataset, Predicted_Label, '测试集识别结果');

 

% ------------------ 可视化函数 ------------------

function showResults(dataset, labels, titleStr)

index = randperm(numel(dataset.Files), 16);

figure('Name', titleStr, 'NumberTitle','off');

for i = 1:16

subplot(4,4,i);

I = readimage(dataset, index(i));

imshow(I);

title(string(labels(index(i))));

end

end