自编码器是早期的神经网络方法之一了，为便于了解各方法识别性能，我们首先用它进行识别。代码如下：
clear
load('JDPig_mlhmslbp_spyr.mat');
m = numel(classe_name);
n = length(y);
label = []
%one-hot编码，将每幅图像的数字类别标签变为只有1个值为0的向量，向量维数与类别个数相同
for i=1:n
label(:,i) = zeros(m,1);
label(y(i),i) = 1;
end
testImg = load('JDTest_mlhmslbp_spyr.mat');
hiddenSize = 10;
%训练自编码器
if ~exist('autoenc.mat')
autoenc1 = trainAutoencoder(X,hiddenSize,...
'L2WeightRegularization',0.001,...
'SparsityRegularization',4,...
'SparsityProportion',0.05,...
'DecoderTransferFunction','purelin');
features1 = encode(autoenc1,X);
fprintf('saving features1\n');
%存储自编码器结果
save('autoenc.mat','features1','autoenc1');
else
load('autoenc.mat','features1','autoenc1');
fprintf('loading features1\n');
end
%训练第二层自编码器
if ~exist('deepnetAutoenc.mat')
hiddenSize = 10;
autoenc2 = trainAutoencoder(features1,hiddenSize,...
'L2WeightRegularization',0.001,...
'SparsityRegularization',4,...
'SparsityProportion',0.05,...
'DecoderTransferFunction','purelin',...
'ScaleData',false);
features2 = encode(autoenc2,features1);
%训练softmax分类层
softnet = trainSoftmaxLayer(features2,label,'LossFunction','crossentropy');
%构造深度网络
deepnet = stack(autoenc1,autoenc2,softnet);
deepnet = train(deepnet,X,label);
%存储第二层自编码器和训练出的深度网络
save('deepnetAutoenc.mat','autoenc2','features2','softnet','deepnet');
fprintf('Saving features2\n');
else
load('deepnetAutoenc.mat','autoenc2','features2','softnet','deepnet');
fprintf('Loading fatures2\n');
end
%计算每幅图像对于各个类的得分
scores = deepnet(testImg.X);
fprintf('Testing images!\n');
load('testName.mat','imgName');
for i=1:length(scores)
for j=1:m
indImg((i-1)*m+j) = imgName(i);
plabel((i-1)*m+j) = j;
prob((i-1)*m+j) = scores(j,i);
end
end
%创建各图像属于每个类的概率表
T = table(indImg',plabel',prob');
%将概率表存为csv文件，用以网站上传并计算最终得分
writetable(T,'resAutoenc.csv');
fprintf('Image recognition finished!\n');
最后写入磁盘的csv文件在上传至京东金融网站后，网站会计算出得分。该网络结构简单，准确率一般，印象中得分在10左右。不过该方法在图像集上测试的结果，准确率也在95%以上。
