神经网络ANN_MLP识别数字
环境:win7+VS2012+OpenCV3
利用OpenCV3中的ANN_MLP神经网络,使用如图所示图片进行训练,实现简单的数字识别功能。
#include
#include
#include
#include
#include
using namespace cv;
using namespace ml;
int main()
{
==========================读取图片创建训练数据==============================
//将所有图片大小统一转化为8*16
const int imageRows = 8;
const int imageCols = 16;
//图片共有10类
const int classSum = 10;
//每类共50张图片
const int imagesSum = 50;
//每一行一个训练图片
float trainingData[classSum*imagesSum][imageRows*imageCols] = {{0}};
//训练样本标签
float labels[classSum*imagesSum][classSum]={{0}};
Mat src, resizeImg, trainImg;
for (int i = 0; i < classSum; i++)
{
//目标文件夹路径
std::string inPath = "E:\\image\\image\\charSamples\\";
char temp[256];
int k = 0;
sprintf_s(temp, "%d", i);
inPath = inPath + temp + "\\*.png";
//用于查找的句柄
long handle;
struct _finddata_t fileinfo;
//第一次查找
handle = _findfirst(inPath.c_str(),&fileinfo);
if(handle == -1)
return -1;
do
{
//找到的文件的文件名
std::string imgname = "E:/image/image/charSamples/";
imgname = imgname + temp + "/" + fileinfo.name;
src = imread(imgname, 0);
if (src.empty())
{
std::cout<<"can not load image \n"<<std::endl;
return -1;
}
//将所有图片大小统一转化为8*16
resize(src, resizeImg, Size(imageRows,imageCols), (0,0), (0,0), INTER_AREA);
threshold(resizeImg, trainImg,0,255,CV_THRESH_BINARY|CV_THRESH_OTSU);
for(int j = 0; jfloat)resizeImg.data[j];
}
// 设置标签数据
for(int j = 0;j < classSum; j++)
{
if(j == i)
labels[i*imagesSum + k][j] = 1;
else
labels[i*imagesSum + k][j] = 0;
}
k++;
} while (!_findnext(handle, &fileinfo));
Mat labelsMat(classSum*imagesSum, classSum, CV_32FC1,labels);
_findclose(handle);
}
//训练样本数据及对应标签
Mat trainingDataMat(classSum*imagesSum, imageRows*imageCols, CV_32FC1, trainingData);
Mat labelsMat(classSum*imagesSum, classSum, CV_32FC1, labels);
//std::cout<<"trainingDataMat: \n"<
//std::cout<<"labelsMat: \n"<
==========================训练部分==============================
Ptrmodel = ANN_MLP::create();
Mat layerSizes = (Mat_<int>(1,5)<128 ,128,128,classSum);
model->setLayerSizes(layerSizes);
model->setTrainMethod(ANN_MLP::BACKPROP, 0.001, 0.1);
model->setActivationFunction(ANN_MLP::SIGMOID_SYM, 1.0, 1.0);
model->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER | TermCriteria::EPS, 10000,0.0001));
Ptr trainData = TrainData::create(trainingDataMat, ROW_SAMPLE, labelsMat);
model->train(trainData);
//保存训练结果
model->save("E:/image/image/MLPModel.xml");
==========================预测部分==============================
//读取测试图像
Mat test, dst;
test = imread("E:/image/image/test.png", 0);;
if (test.empty())
{
std::cout<<"can not load image \n"<<std::endl;
return -1;
}
//将测试图像转化为1*128的向量
resize(test, test, Size(imageRows,imageCols), (0,0), (0,0), INTER_AREA);
threshold(test, test, 0, 255, CV_THRESH_BINARY|CV_THRESH_OTSU);
Mat_<float> testMat(1, imageRows*imageCols);
for (int i = 0; i < imageRows*imageCols; i++)
{
testMat.at<float>(0,i) = (float)test.at(i/8, i%8);
}
//使用训练好的MLP model预测测试图像
model->predict(testMat, dst);
std::cout<<"testMat: \n"<"\n" <<std::endl;
std::cout<<"dst: \n"<"\n"<<std::endl;
double maxVal = 0;
Point maxLoc;
minMaxLoc(dst, NULL, &maxVal, NULL, &maxLoc);
std::cout<<"测试结果:"<"置信度:"<100<<"%"<<std::endl;
imshow("test",test);
waitKey(0);
return 0;
}
从训练过程中可以发现,神经网络的训练时间较长;所以若每次进行识别时都进行训练的话,系统就很难进行实时检测。因此在训练完成时将训练好的模型以.xlm文件进行保存,以后进行图像识别时若没有新的训练样本加入,则可以直接读取训练好的模型进行测试。
#include
#include
#include
#include
#include
using namespace cv;
using namespace ml;
//利用训练完成的神经网络模型进行识别
int main()
{
//将所有图片大小统一转化为8*16
const int imageRows = 8;
const int imageCols = 16;
//读取训练结果
Ptr model = StatModel::load("E:/image/image/MLPModel.xml");
==========================预测部分==============================
//读取测试图像
Mat test, dst;
test = imread("E:/image/image/test.png", 0);;
if (test.empty())
{
std::cout<<"can not load image \n"<<std::endl;
return -1;
}
//将测试图像转化为1*128的向量
resize(test, test, Size(imageRows,imageCols), (0,0), (0,0), INTER_AREA);
threshold(test, test, 0, 255, CV_THRESH_BINARY|CV_THRESH_OTSU);
Mat_<float> testMat(1, imageRows*imageCols);
for (int i = 0; i < imageRows*imageCols; i++)
{
testMat.at<float>(0,i) = (float)test.at(i/8, i%8);
}
//使用训练好的MLP model预测测试图像
model->predict(testMat, dst);
std::cout<<"testMat: \n"<"\n"<<std::endl;
std::cout<<"dst: \n"<"\n"<<std::endl;
double maxVal = 0;
Point maxLoc;
minMaxLoc(dst, NULL, &maxVal, NULL, &maxLoc);
std::cout<<"测试结果:"<"置信度:"<100<<"%"<<std::endl;
imshow("test",test);
waitKey(0);
return 0;
}
相关链接:
训练代码&训练图片
感谢博主思维之际提供的图片数据。