opecv训练数字 0， 2（简单分类）

#include "stdafx.h"

#include <iostream>
//#include <opencv2/core/core.hpp>
//#include <opencv2/ml/ml.hpp>
#include <opencv2/opencv.hpp>
#include <io.h>
using namespace cv;
using namespace std;

#define CLASS_NUM 10

#define MAT_WEIGHT 16
#define MAT_HEIGHT 32

#define uchar unsigned char


void getFeature(Mat &X_train, Mat temp_img, int nIndex)
{
	cvtColor(temp_img, temp_img, CV_BGR2GRAY);
	//判断图像是否加载成功
	if (temp_img.empty())
	{
		cout << "图像加载失败" << endl;
		return ;
	}

	for (size_t i = 0; i < MAT_HEIGHT; i++)
	{
		for (size_t j = 0; j < MAT_WEIGHT; j++)
		{
			X_train.at<float>(nIndex, i*MAT_WEIGHT + j) = (float)temp_img.at<uchar>(i, j)/255;
		}

	}
	
}


 

int main() 
{
	int in_layer = 512;
	int out_layer = 2;
	int SAMPLES = 10;  // 训练样本总量 
	Mat X_train(SAMPLES, in_layer, CV_32FC1);
	Mat Y_train(SAMPLES, out_layer, CV_32FC1);

	// 构建训练数据矩阵 X_train
	Mat xtrain_0 = imread("data\\0\\1.jpg");
	getFeature(X_train, xtrain_0, 0);
	xtrain_0 = imread("data\\0\\2.jpg");
	getFeature(X_train, xtrain_0, 1);
	xtrain_0 = imread("data\\0\\3.jpg");
	getFeature(X_train, xtrain_0, 2);
	xtrain_0 = imread("data\\0\\4.jpg");
	getFeature(X_train, xtrain_0, 3);
	xtrain_0 = imread("data\\0\\5.jpg");
	getFeature(X_train, xtrain_0, 4);


	Mat xtrain_1 = imread("data\\1\\1.png"); 
	getFeature(X_train, xtrain_1, 5); 
	xtrain_1 = imread("data\\1\\2.png");
	getFeature(X_train, xtrain_1, 6);
	xtrain_1 = imread("data\\1\\3.png");
	getFeature(X_train, xtrain_1, 7);
	xtrain_1 = imread("data\\1\\4.png");
	getFeature(X_train, xtrain_1, 8);
	xtrain_1 = imread("data\\1\\5.png");
	getFeature(X_train, xtrain_1, 9);

	//string strPath = "data\\0\\";
	//for (int i = 0; i < 21; i++)
	//{
	//	string strPicPath = strPath + to_string(i) + ".jpg";
	//	Mat mat = imread(strPicPath);
	//	getFeature(X_train, mat,i);
	//}


	// 构建训练标签矩阵 Y_train
	for (int i = 0; i < SAMPLES; i++)
	{
		Y_train.at<float>(i, 0) = 1;
		Y_train.at<float>(i, 1) = 0;
		if (i > SAMPLES / 2 || i == SAMPLES / 2)
		{
			Y_train.at<float>(i, 0) = 0;
			Y_train.at<float>(i, 1) = 1;
		}
	}


	CvANN_MLP_TrainParams params(
		cvTermCriteria(CV_TERMCRIT_ITER + CV_TERMCRIT_EPS, 300, 0.000001),
		CvANN_MLP_TrainParams::BACKPROP,
		0.1,
		0.1);
	int n_W = MAT_WEIGHT * MAT_HEIGHT;
	Mat layers = (Mat_<int>(3, 1) << n_W, 24, 2);

	CvANN_MLP net(layers, CvANN_MLP::SIGMOID_SYM, 0.6667f, 1.7159f);

	net.train(X_train, Y_train, Mat(), Mat(), params);
	net.save("net.xml");

	int numTest = 3;
	Mat X_test(numTest, in_layer, CV_32FC1);
	Mat Y_test(numTest, out_layer, CV_32FC1);
	Mat testImg0 = imread("data\\0\\6.jpg"); 
	cvtColor(testImg0, testImg0, CV_BGR2GRAY);
	Mat testImg1 = imread("data\\1\\6.png");
	cvtColor(testImg1, testImg1, CV_BGR2GRAY);
	Mat testImg2 = imread("data\\1\\7.png");
	cvtColor(testImg2, testImg2, CV_BGR2GRAY);


	
	for (size_t i = 0; i < MAT_HEIGHT; i++)
	{
		for (size_t j = 0; j < MAT_WEIGHT; j++)
		{
			X_test.at<float>(0, i*MAT_WEIGHT + j) = (float)testImg0.at<uchar>(i, j) / 255;
			X_test.at<float>(1, i*MAT_WEIGHT + j) = (float)testImg1.at<uchar>(i, j) / 255;
			X_test.at<float>(2, i*MAT_WEIGHT + j) = (float)testImg2.at<uchar>(i, j) / 255;
		}

	}

	Mat predictions(Y_test.size(), CV_32F);
	net.predict(X_test, predictions);
	cout << predictions << endl;
	return 0;
}

 

最后得到的分类结果还不错：

 

上面用的训练样本比较少，每类只有5张图片，下面采用21张图片的：

 

#include "stdafx.h"

#include <iostream>
//#include <opencv2/core/core.hpp>
//#include <opencv2/ml/ml.hpp>
#include <opencv2/opencv.hpp>
#include <io.h>
using namespace cv;
using namespace std;

#define CLASS_NUM 10

#define MAT_WEIGHT 16
#define MAT_HEIGHT 32

#define uchar unsigned char


void getFeature(Mat &X_train, Mat temp_img, int nIndex)
{
	cvtColor(temp_img, temp_img, CV_BGR2GRAY);
	//判断图像是否加载成功
	if (temp_img.empty())
	{
		cout << "图像加载失败" << endl;
		return ;
	}

	for (size_t i = 0; i < MAT_HEIGHT; i++)
	{
		for (size_t j = 0; j < MAT_WEIGHT; j++)
		{
			X_train.at<float>(nIndex, i*MAT_WEIGHT + j) = (float)temp_img.at<uchar>(i, j)/255;
		}

	}
	
}


 

int main() 
{
	int in_layer = 512;
	int out_layer = 2;
	
	int SAMPLES_0 = 21;
	int SAMPLES_2 = 21;
	int SAMPLES = SAMPLES_0 + SAMPLES_2;  // 训练样本总量 

	Mat X_train(SAMPLES, in_layer, CV_32FC1);
	Mat Y_train(SAMPLES, out_layer, CV_32FC1);

	// 构建训练数据矩阵 X_train
	string strPath = "data\\0\\";
	for (int i = 0; i < 21; i++)
	{
		string strPicPath = strPath + to_string(i+1) + ".jpg";
		Mat mat = imread(strPicPath);
		getFeature(X_train, mat,i);
	}

	strPath = "data\\1\\";
	for (int i = 0; i < SAMPLES_2; i++)
	{
		string strPicPath = strPath + to_string(i + 1) + ".png";
		Mat mat = imread(strPicPath);
		getFeature(X_train, mat, i + SAMPLES_0);
	}


	// 构建训练标签矩阵 Y_train
	for (int i = 0; i < SAMPLES; i++)
	{
		if (i < SAMPLES_0)
		{
			Y_train.at<float>(i, 0) = 1;
			Y_train.at<float>(i, 1) = 0;
		}
		if (i > SAMPLES_0 || SAMPLES_0 == i)
		{
			Y_train.at<float>(i, 0) = 0;
			Y_train.at<float>(i, 1) = 1;
		}
	}


	CvANN_MLP_TrainParams params(
		cvTermCriteria(CV_TERMCRIT_ITER + CV_TERMCRIT_EPS, 300, 0.000001),
		CvANN_MLP_TrainParams::BACKPROP,
		0.1,
		0.1);
	int n_W = MAT_WEIGHT * MAT_HEIGHT;
	Mat layers = (Mat_<int>(3, 1) << n_W, 24, 2);

	CvANN_MLP net(layers, CvANN_MLP::SIGMOID_SYM, 0.6667f, 1.7159f);

	net.train(X_train, Y_train, Mat(), Mat(), params);
	net.save("net.xml");

	int numTest = 3;
	Mat X_test(numTest, in_layer, CV_32FC1);
	Mat Y_test(numTest, out_layer, CV_32FC1);
	Mat testImg0 = imread("data\\0\\6.jpg"); 
	cvtColor(testImg0, testImg0, CV_BGR2GRAY);
	Mat testImg1 = imread("data\\1\\6.png");
	cvtColor(testImg1, testImg1, CV_BGR2GRAY);
	Mat testImg2 = imread("data\\1\\7.png");
	cvtColor(testImg2, testImg2, CV_BGR2GRAY);


	
	for (size_t i = 0; i < MAT_HEIGHT; i++)
	{
		for (size_t j = 0; j < MAT_WEIGHT; j++)
		{
			X_test.at<float>(0, i*MAT_WEIGHT + j) = (float)testImg0.at<uchar>(i, j) / 255;
			X_test.at<float>(1, i*MAT_WEIGHT + j) = (float)testImg1.at<uchar>(i, j) / 255;
			X_test.at<float>(2, i*MAT_WEIGHT + j) = (float)testImg2.at<uchar>(i, j) / 255;
		}

	}

	Mat predictions(Y_test.size(), CV_32F);
	net.predict(X_test, predictions);
	cout << predictions << endl;
	return 0;
}