opencv SVM分类器模块的简单设计

煤矸石分类项目，提取的煤矸石灰度均值和灰度方差作为特征进行分类，SVM的简单代码如下，使用的二次封装的opencv库，在其他机器上运行将头文件和条件编译宏替换成opencv自己的就可以了

#include "sv.h"
#ifdef _DEBUG
#pragma comment(lib,"BoxCV100d.lib")
#pragma comment(lib,"SvRunTimed.lib")
#else
#pragma comment(lib,"BoxCV100.lib")
#pragma comment(lib,"SvRunTime.lib")
#endif

using namespace cv;
using namespace cv::ml;

int main(int argc, char** argv)
{
    int width = 720, height = 720;

    //三通道可视化窗口
    Mat image = Mat::zeros(height, width, CV_8UC3);
    
    int labels[11] = { 1,1,1,1,1,1,
                    -1,-1,-1,-1,-1 };
    Mat LabelsMat(10, 1, CV_32SC1, labels);

　　　 float TrainingData1[11][2] =
　　　{
     {291.4,252},{115.3,186},{277.7,695},{367.5,620},{352.4,645},{117.2,203},
　　  {117.2,129},{122.8,112},{116.4,175},{132.5,78},{106.3,125}
　　  };

    Mat TrainMat(10, 2, CV_32FC1, TrainingData);

    //创建分类器设置参数
    Ptr<SVM> model = SVM::create();
    model->setType(SVM::C_SVC);
    model->setKernel(SVM::LINEAR);

    //设置训练参数
    Ptr<TrainData> tData = TrainData::create(TrainMat, ROW_SAMPLE, LabelsMat);

    //训练分类器
    model->train(tData);

    Vec3b ColorGreen(0, 255, 0), ColorBlue(255, 0, 0);

    //显示svm决策边界
    for(int i=0;i<image.rows;++i)
        for (int j = 0; j < image.cols; ++j)
        {
            //生成测试数据
            Mat SampleMat = (Mat_<float>(1, 2) << j, i);

            //predict
            float response = model->predict(SampleMat);

            if (1 == response)
                image.at<Vec3b>(i, j) = ColorGreen;
            else
                if (-1 == response)
                    image.at<Vec3b>(i, j) = ColorBlue;
        }

    //绘图
    //显示结果
    int thickness = -1;
    int lineType = 8;

    //两类结果一个黑色一个白色
    Scalar c1 = Scalar::all(0);
    Scalar c2 = Scalar::all(255);

    for (int i = 0; i < LabelsMat.rows; ++i)
    {
        //取出每行指针
        const float* v = TrainMat.ptr<float>(i);
        Point pt = Point((int)v[0], (int)v[1]);
        if (1 == labels[i])
            circle(image, pt, 5, c1, thickness, lineType);
        else
            circle(image, pt, 5, c2, thickness, lineType);
    }

    imshow("SVM Simple Example", image);
    waitKey();
}

结果图

 

 

 

 

但是这个程序试将训练数据写死的，在项目中分类器是一个独立的模块，分类器也要达到更高的精度需要更多的训练数据，当训练数据大到一定程度时，全写在程序的数组里是不现实的，还有一种情况是更换分类任务时需要重新训练，这时就需要分类器能够自己读取训练数据。

 

 

 训练数据x1,x2用“，”隔开，标签用“#”隔开。首先进行文件读取，先读进来一行数据，然后解析读进来的字符串，用string.find(',')函数来找到“，”的位置，用string.find('#')来找到标记的位置。当训练数据中包含多种特征时（用一个逗号分隔不够用，需要用多个逗号），可以自己写一个循环，遍历这一行

flag="a";
position=0;
int i=1;
while((position=s.find(flag,position))!=string::npos)
{
    cout<<"position  "<<i<<" : "<<position<<endl;
    position++;
    i++;
}

这个代码以后有需要再加上，其中字符串操作参考https://www.cnblogs.com/lifexy/p/8642163.html

字符串读进来后进行分割，然后进行类型转换储存在vector或者数组中，string类型转数字类型可以使用模板函数，参考https://www.cnblogs.com/lyggqm/p/4562727.html

代码如下

Type stringToNum(const string& str)
{
    istringstream iss(str);
    Type num;
    iss >> num;
    return num;
}

从文件中读取训练数据的完整代码如下

#include<fstream>
#include<sstream>
#include<string>
#include<vector>
#include<iostream>

using namespace std;

template <class Type>
Type stringToNum(const string& str)
{
    istringstream iss(str);
    Type num;
    iss >> num;
    return num;
}

int main(void)
{
    
    ifstream in("com.txt");
    string s;
    //char* temp;
    
    vector<float> Feature1;
    vector<float> Feature2;
    vector<int> TrainLabel;

    while (getline(in, s))//逐行读取数据并存于s中，直至数据全部读取
    {
        //cout << s.c_str() << endl;
        int FeaIndex = s.find(',');
        int LabelIndex = s.find('#');
        string FeaString1(s, 0, FeaIndex);
        Feature1.push_back(stringToNum<float>(FeaString1));

        string FeaString2(s, FeaIndex+1, LabelIndex - FeaIndex-1);
        Feature2.push_back(stringToNum<float>(FeaString2));

        string Label(s, LabelIndex + 1, s.length() - LabelIndex - 1);
        TrainLabel.push_back(stringToNum<int>(Label));
        //cout << "Fea1: " << FeaString1 << endl;
        //cout << "Fea2: " << FeaString2 << endl;
        //cout << "Label: " << Label << endl << endl;
    }
    in.close();

    for (size_t i = 0; i < Feature1.size(); ++i)
    {
        cout << "Feature 1: " << Feature1[i] << endl;
        cout << "Feature 2: " << Feature2[i] << endl;
        cout << "Label: " << TrainLabel[i] << endl;
        cout << endl;
    }
    system("pause");
    return 0;
}

 

 训练数据从文件中读入时是以二维vector形式存储在内存中的，我将它转化为float的二维指针中

float** TrainingData = new float*[TLabel.size()];
    for (int i = 0; i < TLabel.size(); ++i)
        TrainingData[i] = new float[2];
    for (int i = 0; i < TLabel.size(); ++i)
    {
        TrainingData[i][0] = TData.Feature1[i]*10;
        TrainingData[i][1] = TData.Feature2[i]*100;
        /*cout << TrainingData[i][0] << endl << TrainingData[i][1] << endl;*/
    }

在构造训练数据时出现了问题，传入svm的数据需要是Mat格式，而Mat的构造函数无法将二维地址直接传入，所以会导致训练数据集没传入数据

 

 

 Mat赋值方法如下

Mat TrainMat1(TrainLabel.size(), 2, CV_32FC1);
    for (int i = 0; i < TrainLabel.size(); ++i)
    {
        TrainMat1.at<float>(i, 0) = TrainingData.Feature1[i] * 10;
        TrainMat1.at<float>(i, 1) = TrainingData.Feature2[i] * 100;
    }

下面是完整代码

#include "sv.h"
#include<vector>
#include<iostream>
#include<fstream>
#include<sstream>
#include<string>

#ifdef _DEBUG
#pragma comment(lib,"BoxCV100d.lib")
#pragma comment(lib,"SvRunTimed.lib")
#else
#pragma comment(lib,"BoxCV100.lib")
#pragma comment(lib,"SvRunTime.lib")
#endif

using namespace std;
using namespace cv;
using namespace cv::ml;


//------------------------------------------------------------------
// 训练数据
//------------------------------------------------------------------
typedef struct _TRAINING_DATA
{
    vector<float> Feature1;
    vector<float> Feature2;
}TRAINING_DATA;


//------------------------------------------------------------------
// 模板函数，转换类型
//------------------------------------------------------------------
template <class Type>
Type stringToNum(const string& str)
{
    istringstream iss(str);
    Type num;
    iss >> num;
    return num;
}



//------------------------------------------------------------------
// 从文件中读取训练数据，存放到向量中
//------------------------------------------------------------------
void ReadTrainingData(const string& Dir,TRAINING_DATA &TrainingData,vector<int> &TrainLabel)
{
    //打开文件
    ifstream TrainingDataFile(Dir);
    string FileLineData;

    //逐行读取数据并存于s中，直至数据全部读取
    while (getline(TrainingDataFile, FileLineData))
    {
        //寻找分隔符
        int FeaIndex = FileLineData.find(',');
        int LabelIndex = FileLineData.find('#');

        //读取x1
        string FeaString1(FileLineData, 0, FeaIndex);
        TrainingData.Feature1.push_back(stringToNum<float>(FeaString1));

        //读取x2
        string FeaString2(FileLineData, FeaIndex+1, LabelIndex - FeaIndex-1);
        TrainingData.Feature2.push_back(stringToNum<float>(FeaString2));

        //读取标签
        string Label(FileLineData, LabelIndex + 1, FileLineData.length() - LabelIndex - 1);
        TrainLabel.push_back(stringToNum<int>(Label));
    }
    //关闭文件
    TrainingDataFile.close();

    for (size_t i = 0; i < TrainLabel.size(); ++i)
    {
        cout << "Feature 1: " << TrainingData.Feature1[i] << endl;
        cout << "Feature 2: " << TrainingData.Feature2[i] << endl;
        cout << "Label: " << TrainLabel[i] << endl;
        cout << endl;
    }
}


//------------------------------------------------------------------
// 训练SVM分类器
//------------------------------------------------------------------
void TrainSVM(Ptr<SVM> &Model, TRAINING_DATA &TrainingData, vector<int> &TrainLabel)
{
    //创建显示窗口
    int width = 720, height = 720;
    //三通道可视化窗口
    Mat image = Mat::zeros(height, width, CV_8UC3);

    int* labels = new int[TrainLabel.size()];
    for (int i = 0; i < TrainLabel.size(); ++i)
    {
        labels[i] = TrainLabel[i];
        cout << labels[i] << endl;
    }
    Mat LabelsMat(TrainLabel.size(), 1, CV_32SC1, labels);


    Mat TrainMat1(TrainLabel.size(), 2, CV_32FC1);
    for (int i = 0; i < TrainLabel.size(); ++i)
    {
        TrainMat1.at<float>(i, 0) = TrainingData.Feature1[i] * 10;
        TrainMat1.at<float>(i, 1) = TrainingData.Feature2[i] * 100;
    }

    //创建分类器设置参数
    Ptr<SVM> model = SVM::create();
    model->setType(SVM::C_SVC);
    model->setKernel(SVM::LINEAR);

    //设置训练参数
    Ptr<TrainData> tData = TrainData::create(TrainMat1, ROW_SAMPLE, LabelsMat);

    //训练分类器
    model->train(tData);

    Vec3b ColorGreen(0, 255, 0), ColorBlue(255, 0, 0);

    //显示svm决策边界
    for (int i = 0; i < image.rows; ++i)
        for (int j = 0; j < image.cols; ++j)
        {
            //生成测试数据
            Mat SampleMat = (Mat_<float>(1, 2) << j, i);

            //predict
            float response = model->predict(SampleMat);

            if (1 == response)
                image.at<Vec3b>(i, j) = ColorGreen;
            else
                if (-1 == response)
                    image.at<Vec3b>(i, j) = ColorBlue;
        }

    //绘图
    //显示结果
    int thickness = -1;
    int lineType = 8;

    //两类结果一个黑色一个白色
    Scalar c1 = Scalar::all(0);
    Scalar c2 = Scalar::all(255);

    for (int i = 0; i < LabelsMat.rows; ++i)
    {
        //取出每行指针
        const float* v = TrainMat1.ptr<float>(i);
        Point pt = Point((int)v[0], (int)v[1]);
        if (1 == labels[i])
            circle(image, pt, 5, c1, thickness, lineType);
        else
            circle(image, pt, 5, c2, thickness, lineType);
    }

    imshow("SVM Simple Example", image);
    waitKey();
}


int main(int argc, char** argv)
{
    static string dir = "train_data.txt";
    Ptr<SVM> SvmModel;
    TRAINING_DATA TData;
    vector<int> TLabel;

    ReadTrainingData(dir, TData, TLabel);
    TrainSVM(SvmModel, TData, TLabel);

    waitKey();
}