OpenCV——ANN神经网络

ANN—— Artificial Neural Networks 人工神经网络

//定义人工神经网络
    CvANN_MLP bp;
    // Set up BPNetwork's parameters
    CvANN_MLP_TrainParams params;
    params.train_method=CvANN_MLP_TrainParams::BACKPROP;
    params.bp_dw_scale=0.1;
    params.bp_moment_scale=0.1;
    //params.train_method=CvANN_MLP_TrainParams::RPROP;
    //params.rp_dw0 = 0.1;
    //params.rp_dw_plus = 1.2;
    //params.rp_dw_minus = 0.5;
    //params.rp_dw_min = FLT_EPSILON;
    //params.rp_dw_max = 50.;

两种训练方法：BACKPROP 与 RPROP

BACKPROP的两个参数：

RPROP的四个参数：

//  training data
    float labels[][] = {{,,,,},{,,,,},{,,,,}};
    Mat labelsMat(, , CV_32FC1, labels);

    float trainingData[][] = { {,,,,},{,,,,}, {,,,,} };
    Mat trainingDataMat(, , CV_32FC1, trainingData);
// layerSizes设置了有三个隐含层的网络结构：输入层，三个隐含层，输出层。输入层和输出层节点数均为5，中间隐含层每层有两个节点

    Mat layerSizes=(Mat_<int>(,) << ,,,,);

//create第二个参数可以设置每个神经节点的激活函数，默认为CvANN_MLP::SIGMOID_SYM，即Sigmoid函数
//同时提供的其他激活函数有Gauss（CvANN_mlp::GAUSSIAN）和阶跃函数(CvANN_MLP::IDENTITY)。

 bp.create(layerSizes,CvANN_MLP::SIGMOID_SYM);   //CvANN_MLP::SIGMOID_SYM  
 bp.train(trainingDataMat, labelsMat, Mat(),Mat(), params);

//预测新节点
Mat sampleMat = (Mat_<float>(,) << i,j,,,);
            Mat responseMat;
            bp.predict(sampleMat,responseMat);  

float CvANN_MLP::predict(constMat&inputs,Mat&outputs)

图像进行特征提取，把它保存在inputs里，通过调用predict函数，我们得到一个输出向量，它是一个1*nClass的行向量，

其中每一列说明它与该类的相似程度（0-1之间），也可以说是置信度。我们只用对output求一个最大值，就可得到结果。

完整代码：

#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/ml/ml.hpp>
#include <iostream>
#include <string>  

using namespace std;
using namespace cv;  

int main()
{
    CvANN_MLP bp;   

    CvANN_MLP_TrainParams params;
    params.train_method=CvANN_MLP_TrainParams::BACKPROP;  //(Back Propagation,BP)反向传播算法
    params.bp_dw_scale=0.1;
    params.bp_moment_scale=0.1;  

    float labels[][] = {{0.9,0.1},{0.1,0.9},{0.9,0.1},{0.1,0.9},{0.9,0.1},{0.9,0.1},{0.1,0.9},{0.1,0.9},{0.9,0.1},{0.9,0.1}};
    //这里对于样本标记为0.1和0.9而非0和1，主要是考虑到sigmoid函数的输出为一般为0和1之间的数，只有在输入趋近于-∞和+∞才逐渐趋近于0和1，而不可能达到。
    Mat labelsMat(, , CV_32FC1, labels);  

    float trainingData[][] = { {,},{,}, {,}, {,},{,}, {,}, {,},{,}, {,}, {,} };
    Mat trainingDataMat(, , CV_32FC1, trainingData);
    Mat layerSizes=(Mat_<int>(,) << , , , , );                   //5层：输入层，3层隐藏层和输出层，每层均为两个perceptron
    bp.create(layerSizes,CvANN_MLP::SIGMOID_SYM);
    bp.train(trainingDataMat, labelsMat, Mat(),Mat(), params);
    int width = , height = ;
    Mat image = Mat::zeros(height, width, CV_8UC3);
    Vec3b green(,,), blue (,,);  

    for (int i = ; i < image.rows; ++i)
    {
        for (int j = ; j < image.cols; ++j)
        {
            Mat sampleMat = (Mat_<float>(,) << i,j);
            Mat responseMat;
            bp.predict(sampleMat,responseMat);
            float* p=responseMat.ptr<float>();
            //
            if (p[] > p[])
            {
                image.at<Vec3b>(j, i)  = green;
            }
            else
            {
                image.at<Vec3b>(j, i)  = blue;
            }
        }
    }
    // Show the training data
    int thickness = -;
    int lineType = ;
    circle( image, Point(,  ), , Scalar(  ,   ,   ), thickness, lineType);
    circle( image, Point(,  ), , Scalar(  ,   ,   ), thickness, lineType);
    circle( image, Point(,  ), , Scalar(  ,   ,   ), thickness, lineType);
    circle( image, Point(,  ), , Scalar(  ,   ,   ), thickness, lineType);
    circle( image, Point(,  ), , Scalar(, , ), thickness, lineType);
    circle( image, Point(, ), , Scalar(, , ), thickness, lineType);
    circle( image, Point(,  ), , Scalar(, , ), thickness, lineType);
    circle( image, Point(, ), , Scalar(, , ), thickness, lineType);
    circle( image, Point(,  ), , Scalar(, , ), thickness, lineType);
    circle( image, Point(, ), , Scalar(, , ), thickness, lineType);       

    imwrite("result.png", image);        // save the image   

    imshow("BP Simple Example", image); // show it to the user
    waitKey();  

    return ;
}