【模式识别】OpenCV中使用神经网络 CvANN_MLP

原文：http://blog.csdn.net/xiaowei_cqu/article/details/9027617

OpenCV的ml模块实现了人工神经网络（Artificial Neural Networks， ANN）最典型的多层感知器（multi-layer perceptrons, MLP）模型。由于ml模型实现的算法都继承自统一的CvStatModel基类，其训练和预测的接口都是train(),predict()，非常简单。

下面来看神经网络 CvANN_MLP 的使用~

定义神经网络及参数：

 //Setup the BPNetwork
     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.;

可以直接定义CvANN_MLP神经网络，并设置其参数。 BACKPROP表示使用back-propagation的训练方法，RPROP即最简单的propagation训练方法。

使用BACKPROP有两个相关参数：bp_dw_scale即bp_moment_scale：

使用PRPOP有四个相关参数：rp_dw0, rp_dw_plus, rp_dw_minus, rp_dw_min, rp_dw_max：

上述代码中为其默认值。

设置网络层数，训练数据：

 // Set up training data
     float labels[3][5] = {{0,0,0,0,0},{1,1,1,1,1},{0,0,0,0,0}};
     Mat labelsMat(3, 5, CV_32FC1, labels);

     float trainingData[3][5] = { {1,2,3,4,5},{111,112,113,114,115}, {21,22,23,24,25} };
     Mat trainingDataMat(3, 5, CV_32FC1, trainingData);
     Mat layerSizes=(Mat_<int>(1,5) << 5,2,2,2,5);
     bp.create(layerSizes,CvANN_MLP::SIGMOID_SYM);//CvANN_MLP::SIGMOID_SYM
                                                //CvANN_MLP::GAUSSIAN
                                                //CvANN_MLP::IDENTITY
     bp.train(trainingDataMat, labelsMat, Mat(),Mat(), params);

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。

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和阶跃函数。

使用训练好的网络结构分类新的数据：

然后直接使用predict函数，就可以预测新的节点：

 Mat sampleMat = (Mat_<float>(1,5) << i,j,0,0,0);
             Mat responseMat;
             bp.predict(sampleMat,responseMat);

完整程序代码：

 //The example of using BPNetwork in OpenCV
 //Coded by L. Wei
 #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()
 {
     //Setup the BPNetwork
     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.;

     // Set up training data
     float labels[3][5] = {{0,0,0,0,0},{1,1,1,1,1},{0,0,0,0,0}};
     Mat labelsMat(3, 5, CV_32FC1, labels);

     float trainingData[3][5] = { {1,2,3,4,5},{111,112,113,114,115}, {21,22,23,24,25} };
     Mat trainingDataMat(3, 5, CV_32FC1, trainingData);
     Mat layerSizes=(Mat_<int>(1,5) << 5,2,2,2,5);
     bp.create(layerSizes,CvANN_MLP::SIGMOID_SYM);//CvANN_MLP::SIGMOID_SYM
                                                //CvANN_MLP::GAUSSIAN
                                                //CvANN_MLP::IDENTITY
     bp.train(trainingDataMat, labelsMat, Mat(),Mat(), params);


     // Data for visual representation
     int width = 512, height = 512;
     Mat image = Mat::zeros(height, width, CV_8UC3);
     Vec3b green(0,255,0), blue (255,0,0);
     // Show the decision regions given by the SVM
     for (int i = 0; i < image.rows; ++i)
         for (int j = 0; j < image.cols; ++j)
         {
             Mat sampleMat = (Mat_<float>(1,5) << i,j,0,0,0);
             Mat responseMat;
             bp.predict(sampleMat,responseMat);
             float* p=responseMat.ptr<float>(0);
             int response=0;
             for(int i=0;i<5;i++){
             //  cout<<p[i]<<" ";
                 response+=p[i];
             }
             if (response >2)
                 image.at<Vec3b>(j, i)  = green;
             else
                 image.at<Vec3b>(j, i)  = blue;
         }

         // Show the training data
         int thickness = -1;
         int lineType = 8;
         circle( image, Point(501,  10), 5, Scalar(  0,   0,   0), thickness, lineType);
         circle( image, Point(255,  10), 5, Scalar(255, 255, 255), thickness, lineType);
         circle( image, Point(501, 255), 5, Scalar(255, 255, 255), thickness, lineType);
         circle( image, Point( 10, 501), 5, Scalar(255, 255, 255), thickness, lineType);

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

         imshow("BP Simple Example", image); // show it to the user
         waitKey(0);

 }

结果：