在opencv3中的机器学习算法

在opencv3.0中，提供了一个ml.cpp的文件，这里面全是机器学习的算法，共提供了这么几种：

1、正态贝叶斯：normal Bayessian classifier    我已在另外一篇博文中介绍过：在opencv3中实现机器学习之：利用正态贝叶斯分类

2、K最近邻：k nearest neighbors classifier

3、支持向量机：support vectors machine    请参考我的另外一篇博客：在opencv3中实现机器学习之：利用svm(支持向量机)分类

4、决策树： decision tree

5、ADA Boost：adaboost

6、梯度提升决策树：gradient boosted trees

7、随机森林：random forest

8、人工神经网络：artificial neural networks

9、EM算法：expectation-maximization

这些算法在任何一本机器学习书本上都可以介绍过，他们大致的分类过程都很相似，主要分为三个环节：

一、收集样本数据sampleData

二、训练分类器mode

三、对测试数据testData进行预测

不同的地方就是在opencv中的参数设定，假设训练数据为trainingDataMat,且已经标注好labelsMat。待测数据为testMat.

1、正态贝叶斯

 // 创建贝叶斯分类器
  Ptr<NormalBayesClassifier> model=NormalBayesClassifier::create();
    
    // 设置训练数据
  Ptr<TrainData> tData =TrainData::create(trainingDataMat, ROW_SAMPLE, labelsMat);

    //训练分类器
    model->train(tData);
//预测数据
 float response = model->predict(testMat); 

2、K最近邻

 Ptr<KNearest> knn = KNearest::create();  //创建knn分类器
    knn->setDefaultK(K);    //设定k值
    knn->setIsClassifier(true);
    // 设置训练数据
    Ptr<TrainData> tData = TrainData::create(trainingDataMat, ROW_SAMPLE, labelsMat);
    knn->train(tData);
    float response = knn->predict(testMat);

3、支持向量机

Ptr<SVM> svm = SVM::create();    //创建一个分类器
    svm->setType(SVM::C_SVC);    //设置svm类型
    svm->setKernel(SVM::POLY); //设置核函数;
    svm->setDegree(0.5);
    svm->setGamma(1);
    svm->setCoef0(1);
    svm->setNu(0.5);
    svm->setP(0);
    svm->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER+TermCriteria::EPS, 1000, 0.01));
    svm->setC(C);
    Ptr<TrainData> tData = TrainData::create(trainingDataMat, ROW_SAMPLE, labelsMat);
    svm->train(tData);
    float response = svm->predict(testMat);

4、决策树： decision tree

Ptr<DTrees> dtree = DTrees::create();  //创建分类器
    dtree->setMaxDepth(8);   //设置最大深度
    dtree->setMinSampleCount(2);  
    dtree->setUseSurrogates(false);
    dtree->setCVFolds(0); //交叉验证
    dtree->setUse1SERule(false);
    dtree->setTruncatePrunedTree(false);
    Ptr<TrainData> tData = TrainData::create(trainingDataMat, ROW_SAMPLE, labelsMat);
    dtree->train(tData);
    float response = dtree->predict(testMat);

5、ADA Boost：adaboost

 Ptr<Boost> boost = Boost::create();
    boost->setBoostType(Boost::DISCRETE);
    boost->setWeakCount(100);
    boost->setWeightTrimRate(0.95);
    boost->setMaxDepth(2);
    boost->setUseSurrogates(false);
    boost->setPriors(Mat());
    Ptr<TrainData> tData = TrainData::create(trainingDataMat, ROW_SAMPLE, labelsMat);
    boost->train(tData);
    float response = boost->predict(testMat);

6、梯度提升决策树：gradient boosted trees

此算法在opencv3.0中被注释掉了，原因未知，因此此处提供一个老版本的算法。

GBTrees::Params params( GBTrees::DEVIANCE_LOSS, // loss_function_type
                         100, // weak_count
                         0.1f, // shrinkage
                         1.0f, // subsample_portion
                         2, // max_depth
                         false // use_surrogates )
                         );
    Ptr<TrainData> tData = TrainData::create(trainingDataMat, ROW_SAMPLE, labelsMat);
    Ptr<GBTrees> gbtrees = StatModel::train<GBTrees>(tData, params);
    float response = gbtrees->predict(testMat);

7、随机森林：random forest

   Ptr<RTrees> rtrees = RTrees::create();
    rtrees->setMaxDepth(4);
    rtrees->setMinSampleCount(2);
    rtrees->setRegressionAccuracy(0.f);
    rtrees->setUseSurrogates(false);
    rtrees->setMaxCategories(16);
    rtrees->setPriors(Mat());
    rtrees->setCalculateVarImportance(false);
    rtrees->setActiveVarCount(1);
    rtrees->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER, 5, 0));
   Ptr<TrainData> tData = TrainData::create(trainingDataMat, ROW_SAMPLE, labelsMat);
   rtrees->train(tData);
   float response = rtrees->predict(testMat);

8、人工神经网络：artificial neural networks

 Ptr<ANN_MLP> ann = ANN_MLP::create();
    ann->setLayerSizes(layer_sizes);
    ann->setActivationFunction(ANN_MLP::SIGMOID_SYM, 1, 1);
    ann->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER+TermCriteria::EPS, 300, FLT_EPSILON));
    ann->setTrainMethod(ANN_MLP::BACKPROP, 0.001);
    Ptr<TrainData> tData = TrainData::create(trainingDataMat, ROW_SAMPLE, labelsMat);
    ann->train(tData);
    float response = ann->predict(testMat);

9、EM算法：expectation-maximization

EM算法与前面的稍微有点不同，它需要创建很多个model,将trainingDataMat分成很多个modelSamples，每个modelSamples训练出一个model

训练核心代码为：

 int nmodels = (int)labelsMat.size();
    vector<Ptr<EM> > em_models(nmodels);
    Mat modelSamples;

    for( i = 0; i < nmodels; i++ )
    {
        const int componentCount = 3;

        modelSamples.release();
        for (j = 0; j < labelsMat.rows; j++)
        {
            if (labelsMat.at<int>(j,0)== i)
                modelSamples.push_back(trainingDataMat.row(j));
        }

        // learn models
        if( !modelSamples.empty() )
        {
            Ptr<EM> em = EM::create();
            em->setClustersNumber(componentCount);
            em->setCovarianceMatrixType(EM::COV_MAT_DIAGONAL);
            em->trainEM(modelSamples, noArray(), noArray(), noArray());
            em_models[i] = em;
        }
    }

预测：

 Mat logLikelihoods(1, nmodels, CV_64FC1, Scalar(-DBL_MAX));
 for( i = 0; i < nmodels; i++ )
            {
                if( !em_models[i].empty() )
                    logLikelihoods.at<double>(i) = em_models[i]->predict2(testMat, noArray())[0];
            }

 

这么多的机器学习算法，在实际用途中照我的理解其实只需要掌握svm算法就可以了。

ANN算法在opencv中也叫多层感知机，因此在训练的时候，需要分多层。

EM算法需要为每一类创建一个model。

其中一些算法的具体代码练习：在opencv3中的机器学习算法练习：对OCR进行分类