?????????????? 原文鏈接：OpenCV3 Java 機(jī)器學(xué)習(xí)使用方法匯總??????????????????????????????

?前言

????????? 按道理來說，C++版本的OpenCV訓(xùn)練的版本XML文件，在java中可以無縫使用。但要注意OpenCV本身的版本問題。從2.4 到3.x版本出現(xiàn)了很大的改變，XML文件本身的存儲格式本身也不同，不能通用。

????????? opencv提供了非常多的機(jī)器學(xué)習(xí)算法用于研究。這里對這些算法進(jìn)行分類學(xué)習(xí)和研究，以拋磚引玉。這里使用的機(jī)器學(xué)習(xí)算法包括：人工神經(jīng)網(wǎng)絡(luò)，boost,決策樹，最近鄰，邏輯回歸，貝葉斯，隨機(jī)森林，SVM等算法等。

????????? 機(jī)器學(xué)習(xí)的過程相同，都要經(jīng)歷1、收集樣本數(shù)據(jù)sampleData2.訓(xùn)練分類器mode3.對測試數(shù)據(jù)testData進(jìn)行預(yù)測。這里使用一個在別處看到的例子，利用身高體重等原始信息預(yù)測男女的概率。通過一些簡單的數(shù)據(jù)學(xué)習(xí)，用測試數(shù)據(jù)預(yù)測男女概率。

實(shí)例代碼：

import org.opencv.core.Core; import org.opencv.core.CvType; import org.opencv.core.Mat; import org.opencv.core.TermCriteria; import org.opencv.ml.ANN_MLP; import org.opencv.ml.Boost; import org.opencv.ml.DTrees; import org.opencv.ml.KNearest; import org.opencv.ml.LogisticRegression; import org.opencv.ml.Ml; import org.opencv.ml.NormalBayesClassifier; import org.opencv.ml.RTrees; import org.opencv.ml.SVM; import org.opencv.ml.SVMSGD; import org.opencv.ml.TrainData; public class ML { public static void main(String[] args) { System.loadLibrary(Core.NATIVE_LIBRARY_NAME); // 訓(xùn)練數(shù)據(jù)，兩個維度，表示身高和體重 float[] trainingData = { 186, 80, 185, 81, 160, 50, 161, 48 }; // 訓(xùn)練標(biāo)簽數(shù)據(jù)，前兩個表示男生0，后兩個表示女生1,由于使用了多種機(jī)器學(xué)習(xí)算法，他們的輸入有些不一樣，所以labelsMat有三種 float[] labels = { 0f, 0f, 0f, 0f, 1f, 1f, 1f, 1f }; int[] labels2 = { 0, 0, 1, 1 }; float[] labels3 = { 0, 0, 1, 1 }; // 測試數(shù)據(jù),先男后女 float[] test = { 184, 79, 159, 50 }; Mat trainingDataMat = new Mat(4, 2, CvType.CV_32FC1); trainingDataMat.put(0, 0, trainingData); Mat labelsMat = new Mat(4, 2, CvType.CV_32FC1); labelsMat.put(0, 0, labels); Mat labelsMat2 = new Mat(4, 1, CvType.CV_32SC1); labelsMat2.put(0, 0, labels2); Mat labelsMat3 = new Mat(4, 1, CvType.CV_32FC1); labelsMat3.put(0, 0, labels3); Mat sampleMat = new Mat(2, 2, CvType.CV_32FC1); sampleMat.put(0, 0, test); MyAnn(trainingDataMat, labelsMat, sampleMat); MyBoost(trainingDataMat, labelsMat2, sampleMat); MyDtrees(trainingDataMat, labelsMat2, sampleMat); MyKnn(trainingDataMat, labelsMat3, sampleMat); MyLogisticRegression(trainingDataMat, labelsMat3, sampleMat); MyNormalBayes(trainingDataMat, labelsMat2, sampleMat); MyRTrees(trainingDataMat, labelsMat2, sampleMat); MySvm(trainingDataMat, labelsMat2, sampleMat); MySvmsgd(trainingDataMat, labelsMat2, sampleMat); } // 人工神經(jīng)網(wǎng)絡(luò) public static Mat MyAnn(Mat trainingData, Mat labels, Mat testData) { // train data using aNN TrainData td = TrainData.create(trainingData, Ml.ROW_SAMPLE, labels); Mat layerSizes = new Mat(1, 4, CvType.CV_32FC1); // 含有兩個隱含層的網(wǎng)絡(luò)結(jié)構(gòu)，輸入、輸出層各兩個節(jié)點(diǎn)，每個隱含層含兩個節(jié)點(diǎn) layerSizes.put(0, 0, new float[] { 2, 2, 2, 2 }); ANN_MLP ann = ANN_MLP.create(); ann.setLayerSizes(layerSizes); ann.setTrainMethod(ANN_MLP.BACKPROP); ann.setBackpropWeightScale(0.1); ann.setBackpropMomentumScale(0.1); ann.setActivationFunction(ANN_MLP.SIGMOID_SYM, 1, 1); ann.setTermCriteria(new TermCriteria(TermCriteria.MAX_ITER + TermCriteria.EPS, 300, 0.0)); boolean success = ann.train(td.getSamples(), Ml.ROW_SAMPLE, td.getResponses()); System.out.println("Ann training result: " + success); // ann.save("D:/bp.xml");//存儲模型 // ann.load("D:/bp.xml");//讀取模型 // 測試數(shù)據(jù) Mat responseMat = new Mat(); ann.predict(testData, responseMat, 0); System.out.println("Ann responseMat:\n" + responseMat.dump()); for (int i = 0; i < responseMat.size().height; i++) { if (responseMat.get(i, 0)[0] + responseMat.get(i, i)[0] >= 1) System.out.println("Girl\n"); if (responseMat.get(i, 0)[0] + responseMat.get(i, i)[0] < 1) System.out.println("Boy\n"); } return responseMat; } // Boost public static Mat MyBoost(Mat trainingData, Mat labels, Mat testData) { Boost boost = Boost.create(); // boost.setBoostType(Boost.DISCRETE); boost.setBoostType(Boost.GENTLE); boost.setWeakCount(2); boost.setWeightTrimRate(0.95); boost.setMaxDepth(2); boost.setUseSurrogates(false); boost.setPriors(new Mat()); TrainData td = TrainData.create(trainingData, Ml.ROW_SAMPLE, labels); boolean success = boost.train(td.getSamples(), Ml.ROW_SAMPLE, td.getResponses()); System.out.println("Boost training result: " + success); // boost.save("D:/bp.xml");//存儲模型 Mat responseMat = new Mat(); float response = boost.predict(testData, responseMat, 0); System.out.println("Boost responseMat:\n" + responseMat.dump()); for (int i = 0; i < responseMat.height(); i++) { if (responseMat.get(i, 0)[0] == 0) System.out.println("Boy\n"); if (responseMat.get(i, 0)[0] == 1) System.out.println("Girl\n"); } return responseMat; } // 決策樹 public static Mat MyDtrees(Mat trainingData, Mat labels, Mat testData) { DTrees dtree = DTrees.create(); // 創(chuàng)建分類器 dtree.setMaxDepth(8); // 設(shè)置最大深度 dtree.setMinSampleCount(2); dtree.setUseSurrogates(false); dtree.setCVFolds(0); // 交叉驗(yàn)證 dtree.setUse1SERule(false); dtree.setTruncatePrunedTree(false); TrainData td = TrainData.create(trainingData, Ml.ROW_SAMPLE, labels); boolean success = dtree.train(td.getSamples(), Ml.ROW_SAMPLE, td.getResponses()); System.out.println("Dtrees training result: " + success); // dtree.save("D:/bp.xml");//存儲模型 Mat responseMat = new Mat(); float response = dtree.predict(testData, responseMat, 0); System.out.println("Dtrees responseMat:\n" + responseMat.dump()); for (int i = 0; i < responseMat.height(); i++) { if (responseMat.get(i, 0)[0] == 0) System.out.println("Boy\n"); if (responseMat.get(i, 0)[0] == 1) System.out.println("Girl\n"); } return responseMat; } // K最鄰近 public static Mat MyKnn(Mat trainingData, Mat labels, Mat testData) { final int K = 2; TrainData td = TrainData.create(trainingData, Ml.ROW_SAMPLE, labels); KNearest knn = KNearest.create(); boolean success = knn.train(trainingData, Ml.ROW_SAMPLE, labels); System.out.println("Knn training result: " + success); // knn.save("D:/bp.xml");//存儲模型 // find the nearest neighbours of test data Mat results = new Mat(); Mat neighborResponses = new Mat(); Mat dists = new Mat(); knn.findNearest(testData, K, results, neighborResponses, dists); System.out.println("results:\n" + results.dump()); System.out.println("Knn neighborResponses:\n" + neighborResponses.dump()); System.out.println("dists:\n" + dists.dump()); for (int i = 0; i < results.height(); i++) { if (results.get(i, 0)[0] == 0) System.out.println("Boy\n"); if (results.get(i, 0)[0] == 1) System.out.println("Girl\n"); } return results; } // 邏輯回歸 public static Mat MyLogisticRegression(Mat trainingData, Mat labels, Mat testData) { LogisticRegression lr = LogisticRegression.create(); TrainData td = TrainData.create(trainingData, Ml.ROW_SAMPLE, labels); boolean success = lr.train(td.getSamples(), Ml.ROW_SAMPLE, td.getResponses()); System.out.println("LogisticRegression training result: " + success); // lr.save("D:/bp.xml");//存儲模型 Mat responseMat = new Mat(); float response = lr.predict(testData, responseMat, 0); System.out.println("LogisticRegression responseMat:\n" + responseMat.dump()); for (int i = 0; i < responseMat.height(); i++) { if (responseMat.get(i, 0)[0] == 0) System.out.println("Boy\n"); if (responseMat.get(i, 0)[0] == 1) System.out.println("Girl\n"); } return responseMat; } // 貝葉斯 public static Mat MyNormalBayes(Mat trainingData, Mat labels, Mat testData) { NormalBayesClassifier nb = NormalBayesClassifier.create(); TrainData td = TrainData.create(trainingData, Ml.ROW_SAMPLE, labels); boolean success = nb.train(td.getSamples(), Ml.ROW_SAMPLE, td.getResponses()); System.out.println("NormalBayes training result: " + success); // nb.save("D:/bp.xml");//存儲模型 Mat responseMat = new Mat(); float response = nb.predict(testData, responseMat, 0); System.out.println("NormalBayes responseMat:\n" + responseMat.dump()); for (int i = 0; i < responseMat.height(); i++) { if (responseMat.get(i, 0)[0] == 0) System.out.println("Boy\n"); if (responseMat.get(i, 0)[0] == 1) System.out.println("Girl\n"); } return responseMat; } // 隨機(jī)森林 public static Mat MyRTrees(Mat trainingData, Mat labels, Mat testData) { RTrees rtrees = RTrees.create(); rtrees.setMaxDepth(4); rtrees.setMinSampleCount(2); rtrees.setRegressionAccuracy(0.f); rtrees.setUseSurrogates(false); rtrees.setMaxCategories(16); rtrees.setPriors(new Mat()); rtrees.setCalculateVarImportance(false); rtrees.setActiveVarCount(1); rtrees.setTermCriteria(new TermCriteria(TermCriteria.MAX_ITER, 5, 0)); TrainData tData = TrainData.create(trainingData, Ml.ROW_SAMPLE, labels); boolean success = rtrees.train(tData.getSamples(), Ml.ROW_SAMPLE, tData.getResponses()); System.out.println("Rtrees training result: " + success); // rtrees.save("D:/bp.xml");//存儲模型 Mat responseMat = new Mat(); rtrees.predict(testData, responseMat, 0); System.out.println("Rtrees responseMat:\n" + responseMat.dump()); for (int i = 0; i < responseMat.height(); i++) { if (responseMat.get(i, 0)[0] == 0) System.out.println("Boy\n"); if (responseMat.get(i, 0)[0] == 1) System.out.println("Girl\n"); } return responseMat; } // 支持向量機(jī) public static Mat MySvm(Mat trainingData, Mat labels, Mat testData) { SVM svm = SVM.create(); svm.setKernel(SVM.LINEAR); svm.setType(SVM.C_SVC); TermCriteria criteria = new TermCriteria(TermCriteria.EPS + TermCriteria.MAX_ITER, 1000, 0); svm.setTermCriteria(criteria); svm.setGamma(0.5); svm.setNu(0.5); svm.setC(1); TrainData td = TrainData.create(trainingData, Ml.ROW_SAMPLE, labels); boolean success = svm.train(td.getSamples(), Ml.ROW_SAMPLE, td.getResponses()); System.out.println("Svm training result: " + success); // svm.save("D:/bp.xml");//存儲模型 // svm.load("D:/bp.xml");//讀取模型 Mat responseMat = new Mat(); svm.predict(testData, responseMat, 0); System.out.println("SVM responseMat:\n" + responseMat.dump()); for (int i = 0; i < responseMat.height(); i++) { if (responseMat.get(i, 0)[0] == 0) System.out.println("Boy\n"); if (responseMat.get(i, 0)[0] == 1) System.out.println("Girl\n"); } return responseMat; } // SGD支持向量機(jī) public static Mat MySvmsgd(Mat trainingData, Mat labels, Mat testData) { SVMSGD Svmsgd = SVMSGD.create(); TermCriteria criteria = new TermCriteria(TermCriteria.EPS + TermCriteria.MAX_ITER, 1000, 0); Svmsgd.setTermCriteria(criteria); Svmsgd.setInitialStepSize(2); Svmsgd.setSvmsgdType(SVMSGD.SGD); Svmsgd.setMarginRegularization(0.5f); boolean success = Svmsgd.train(trainingData, Ml.ROW_SAMPLE, labels); System.out.println("SVMSGD training result: " + success); // svm.save("D:/bp.xml");//存儲模型 // svm.load("D:/bp.xml");//讀取模型 Mat responseMat = new Mat(); Svmsgd.predict(testData, responseMat, 0); System.out.println("SVMSGD responseMat:\n" + responseMat.dump()); for (int i = 0; i < responseMat.height(); i++) { if (responseMat.get(i, 0)[0] == 0) System.out.println("Boy\n"); if (responseMat.get(i, 0)[0] == 1) System.out.println("Girl\n"); } return responseMat; } }

備注：作者的代碼運(yùn)行無誤，可直接測試。

創(chuàng)作挑戰(zhàn)賽新人創(chuàng)作獎勵來咯，堅持創(chuàng)作打卡瓜分現(xiàn)金大獎

總結(jié)

以上是生活随笔為你收集整理的OpenCV3 Java 机器学习使用方法汇总的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。