From 139acaf74b135dbb0239bca3150cfddb1ca572a8 Mon Sep 17 00:00:00 2001 From: lidapeng Date: Sat, 1 Feb 2020 14:00:48 +0800 Subject: [PATCH] =?UTF-8?q?=E5=A4=A7=E5=B9=85=E5=BA=A6=E6=8F=90=E9=AB=98?= =?UTF-8?q?=E4=BA=86=E7=B2=BE=E5=87=86=E6=A8=A1=E5=BC=8F=E7=9A=84=E8=BF=90?= =?UTF-8?q?=E8=A1=8C=E9=80=9F=E5=BA=A6=EF=BC=8C=E8=88=8D=E5=BC=83=E4=BA=86?= =?UTF-8?q?=E9=83=A8=E5=88=86=E7=B2=BE=E5=BA=A6?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- .../wlld/imageRecognition/Convolution.java | 4 +- .../org/wlld/imageRecognition/KMatrix.java | 52 +++++++ .../org/wlld/imageRecognition/Operation.java | 61 ++++---- .../wlld/imageRecognition/TempleConfig.java | 34 ++++- .../org/wlld/nerveEntity/HiddenNerve.java | 2 +- .../org/wlld/nerveEntity/ModelParameter.java | 10 ++ src/main/java/org/wlld/nerveEntity/Nerve.java | 8 +- .../java/org/wlld/nerveEntity/OutNerve.java | 19 +-- .../org/wlld/nerveEntity/SensoryNerve.java | 2 +- src/test/java/org/wlld/HelloWorld.java | 133 +++++++++++++----- src/test/java/org/wlld/Ma.java | 3 +- 11 files changed, 239 insertions(+), 89 deletions(-) create mode 100644 src/main/java/org/wlld/imageRecognition/KMatrix.java diff --git a/src/main/java/org/wlld/imageRecognition/Convolution.java b/src/main/java/org/wlld/imageRecognition/Convolution.java index 46e7583..25579af 100644 --- a/src/main/java/org/wlld/imageRecognition/Convolution.java +++ b/src/main/java/org/wlld/imageRecognition/Convolution.java @@ -59,8 +59,8 @@ public class Convolution { List frameBodies = new ArrayList<>(); double xNub = frame.getLengthHeight(); double yNub = frame.getLengthWidth(); - for (int i = 0; i < x - xFrame; i += xNub) { - for (int j = 0; j < y - yFrame; j += yNub) { + for (int i = 0; i <= x - xFrame; i += xNub) { + for (int j = 0; j <= y - yFrame; j += yNub) { FrameBody frameBody = new FrameBody(); Matrix myMatrix = matrix.getSonOfMatrix(i, j, xFrame, yFrame); frameBody.setMatrix(myMatrix); diff --git a/src/main/java/org/wlld/imageRecognition/KMatrix.java b/src/main/java/org/wlld/imageRecognition/KMatrix.java new file mode 100644 index 0000000..3958f85 --- /dev/null +++ b/src/main/java/org/wlld/imageRecognition/KMatrix.java @@ -0,0 +1,52 @@ +package org.wlld.imageRecognition; + +import org.wlld.MatrixTools.Matrix; +import org.wlld.MatrixTools.MatrixOperation; +import org.wlld.tools.ArithUtil; + +/** + * @author lidapeng + * @description K均值聚类矩阵 + * @date 9:05 上午 2020/2/1 + */ +public class KMatrix { + private Matrix sigmaMatrix;//矩阵和 + private int nub;//加和次数 + private boolean isFinish = false;//是否结算 + + KMatrix(int x, int y) { + sigmaMatrix = new Matrix(x, y); + } + + public double getEDist(Matrix matrix) throws Exception {//返回欧式距离 + if (isFinish && matrix.getX() == sigmaMatrix.getX() + && matrix.getY() == sigmaMatrix.getY()) { + double sigma = 0; + for (int i = 0; i < matrix.getX(); i++) { + for (int j = 0; j < matrix.getY(); j++) { + double sub = ArithUtil.sub(matrix.getNumber(i, j), sigmaMatrix.getNumber(i, j)); + sigma = ArithUtil.add(Math.pow(sub, 2), sigma); + } + } + sigma = Math.sqrt(sigma); + return sigma; + } else { + throw new Exception("K is not finish or matrix not equal"); + } + } + + public void addMatrix(Matrix matrix) throws Exception { + sigmaMatrix = MatrixOperation.add(sigmaMatrix, matrix); + nub++; + } + + public void getK() throws Exception {//结算K均值 + if (nub != 0) { + double aNub = ArithUtil.div(1, nub); + MatrixOperation.mathMul(sigmaMatrix, aNub); + isFinish = true; + } else { + throw new Exception("not value"); + } + } +} diff --git a/src/main/java/org/wlld/imageRecognition/Operation.java b/src/main/java/org/wlld/imageRecognition/Operation.java index 247d47c..aaac164 100644 --- a/src/main/java/org/wlld/imageRecognition/Operation.java +++ b/src/main/java/org/wlld/imageRecognition/Operation.java @@ -65,17 +65,10 @@ public class Operation {//进行计算 } //卷积核学习 - public void learning(Matrix matrix, Map tagging, boolean isNerveStudy) throws Exception { + public void learning(Matrix matrix, int tagging, boolean isNerveStudy) throws Exception { if (templeConfig.getStudyPattern() == StudyPattern.Accuracy_Pattern) { Border border = null; - int id = 0; - for (Map.Entry entry : tagging.entrySet()) { - if (entry.getValue() > 0.5) { - id = entry.getKey(); - break; - } - } - if (templeConfig.isHavePosition() && isNerveStudy && id > 0) { + if (templeConfig.isHavePosition() && isNerveStudy && tagging > 0) { outBack = imageBack; border = convolution.borderOnce(matrix, templeConfig); } @@ -88,12 +81,19 @@ public class Operation {//进行计算 isKernelStudy, tagging, matrixBack); if (isNerveStudy) { Matrix myMatrix = matrixBack.getMatrix(); - if (templeConfig.isHavePosition() && id > 0) { - border.end(myMatrix, id); + if (templeConfig.isHavePosition() && tagging > 0) { + border.end(myMatrix, tagging); } - long eventId = matrixBack.getEventId(); - List featureList = getFeatureList(myMatrix); - intoNerve(eventId, featureList, templeConfig.getSensoryNerves(), true, tagging, outBack); + //System.out.println(myMatrix.getString()); + //进行聚类 + Map kMatrixMap = templeConfig.getkMatrixMap(); + if (kMatrixMap.containsKey(tagging)) { + KMatrix kMatrix = kMatrixMap.get(tagging); + kMatrix.addMatrix(myMatrix); + } else { + throw new Exception("not find tag"); + } + //intoNerve(1, featureList, templeConfig.getSensoryNerves(), true, tagging, outBack); } } else { throw new Exception("pattern is wrong"); @@ -106,7 +106,7 @@ public class Operation {//进行计算 List list = convolution(matrix, null); intoNerve(eventId, list, templeConfig.getSensoryNerves(), false, null, outBack); } else if (templeConfig.getStudyPattern() == StudyPattern.Accuracy_Pattern) { - see(matrix, eventId); + throw new Exception("studyPattern not right"); } } @@ -138,7 +138,7 @@ public class Operation {//进行计算 } else if (templeConfig.getStudyPattern() == StudyPattern.Accuracy_Pattern) { for (FrameBody frameBody : frameBodies) { intoNerve2(eventId, frameBody.getMatrix(), templeConfig.getConvolutionNerveManager().getSensoryNerves(), - false, null, matrixBack); + false, -1, matrixBack); Matrix myMatrix = matrixBack.getMatrix(); //卷积层输出即边框回归的输入的特征向量 frameBody.setEndMatrix(myMatrix); @@ -279,13 +279,24 @@ public class Operation {//进行计算 } //图像视觉 Accuracy 模式 - private void see(Matrix matrix, long eventId) throws Exception { + public int toSee(Matrix matrix) throws Exception { if (templeConfig.getStudyPattern() == StudyPattern.Accuracy_Pattern) { - intoNerve2(eventId, matrix, templeConfig.getConvolutionNerveManager().getSensoryNerves(), - false, null, matrixBack); + intoNerve2(2, matrix, templeConfig.getConvolutionNerveManager().getSensoryNerves(), + false, -1, matrixBack); Matrix myMatrix = matrixBack.getMatrix(); - List featureList = getFeatureList(myMatrix); - intoNerve(eventId, featureList, templeConfig.getSensoryNerves(), false, null, outBack); + int id = 0; + double distEnd = 0; + Map kMatrixMap = templeConfig.getkMatrixMap(); + for (Map.Entry entry : kMatrixMap.entrySet()) { + int key = entry.getKey(); + KMatrix kMatrix = entry.getValue(); + double dist = kMatrix.getEDist(myMatrix); + if (distEnd == 0 || dist < distEnd) { + id = key; + distEnd = dist; + } + } + return id; } else { throw new Exception("pattern is wrong"); } @@ -311,23 +322,23 @@ public class Operation {//进行计算 List list = new ArrayList<>(); for (int i = 0; i < matrix.getX(); i++) { for (int j = 0; j < matrix.getY(); j++) { - double nub = ArithUtil.div(matrix.getNumber(i, j), 10); + double nub = matrix.getNumber(i, j); list.add(nub); } } return list; } - private void intoNerve(long eventId, List featurList, List sensoryNerveList + private void intoNerve(long eventId, List featureList, List sensoryNerveList , boolean isStudy, Map map, OutBack imageBack) throws Exception { for (int i = 0; i < sensoryNerveList.size(); i++) { - sensoryNerveList.get(i).postMessage(eventId, featurList.get(i), isStudy, map + sensoryNerveList.get(i).postMessage(eventId, featureList.get(i), isStudy, map , imageBack); } } private void intoNerve2(long eventId, Matrix featur, List sensoryNerveList - , boolean isKernelStudy, Map E, OutBack outBack) throws Exception { + , boolean isKernelStudy, int E, OutBack outBack) throws Exception { for (int i = 0; i < sensoryNerveList.size(); i++) { sensoryNerveList.get(i).postMatrixMessage(eventId, featur, isKernelStudy, E, outBack); } diff --git a/src/main/java/org/wlld/imageRecognition/TempleConfig.java b/src/main/java/org/wlld/imageRecognition/TempleConfig.java index 916a639..8edeb56 100644 --- a/src/main/java/org/wlld/imageRecognition/TempleConfig.java +++ b/src/main/java/org/wlld/imageRecognition/TempleConfig.java @@ -28,6 +28,7 @@ public class TempleConfig { private int studyPattern;//学习模式 private boolean isHavePosition = false;//是否需要锁定物体位置 private Map borderBodyMap = new HashMap<>();//border特征集合 + private Map kMatrixMap = new HashMap<>();//K均值矩阵集合 private Frame frame;//先验边框 private double th = 0.6;//标准阈值 private boolean boxReady = false;//边框已经学习完毕 @@ -65,6 +66,16 @@ public class TempleConfig { return borderBodyMap; } + public Map getkMatrixMap() { + return kMatrixMap; + } + + public void clustering() throws Exception {//进行聚类结算 + for (Map.Entry kMatrixEntry : kMatrixMap.entrySet()) { + kMatrixEntry.getValue().getK(); + } + } + private void border(BorderBody borderBody) throws Exception { Matrix parameter = borderBody.getX();//参数矩阵 Matrix tx = borderBody.getTx(); @@ -160,11 +171,13 @@ public class TempleConfig { height = height / 3; deep++; } - //加载各识别分类的期望矩阵 - double nub = ArithUtil.div(10, classificationNub);//每个分类期望参数的跨度 + matrixMap.put(0, new Matrix(height, width)); + double nub = 10;//每个分类期望参数的跨度 + kMatrixMap.put(0, new KMatrix(height, width)); for (int k = 0; k < classificationNub; k++) { Matrix matrix = new Matrix(height, width);//初始化期望矩阵 + kMatrixMap.put(k + 1, new KMatrix(height, width)); double t = (k + 1) * nub;//期望矩阵的分类参数数值 for (int i = 0; i < height; i++) {//给期望矩阵注入期望参数 for (int j = 0; j < width - 1; j++) { @@ -175,17 +188,23 @@ public class TempleConfig { } convolutionNerveManager = new NerveManager(1, 1, 1, deep - 1, new ReLu(), true); - initNerveManager(initPower, width * height, 2); + //initNerveManager(initPower, width * height, 2); + convolutionNerveManager.setMatrixMap(matrixMap);//给卷积网络管理器注入期望矩阵 convolutionNerveManager.init(initPower, true); } public ModelParameter getModel() throws Exception {//获取模型参数 - ModelParameter modelParameter = nerveManager.getModelParameter(); + ModelParameter modelParameter = new ModelParameter(); if (studyPattern == StudyPattern.Accuracy_Pattern) { ModelParameter modelParameter1 = convolutionNerveManager.getModelParameter(); modelParameter.setDymNerveStudies(modelParameter1.getDymNerveStudies()); modelParameter.setDymOutNerveStudy(modelParameter1.getDymOutNerveStudy()); + //modelParameter.setkMatrixMap(kMatrixMap); + } else if (studyPattern == StudyPattern.Speed_Pattern) { + ModelParameter modelParameter1 = nerveManager.getModelParameter(); + modelParameter.setDepthNerves(modelParameter1.getDepthNerves()); + modelParameter.setOutNevers(modelParameter1.getOutNevers()); } if (isHavePosition) {//存在边框学习模型参数 modelParameter.setBorderBodyMap(borderBodyMap); @@ -226,9 +245,14 @@ public class TempleConfig { //注入模型参数 public void insertModel(ModelParameter modelParameter) throws Exception { - nerveManager.insertModelParameter(modelParameter); if (studyPattern == StudyPattern.Accuracy_Pattern) { convolutionNerveManager.insertModelParameter(modelParameter); + Map kMatrixs = modelParameter.getkMatrixMap(); + if (kMatrixs != null && kMatrixs.size() == kMatrixMap.size()) { + kMatrixMap = modelParameter.getkMatrixMap(); + } + } else if (studyPattern == StudyPattern.Speed_Pattern) { + nerveManager.insertModelParameter(modelParameter); } Frame frame = modelParameter.getFrame(); Map borderBodyMap = modelParameter.getBorderBodyMap(); diff --git a/src/main/java/org/wlld/nerveEntity/HiddenNerve.java b/src/main/java/org/wlld/nerveEntity/HiddenNerve.java index 2dbbfb3..47ddb9b 100644 --- a/src/main/java/org/wlld/nerveEntity/HiddenNerve.java +++ b/src/main/java/org/wlld/nerveEntity/HiddenNerve.java @@ -41,7 +41,7 @@ public class HiddenNerve extends Nerve { @Override protected void inputMatrix(long eventId, Matrix matrix, boolean isStudy - , Map E, OutBack outBack) throws Exception { + , int E, OutBack outBack) throws Exception { Matrix myMatrix = dynamicNerve(matrix, eventId, isStudy);//处理过的矩阵 sendMatrix(eventId, myMatrix, isStudy, E, outBack); } diff --git a/src/main/java/org/wlld/nerveEntity/ModelParameter.java b/src/main/java/org/wlld/nerveEntity/ModelParameter.java index 29a3017..10e9903 100644 --- a/src/main/java/org/wlld/nerveEntity/ModelParameter.java +++ b/src/main/java/org/wlld/nerveEntity/ModelParameter.java @@ -1,6 +1,7 @@ package org.wlld.nerveEntity; import org.wlld.MatrixTools.Matrix; +import org.wlld.imageRecognition.KMatrix; import org.wlld.imageRecognition.border.BorderBody; import org.wlld.imageRecognition.border.Frame; @@ -21,12 +22,21 @@ public class ModelParameter { private List dymNerveStudies = new ArrayList<>();//动态神经元隐层 private DymNerveStudy dymOutNerveStudy = new DymNerveStudy();//动态神经元输出层 private Map borderBodyMap = new HashMap<>();//border特征集合 + private Map kMatrixMap = new HashMap<>();//K均值矩阵集合 private Frame frame;//先验边框 public Map getBorderBodyMap() { return borderBodyMap; } + public Map getkMatrixMap() { + return kMatrixMap; + } + + public void setkMatrixMap(Map kMatrixMap) { + this.kMatrixMap = kMatrixMap; + } + public void setBorderBodyMap(Map borderBodyMap) { this.borderBodyMap = borderBodyMap; } diff --git a/src/main/java/org/wlld/nerveEntity/Nerve.java b/src/main/java/org/wlld/nerveEntity/Nerve.java index b7043e4..f041122 100644 --- a/src/main/java/org/wlld/nerveEntity/Nerve.java +++ b/src/main/java/org/wlld/nerveEntity/Nerve.java @@ -77,11 +77,11 @@ public abstract class Nerve { this.studyPoint = studyPoint; } - public void sendMessage(long enevtId, double parameter, boolean isStudy, Map E + public void sendMessage(long eventId, double parameter, boolean isStudy, Map E , OutBack outBack) throws Exception { if (son.size() > 0) { for (Nerve nerve : son) { - nerve.input(enevtId, parameter, isStudy, E, outBack); + nerve.input(eventId, parameter, isStudy, E, outBack); } } else { throw new Exception("this layer is lastIndex"); @@ -123,7 +123,7 @@ public abstract class Nerve { } public void sendMatrix(long eventId, Matrix parameter, boolean isStudy, - Map E, OutBack outBack) throws Exception { + int E, OutBack outBack) throws Exception { if (son.size() > 0) { for (Nerve nerve : son) { nerve.inputMatrix(eventId, parameter, isStudy, E, outBack); @@ -154,7 +154,7 @@ public abstract class Nerve { } - protected void inputMatrix(long eventId, Matrix matrix, boolean isKernelStudy, Map E, OutBack outBack) throws Exception {//输入动态矩阵 + protected void inputMatrix(long eventId, Matrix matrix, boolean isKernelStudy, int E, OutBack outBack) throws Exception {//输入动态矩阵 } private void backGetMessage(double parameter, long eventId) throws Exception {//反向传播 diff --git a/src/main/java/org/wlld/nerveEntity/OutNerve.java b/src/main/java/org/wlld/nerveEntity/OutNerve.java index d4c6c3d..58d6034 100644 --- a/src/main/java/org/wlld/nerveEntity/OutNerve.java +++ b/src/main/java/org/wlld/nerveEntity/OutNerve.java @@ -19,7 +19,6 @@ import java.util.Map; */ public class OutNerve extends Nerve { private Map matrixMapE;//主键与期望矩阵的映射 - private Matrix matrixF; public OutNerve(int id, int upNub, int downNub, double studyPoint, boolean init, ActiveFunction activeFunction, boolean isDynamic) throws Exception { @@ -58,22 +57,12 @@ public class OutNerve extends Nerve { @Override protected void inputMatrix(long eventId, Matrix matrix, boolean isKernelStudy - , Map E, OutBack outBack) throws Exception { + , int E, OutBack outBack) throws Exception { Matrix myMatrix = dynamicNerve(matrix, eventId, isKernelStudy); - if (matrixF == null) { - matrixF = new Matrix(myMatrix.getX(), myMatrix.getY()); - } if (isKernelStudy) {//回传 - for (Map.Entry entry : E.entrySet()) { - double g; - if (entry.getValue() > 0.5) {//正模板 - Matrix matrix1 = matrixMapE.get(entry.getKey()); - g = getGradient(myMatrix, matrix1); - } else { - g = getGradient(myMatrix, matrixF); - } - backMatrix(g, eventId); - }//所有训练集的卷积核训练结束,才需要再次训练全连接层 + Matrix matrix1 = matrixMapE.get(E); + double g = getGradient(myMatrix, matrix1); + backMatrix(g, eventId); } else {//卷积层输出 if (outBack != null) { outBack.getBackMatrix(myMatrix, eventId); diff --git a/src/main/java/org/wlld/nerveEntity/SensoryNerve.java b/src/main/java/org/wlld/nerveEntity/SensoryNerve.java index 2626ae1..6c1beba 100644 --- a/src/main/java/org/wlld/nerveEntity/SensoryNerve.java +++ b/src/main/java/org/wlld/nerveEntity/SensoryNerve.java @@ -25,7 +25,7 @@ public class SensoryNerve extends Nerve { } public void postMatrixMessage(long eventId, Matrix parameter, boolean isKernelStudy - , Map E, OutBack outBack) throws Exception { + , int E, OutBack outBack) throws Exception { sendMatrix(eventId, parameter, isKernelStudy, E, outBack); } diff --git a/src/test/java/org/wlld/HelloWorld.java b/src/test/java/org/wlld/HelloWorld.java index c060cbd..df86e44 100644 --- a/src/test/java/org/wlld/HelloWorld.java +++ b/src/test/java/org/wlld/HelloWorld.java @@ -8,11 +8,9 @@ import org.wlld.imageRecognition.Operation; import org.wlld.imageRecognition.Picture; import org.wlld.imageRecognition.TempleConfig; import org.wlld.imageRecognition.border.Frame; -import org.wlld.imageRecognition.border.FrameBody; import org.wlld.nerveEntity.ModelParameter; import java.util.HashMap; -import java.util.List; import java.util.Map; /** @@ -22,16 +20,49 @@ import java.util.Map; */ public class HelloWorld { public static void main(String[] args) throws Exception { - testPic(); + test(); + //testPic(); //testModel(); } + public static void test() throws Exception { + Picture picture = new Picture(); + TempleConfig templeConfig = new TempleConfig(); + ModelParameter modelParameter = JSONObject.parseObject(ModelData.DATA, ModelParameter.class); + templeConfig.init(StudyPattern.Accuracy_Pattern, true, 3204, 4032, 1); + templeConfig.insertModel(modelParameter); + Operation operation = new Operation(templeConfig); + for (int i = 1; i < 120; i++) {//神经网络学习 + System.out.println("study==" + i); + //读取本地URL地址图片,并转化成矩阵 + Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/c/c" + i + ".png"); + Matrix wrong = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/b/b" + i + ".png"); + //将图像矩阵和标注加入进行学习，Accuracy_Pattern 模式 进行第二次学习 + //第二次学习的时候，第三个参数必须是 true + operation.learning(right, 1, true); + operation.learning(wrong, 0, true); + } + templeConfig.clustering();//进行聚类 + + for (int j = 121; j < 140; j++) { + Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/c/c" + j + ".png"); + Matrix wrong = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/b/b" + j + ".png"); + int rightId = operation.toSee(right); + int wrongId = operation.toSee(wrong); + System.out.println("right==" + rightId); + System.out.println("wrong==" + wrongId); + } + ModelParameter modelParameter1 = templeConfig.getModel(); + String a = JSON.toJSONString(modelParameter1); + System.out.println(a); + } + public static void testPic() throws Exception { //测试SPEED模式学习过程 //初始化图像转矩阵类:作用就是说将一个图片文件转化为矩阵类 Picture picture = new Picture(); //初始化配置模板类,设置模式为SPEED_PATTERN模式 即速度模式 - TempleConfig templeConfig = getTemple(true, StudyPattern.Speed_Pattern); + TempleConfig templeConfig = getTemple(true, StudyPattern.Speed_Pattern, false); //初始化计算类，并将配置模版和输出回调类载入计算类 //运算类有两个构造一个是配置回调类，一个是不配置回调类 //若使用定位功能，则无需配置回调类，若不启用，则要配置回调类 @@ -46,7 +77,7 @@ public class HelloWorld { rightTagging.put(1, 1.0); wrongTagging.put(1, 0.0); // 例如上面的标注了 只有一种分类，第一个MAP是true标注，第二个map是false标注 - for (int i = 1; i < 999; i++) { + for (int i = 1; i < 1000; i++) { System.out.println("开始学习1==" + i); //读取本地URL地址图片(适用于电脑本地图片),并转化成矩阵 //注意学习图片至少要一千张+同物体的不同图片，学习的越多就越准，拿同样的图片反复循环学习是没用的 @@ -59,6 +90,15 @@ public class HelloWorld { //wrong这个矩阵是错误的图片，所以要配置上面错误的标注0.0 学习 告诉计算机这个图片是错误的 operation.study(wrong, wrongTagging); } + + Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/test/a101.png"); + Matrix wrong = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/b/b1000.png"); + ma.setNub(1); + operation.look(wrong, 3); + ma.setNub(0); + operation.look(right, 2); + + //如果启用物体坐标定位，则在学习结束的时候，一定要执行boxStudy方法 //若不启用，则请不要使用，否则会报错 //templeConfig.boxStudy(); @@ -68,32 +108,43 @@ public class HelloWorld { ModelParameter modelParameter = templeConfig.getModel(); //将模型MODEL转化成JSON 字符串 保存到数据库 留待下次服务启动的时候，识别提取用 String model = JSON.toJSONString(modelParameter); + System.out.println(model); //以上就是SPEED模式下的学习全过程，识别的过程就是再次初始化，将学习结果注入之后使用 //识别过程 //将从数据库取出的JSON字符串转化为模型MODEL - ModelParameter modelParameter1 = JSONObject.parseObject(model, ModelParameter.class); - //初始化模型配置 - TempleConfig templeConfig1 = getTemple(false, StudyPattern.Speed_Pattern); - //注入之前学习结果的模型MODEL到配置模版里面，将学习结果注入就可以使用识别了 - templeConfig1.insertModel(modelParameter1); - //将配置模板配置到运算类 - Operation operation1 = new Operation(templeConfig1); - //获取本地图片字节码转化成降纬后的灰度矩阵 - Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/test/a101.png"); - Matrix wrong = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/b/b1000.png"); - //进行图像识别 参数说明 eventId,事件id,因为输出结果是在回调类回调的，所以必须有个主键去判断事件 - //说明你回调是响应的哪一次调用的ID,所以每一次识别调用，请用不同的id - operation1.look(wrong, 3); - operation1.look(right, 2); +// ModelParameter modelParameter1 = JSONObject.parseObject(model, ModelParameter.class); +// //初始化模型配置 +// TempleConfig templeConfig1 = getTemple(false, StudyPattern.Speed_Pattern, false); +// //注入之前学习结果的模型MODEL到配置模版里面，将学习结果注入就可以使用识别了 +// templeConfig1.insertModel(modelParameter1); +// //将配置模板配置到运算类 +// Operation operation1 = new Operation(templeConfig1); +// //获取本地图片字节码转化成降纬后的灰度矩阵 +// Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/test/a101.png"); +// Matrix wrong = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/b/b1000.png"); +// //进行图像识别 参数说明 eventId,事件id,因为输出结果是在回调类回调的，所以必须有个主键去判断事件 +// //说明你回调是响应的哪一次调用的ID,所以每一次识别调用，请用不同的id +// operation1.look(wrong, 3); +// operation1.look(right, 2); //若启用定位功能检测请使用lookWithPosition,若没有启用，使用检测会报错 //返回map,主键是分类id,值是该图片中此分类有多少个物体，每个物体的具体位置坐标的大小 //Map> map = operation1.lookWithPosition(right, 4); } - public static TempleConfig getTemple(boolean isFirst, int pattern) throws Exception { + public static TempleConfig getTemple(boolean isFirst, int pattern + , boolean isPosition) throws Exception { //创建一个配置模板类，作用：主要是保存及载入一些配置参数用 TempleConfig templeConfig = new TempleConfig(); + if (isPosition) { + templeConfig.setHavePosition(true); + Frame frame = new Frame(); + frame.setWidth(3024); + frame.setHeight(4032); + frame.setLengthHeight(100); + frame.setLengthWidth(100); + templeConfig.setFrame(frame); + } //全连接层深度,选填可不填 不填默认值为2 //这就像人类大脑的意识深度原理一样，深度学习越深，训练结果越准，但是训练量成几何倍数增加 //比如默认深度是2 需要 正负模板各一千+张照片进行训练。识别率70%（数值只是举个例子，不是具体数值） @@ -101,7 +152,7 @@ public class HelloWorld { //以此类推，，内存允许的情况下，深度无限 识别率无限接近与百分之百 //但是有极限，即超过某个深度，即使再增加深度，识别率反而会下降。需要具体不断尝试找到 合适的深度 //注意：若深度提升，训练量没有成倍增长，则准确度反而更低！ - templeConfig.setDeep(2); + //templeConfig.setDeep(2); //启用定位学习 注意启用在图片中对某个物体进行定位，要注意 //学习的图片必须除了学习的物体以外，其他位置都是白色或者空白(即用PS扣空)。 //即该图片除了这个物体，没有其他任何干扰杂色（一个像素的杂色都不可以有） @@ -134,11 +185,11 @@ public class HelloWorld { public static void testModel() throws Exception { // 模型参数获取及注入 实例 - TempleConfig templeConfig = getTemple(true, StudyPattern.Accuracy_Pattern); + TempleConfig templeConfig = getTemple(true, StudyPattern.Accuracy_Pattern, false); ModelParameter modelParameter1 = templeConfig.getModel(); String model = JSON.toJSONString(modelParameter1); System.out.println(model); - TempleConfig templeConfig2 = getTemple(false, StudyPattern.Accuracy_Pattern); + TempleConfig templeConfig2 = getTemple(false, StudyPattern.Accuracy_Pattern, false); ModelParameter modelParameter3 = JSONObject.parseObject(model, ModelParameter.class); templeConfig2.insertModel(modelParameter3); ModelParameter modelParameter2 = templeConfig2.getModel(); @@ -150,14 +201,15 @@ public class HelloWorld { public static void testPic2() throws Exception { //测试Accuracy_Pattern 模式学习过程，跟SPEED模式相同的部分就不再说明了 Picture picture = new Picture(); - TempleConfig templeConfig = getTemple(true, StudyPattern.Accuracy_Pattern); - Operation operation = new Operation(templeConfig); + TempleConfig templeConfig = getTemple(true, StudyPattern.Accuracy_Pattern, false); + Ma ma = new Ma(); + Operation operation = new Operation(templeConfig, ma); //标注主键为 第几种分类，值为标注 1 是TRUE 0是FALSE Map rightTagging = new HashMap<>();//分类标注 Map wrongTagging = new HashMap<>();//分类标注 rightTagging.put(1, 1.0); wrongTagging.put(1, 0.0); - for (int i = 1; i < 2; i++) { + for (int i = 1; i < 100; i++) { System.out.println("开始学习1==" + i); //读取本地URL地址图片,并转化成矩阵 Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/c/c" + i + ".png"); @@ -166,24 +218,35 @@ public class HelloWorld { //这里使用learning方法，前两个参数与SPEED模式相同，多了一个第三个参数 //第一次学习的时候 这个参数必须是 false //最后一个参数id - operation.learning(right, rightTagging, false); - operation.learning(wrong, wrongTagging, false); + operation.learning(right, 1, false); + operation.learning(wrong, 0, false); } - for (int i = 1; i < 2; i++) {//神经网络学习 + for (int i = 1; i < 300; i++) {//神经网络学习 System.out.println("开始学习2==" + i); //读取本地URL地址图片,并转化成矩阵 Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/c/c" + i + ".png"); Matrix wrong = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/b/b" + i + ".png"); //将图像矩阵和标注加入进行学习，Accuracy_Pattern 模式 进行第二次学习 //第二次学习的时候，第三个参数必须是 true - operation.learning(right, rightTagging, true); - operation.learning(wrong, wrongTagging, true); + operation.learning(right, 1, true); + operation.learning(wrong, 0, true); + } + ModelParameter modelParameter = templeConfig.getModel(); + String st = JSON.toJSONString(modelParameter); + System.out.println(st); + for (int i = 300; i < 320; i++) {//神经网络学习 + //读取本地URL地址图片,并转化成矩阵 + Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/c/c" + i + ".png"); + Matrix wrong = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/b/b" + i + ".png"); + //将图像矩阵和标注加入进行学习，Accuracy_Pattern 模式 进行第二次学习 + //第二次学习的时候，第三个参数必须是 true + ma.setNub(1); + operation.look(right, 100 + i); + ma.setNub(0); + operation.look(wrong, 200 + i); } - Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/test/a101.png"); - Matrix wrong = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/b/b1000.png"); //进行图像识别，Accuracy_Pattern 模式学习结果获取和注入，跟SPEED模式一致 //若有疑问可以参考一下 testModel()方法 - operation.look(right, 2); - operation.look(wrong, 3); + } } diff --git a/src/test/java/org/wlld/Ma.java b/src/test/java/org/wlld/Ma.java index d690519..dca7a26 100644 --- a/src/test/java/org/wlld/Ma.java +++ b/src/test/java/org/wlld/Ma.java @@ -17,7 +17,8 @@ public class Ma implements OutBack { @Override public void getBack(double out, int id, long eventId) { - System.out.println("id==" + id + ",out==" + out + ",nub==" + nub); + System.out.println("id==" + id + ",out==" + out + ",nub==" + nub + + ",eid==" + eventId); } @Override