m53297601
/
easyAi
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

修改特征向量量化BUG

Browse Source
pull/1/head
lidapeng 5 years ago
parent f1cf3de0f2
commit 942e651221
6 changed files with 104 additions and 43 deletions
Show Stats Download Patch File Download Diff File

9
src/main/java/org/wlld/MatrixTools/MatrixOperation.java
Unescape View File

@ -212,8 +212,13 @@ public class MatrixOperation {
public static double getNorm(Matrix matrix) throws Exception {//求向量范数
if (matrix.getY() == 1 || matrix.getX() == 1) {
Matrix matrix1 = transPosition(matrix);
return Math.sqrt(mulMatrix(matrix1, matrix).getNumber(0, 0));
double nub = 0;
for (int i = 0; i < matrix.getX(); i++) {
for (int j = 0; j < matrix.getY(); j++) {
nub = ArithUtil.add(Math.pow(matrix.getNumber(i, j), 2), nub);
}
}
return Math.sqrt(nub);
} else {
throw new Exception("this matrix is not vector");
}

33
src/main/java/org/wlld/imageRecognition/Operation.java
Unescape View File

@ -5,10 +5,7 @@ import org.wlld.MatrixTools.Matrix;
import org.wlld.MatrixTools.MatrixOperation;
import org.wlld.config.StudyPattern;
import org.wlld.i.OutBack;
import org.wlld.imageRecognition.border.Border;
import org.wlld.imageRecognition.border.BorderBody;
import org.wlld.imageRecognition.border.Frame;
import org.wlld.imageRecognition.border.FrameBody;
import org.wlld.imageRecognition.border.*;
import org.wlld.nerveEntity.SensoryNerve;
import org.wlld.tools.ArithUtil;
@ -80,10 +77,17 @@ public class Operation {//进行计算
intoNerve2(1, matrix, templeConfig.getConvolutionNerveManager().getSensoryNerves(),
isKernelStudy, tagging, matrixBack);
if (isNerveStudy) {
//卷积后的结果
Matrix myMatrix = matrixBack.getMatrix();
if (templeConfig.isHavePosition() && tagging > 0) {
border.end(myMatrix, tagging);
}
LVQ lvq = templeConfig.getLvq();
Matrix vector = MatrixOperation.matrixToVector(myMatrix, true);
MatrixBody matrixBody = new MatrixBody();
matrixBody.setMatrix(vector);
matrixBody.setId(tagging);
lvq.insertMatrixBody(matrixBody);
//进行聚类
Map<Integer, KMatrix> kMatrixMap = templeConfig.getkMatrixMap();
if (kMatrixMap.containsKey(tagging)) {
@ -283,12 +287,31 @@ public class Operation {//进行计算
intoNerve2(2, matrix, templeConfig.getConvolutionNerveManager().getSensoryNerves(),
false, -1, matrixBack);
Matrix myMatrix = matrixBack.getMatrix();
return getClassificationId(myMatrix);
Matrix vector = MatrixOperation.matrixToVector(myMatrix, true);
return getClassificationId2(vector);
} else {
throw new Exception("pattern is wrong");
}
}
private int getClassificationId2(Matrix myVector) throws Exception {
int id = 0;
double distEnd = 0;
LVQ lvq = templeConfig.getLvq();
MatrixBody[] matrixBodies = lvq.getModel();
for (int i = 0; i < matrixBodies.length; i++) {
MatrixBody matrixBody = matrixBodies[i];
Matrix vector = matrixBody.getMatrix();
double dist = lvq.vectorEqual(myVector, vector);
if (distEnd == 0 || dist < distEnd) {
id = matrixBody.getId();
distEnd = dist;
}
}
return id;
}
private int getClassificationId(Matrix myMatrix) throws Exception {
int id = 0;
double distEnd = 0;

20
src/main/java/org/wlld/imageRecognition/TempleConfig.java
Unescape View File

@ -7,6 +7,7 @@ import org.wlld.function.ReLu;
import org.wlld.function.Sigmod;
import org.wlld.imageRecognition.border.BorderBody;
import org.wlld.imageRecognition.border.Frame;
import org.wlld.imageRecognition.border.LVQ;
import org.wlld.nerveCenter.NerveManager;
import org.wlld.nerveEntity.ModelParameter;
import org.wlld.nerveEntity.SensoryNerve;
@ -30,10 +31,20 @@ public class TempleConfig {
private boolean isHavePosition = false;//是否需要锁定物体位置
private Map<Integer, BorderBody> borderBodyMap = new HashMap<>();//border特征集合
private Map<Integer, KMatrix> kMatrixMap = new HashMap<>();//K均值矩阵集合
private LVQ lvq;
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
private double iouTh = 0.5;//IOU阈值
private int lvqNub = 50;//lvq循环次数，默认50
public int getLvqNub() {
return lvqNub;
}
public void setLvqNub(int lvqNub) {
this.lvqNub = lvqNub;
}
public double getIouTh() {
return iouTh;
@ -77,6 +88,14 @@ public class TempleConfig {
}
}
public void startLvq() throws Exception {//进行量化
lvq.start();
}
public LVQ getLvq() {
return lvq;
}
private void border(BorderBody borderBody) throws Exception {
Matrix parameter = borderBody.getX();//参数矩阵
Matrix tx = borderBody.getTx();
@ -166,6 +185,7 @@ public class TempleConfig {
private void initConvolutionVision(boolean initPower, int width, int height) throws Exception {
int deep = 0;
lvq = new LVQ(classificationNub + 1, lvqNub);
Map<Integer, Matrix> matrixMap = new HashMap<>();//主键与期望矩阵的映射
while (width > 5 && height > 5) {
width = width / 3;

3
src/main/java/org/wlld/imageRecognition/border/KClustering.java
Unescape View File

@ -15,7 +15,7 @@ public class KClustering {
private List<MatrixBody> matrixList = new ArrayList<>();//聚类集合
private int length;//向量长度
private int speciesQuantity;//种类数量
private Matrix[] matrices = new Matrix[speciesQuantity];//均值K
private Matrix[] matrices;//均值K
private Map<Integer, List<MatrixBody>> clusterMap = new HashMap<>();//簇
public Matrix[] getMatrices() {
@ -28,6 +28,7 @@ public class KClustering {
public KClustering(int speciesQuantity) {
this.speciesQuantity = speciesQuantity;
matrices = new Matrix[speciesQuantity];
for (int i = 0; i < speciesQuantity; i++) {
clusterMap.put(i, new ArrayList<>());
}

51
src/main/java/org/wlld/imageRecognition/border/LVQ.java
Unescape View File

@ -2,7 +2,6 @@ package org.wlld.imageRecognition.border;
import org.wlld.MatrixTools.Matrix;
import org.wlld.MatrixTools.MatrixOperation;
import org.wlld.tools.ArithUtil;
import java.util.ArrayList;
import java.util.List;
@ -15,16 +14,27 @@ import java.util.Random;
*/
public class LVQ {
private int typeNub;//原型聚类个数,即分类个数
private MatrixBody[] model = new MatrixBody[typeNub];//原型向量
private MatrixBody[] model;//原型向量
private List<MatrixBody> matrixList = new ArrayList<>();
private double studyPoint = 0.1;//量化学习率
private int length;//向量长度
private boolean isReady = false;
private int lvqNub = 50;
public LVQ(int typeNub) {
public boolean isReady() {
return isReady;
}
public LVQ(int typeNub, int lvqNub) {
this.typeNub = typeNub;
this.lvqNub = lvqNub;
model = new MatrixBody[typeNub];
}
public MatrixBody[] getModel() {
public MatrixBody[] getModel() throws Exception {
if (!isReady) {
throw new Exception("not study");
}
return model;
}
@ -46,31 +56,32 @@ public class LVQ {
}
}
private double study() throws Exception {
double error = 0;
private void study() throws Exception {
for (MatrixBody matrixBody : matrixList) {
Matrix matrix = matrixBody.getMatrix();//特征向量
long type = matrixBody.getId();//类别
double distEnd = 0;
int id = 0;
for (int i = 0; i < typeNub; i++) {
MatrixBody modelBody = model[i];
Matrix modelMatrix = modelBody.getMatrix();
long id = modelBody.getId();
boolean isRight = id == type;//类别是否相同
//对矩阵进行修正
Matrix matrix1 = op(matrix, modelMatrix, isRight);
//修正矩阵与原矩阵的范数差
double dist = vectorEqual(modelMatrix, matrix1);
//将修正后的向量进行赋值
modelBody.setMatrix(matrix1);
//统计变化值
error = ArithUtil.add(error, dist);
double dist = vectorEqual(modelMatrix, matrix);
if (distEnd == 0 || dist < distEnd) {
id = matrixBody.getId();
distEnd = dist;
}
}
MatrixBody modelBody = model[id];
Matrix modelMatrix = modelBody.getMatrix();
boolean isRight = id == type;
Matrix matrix1 = op(matrix, modelMatrix, isRight);
modelBody.setMatrix(matrix1);
}
return error;
}
//比较两个向量之间的范数差
private double vectorEqual(Matrix matrix1, Matrix matrix2) throws Exception {
public double vectorEqual(Matrix matrix1, Matrix matrix2) throws Exception {
Matrix matrix = MatrixOperation.sub(matrix1, matrix2);
return MatrixOperation.getNorm(matrix);
}
@ -100,9 +111,9 @@ public class LVQ {
model[i] = matrixBody;
}
//初始化完成
for (int k = 0; k < 1000; k++) {
double error = study();
System.out.println("error==" + error);
for (int i = 0; i < lvqNub; i++) {
study();
}
isReady = true;
}
}

31
src/test/java/org/wlld/HelloWorld.java
Unescape View File

@ -30,18 +30,18 @@ public class HelloWorld {
public static void test() throws Exception {
Picture picture = new Picture();
TempleConfig templeConfig = new TempleConfig();
templeConfig.setHavePosition(true);
Frame frame = new Frame();
frame.setWidth(3024);
frame.setHeight(4032);
frame.setLengthHeight(100);
frame.setLengthWidth(100);
templeConfig.setFrame(frame);
//templeConfig.setHavePosition(true);
// Frame frame = new Frame();
// frame.setWidth(3024);
// frame.setHeight(4032);
// frame.setLengthHeight(100);
// frame.setLengthWidth(100);
// templeConfig.setFrame(frame);
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 < 20; i++) {//faster rcnn神经网络学习
for (int i = 1; i < 30; i++) {//faster rcnn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");
@ -51,16 +51,17 @@ public class HelloWorld {
operation.learning(right, 1, true);
operation.learning(wrong, 0, true);
}
templeConfig.boxStudy();//边框回归
templeConfig.clustering();//进行聚类
templeConfig.startLvq();
//templeConfig.boxStudy();//边框回归
//templeConfig.clustering();//进行聚类
// ModelParameter modelParameter1 = templeConfig.getModel();
// String a = JSON.toJSONString(modelParameter1);
// System.out.println(a);
for (int j = 1; j < 2; j++) {
Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/c/c" + j + ".png");
Map<Integer, List<FrameBody>> map = operation.lookWithPosition(right, j);
System.out.println("j===" + j);
}
// for (int j = 1; j < 2; j++) {
// Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/c/c" + j + ".png");
// Map<Integer, List<FrameBody>> map = operation.lookWithPosition(right, j);
// System.out.println("j===" + j);
// }
//测试集图片,进行识别测试
// for (int j = 121; j < 140; j++) {
// Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/c/c" + j + ".png");

Write Preview
Loading…
Cancel
Save