p86743915
/
pp6hrvtlm
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

提交一个备份版本

Browse Source
devA
yuxue 5 years ago
parent b6154ef642
commit 591dc5e9a1
1 changed files with 132 additions and 166 deletions
Show Stats Download Patch File Download Diff File

292
src/main/java/com/yuxue/train/ANNTrain.java
Unescape View File

@ -1,30 +1,33 @@
package com.yuxue.train;
import static org.bytedeco.javacpp.opencv_core.CV_32F;
import static org.bytedeco.javacpp.opencv_core.CV_32FC1;
import static org.bytedeco.javacpp.opencv_core.CV_32SC1;
import static org.bytedeco.javacpp.opencv_core.getTickCount;
import static org.bytedeco.javacpp.opencv_imgproc.resize;
import static org.bytedeco.javacpp.opencv_ml.ROW_SAMPLE;
import java.util.Vector;
import org.bytedeco.javacpp.opencv_core.FileStorage;
import org.bytedeco.javacpp.opencv_core.Mat;
import org.bytedeco.javacpp.opencv_core.Scalar;
import org.bytedeco.javacpp.opencv_core.Size;
import org.bytedeco.javacpp.opencv_core.TermCriteria;
import org.bytedeco.javacpp.opencv_core;
import org.bytedeco.javacpp.opencv_imgcodecs;
import org.bytedeco.javacpp.opencv_ml.ANN_MLP;
import org.bytedeco.javacpp.opencv_ml.TrainData;
import com.yuxue.easypr.core.CoreFunc;
import com.yuxue.enumtype.Direction;
import com.yuxue.util.Convert;
import com.yuxue.util.FileUtil;
/**
* 基于org.bytedeco.javacpp包实现的训练
*
* 图片文字识别训练
* 训练出来的库文件，用于判断切图是否包含车牌
*
* 训练的svm.xml应用：
* 1、替换res/model/svm.xml文件
* 2、修改com.yuxue.easypr.core.PlateJudge.plateJudge(Mat) 方法
* 将样本处理方法切换一下，即将对应被注释掉的模块代码取消注释
*
* @author yuxue
* @date 2020-05-14 22:16
@ -33,171 +36,110 @@ public class ANNTrain {
private ANN_MLP ann = ANN_MLP.create();
// 中国车牌
private final char strCharacters[] = { '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', /* 没有I */
'J', 'K', 'L', 'M', 'N', /* 没有O */'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z' };
private final int numCharacter = 34; /* 没有I和0,10个数字与24个英文字符之和 */
// 以下都是我训练时用到的中文字符数据，并不全面，有些省份没有训练数据所以没有字符
// 有些后面加数字2的表示在训练时常看到字符的一种变形，也作为训练数据存储
private final String strChinese[] = { "zh_cuan" /* 川 */, "zh_e" /* 鄂 */, "zh_gan" /* 赣 */, "zh_hei" /* 黑 */, "zh_hu" /* 沪 */, "zh_ji" /* 冀 */, "zh_jl" /* 吉 */, "zh_jin" /* 津 */, "zh_jing" /* 京 */,
"zh_shan" /* 陕 */, "zh_liao" /* 辽 */, "zh_lu" /* 鲁 */, "zh_min" /* 闽 */, "zh_ning" /* 宁 */, "zh_su" /* 苏 */, "zh_sx" /* 晋 */, "zh_wan" /* 皖 */, "zh_yu" /* 豫 */, "zh_yue" /* 粤 */,
"zh_zhe" /* 浙 */ };
private final int numAll = 54; /* 34+20=54 */
public Mat features(Mat in, int sizeData) {
// Histogram features
float[] vhist = CoreFunc.projectedHistogram(in, Direction.VERTICAL);
float[] hhist = CoreFunc.projectedHistogram(in, Direction.HORIZONTAL);
// Low data feature
Mat lowData = new Mat();
resize(in, lowData, new Size(sizeData, sizeData));
// Last 10 is the number of moments components
int numCols = vhist.length + hhist.length + lowData.cols() * lowData.cols();
Mat out = Mat.zeros(1, numCols, CV_32F).asMat();
// Asign values to feature,ANN的样本特征为水平、垂直直方图和低分辨率图像所组成的矢量
int j = 0;
for (int i = 0; i < vhist.length; i++, ++j) {
out.ptr(j).put(Convert.getBytes(vhist[i]));
}
for (int i = 0; i < hhist.length; i++, ++j) {
out.ptr(j).put(Convert.getBytes(hhist[i]));
}
for (int x = 0; x < lowData.cols(); x++) {
for (int y = 0; y < lowData.rows(); y++, ++j) {
float val = lowData.ptr(x, y).get() & 0xFF;
out.ptr(j).put(Convert.getBytes(val));
}
}
return out;
}
/**
*
* @param TrainData 训练样本数据
* @param classes 数据对应的标签
* @param nNeruns
*/
public void annTrain(Mat trainingData, Mat classes, int nNeruns) {
ann.clear();
Mat layers = new Mat(1, 3, CV_32SC1);
layers.ptr(0).put(Convert.getBytes(trainingData.cols()));
layers.ptr(1).put(Convert.getBytes(nNeruns));
layers.ptr(2).put(Convert.getBytes(numAll));
// Prepare trainClases
// Create a mat with n trained data by m classes
Mat trainClasses = new Mat();
trainClasses.create(trainingData.rows(), numAll, CV_32FC1);
for (int i = 0; i < trainClasses.rows(); i++) {
for (int k = 0; k < trainClasses.cols(); k++) {
// If class of data i is same than a k class
if (k == Convert.toInt(classes.ptr(i)))
trainClasses.ptr(i, k).put(Convert.getBytes(1f));
else
trainClasses.ptr(i, k).put(Convert.getBytes(0f));
}
}
Mat weights = new Mat(1, trainingData.rows(), CV_32FC1, Scalar.all(1));
// Learn classifier
TrainData train_data = TrainData.create(trainingData, ROW_SAMPLE, trainClasses);
// 中国车牌; 34个字符; 没有 字母I、字母O
private final char strCharacters[] = { '0', '1', '2', '3', '4', '5', '6', '7', '8', '9',
'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', /* 没有I */ 'J', 'K', 'L', 'M', 'N', /* 没有O */'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z' };
/*
ann_->setTrainMethod(cv::ml::ANN_MLP::TrainingMethods::BACKPROP);
ann_->setTermCriteria(cvTermCriteria(CV_TERMCRIT_ITER, 30000, 0.0001));
ann_->setBackpropWeightScale(0.1);
ann_->setBackpropMomentumScale(0.1);*/
ann.setLayerSizes(layers);
ann.setActivationFunction(ANN_MLP.SIGMOID_SYM, 1, 1);
ann.setTrainMethod(ANN_MLP.BACKPROP);
TermCriteria criteria = new TermCriteria(TermCriteria.MAX_ITER, 30000, 0.0001);
ann.setTermCriteria(criteria);
ann.setBackpropWeightScale(0.1);
ann.setBackpropMomentumScale(0.1);
ann.train(train_data);
}
private final int numCharacter = strCharacters.length;
public int saveTrainData() {
// 并不全面，有些省份没有训练数据所以没有字符
// 有些后面加数字2的表示在训练时常看到字符的一种变形，也作为训练数据存储
private final String strChinese[] = {
"zh_cuan", /*川*/
"zh_e", /*鄂*/
"zh_gan", /*赣*/
"zh_gan1", /*甘*/
"zh_gui", /*贵*/
"zh_gui1", /*桂*/
"zh_hei", /*黑*/
"zh_hu", /*沪*/
"zh_ji", /*冀*/
"zh_jin", /*津*/
"zh_jing", /*京*/
"zh_jl", /*吉*/
"zh_liao", /*辽*/
"zh_lu", /*鲁*/
"zh_meng", /*蒙*/
"zh_min", /*闽*/
"zh_ning", /*宁*/
"zh_qing", /*青*/
"zh_qiong", /*琼*/
"zh_shan", /*陕*/
"zh_su", /*苏*/
"zh_sx", /*晋*/
"zh_wan", /*皖*/
"zh_xiang", /*湘*/
"zh_xin", /*新*/
"zh_yu", /*豫*/
"zh_yu1", /*渝*/
"zh_yue", /*粤*/
"zh_yun", /*云*/
"zh_zang", /*藏*/
"zh_zhe" /*浙*/
};
private final int numAll = strCharacters.length + strChinese.length;
// 默认的训练操作的根目录
private static final String DEFAULT_PATH = "D:/PlateDetect/train/chars_recognise_ann/";
// 训练模型文件保存位置
private static final String DATA_PATH = DEFAULT_PATH + "ann_data.xml";
private static final String MODEL_PATH = DEFAULT_PATH + "ann.xml";
public int saveTrainData(int _predictsize) {
System.out.println("Begin saveTrainData");
Mat classes = new Mat();
Mat trainingDataf5 = new Mat();
Mat trainingDataf10 = new Mat();
Mat trainingDataf15 = new Mat();
Mat trainingDataf20 = new Mat();
Mat trainingDataf = new Mat();
Vector<Integer> trainingLabels = new Vector<Integer>();
String path = "res/train/data/chars_recognise_ann/chars2/chars2";
// 加载数字及字母字符
for (int i = 0; i < numCharacter; i++) {
System.out.println("Character: " + strCharacters[i]);
String str = path + '/' + strCharacters[i];
String str = DEFAULT_PATH + strCharacters[i];
System.err.println(str);
Vector<String> files = new Vector<String>();
FileUtil.getFiles(str, files);
int size = (int) files.size();
for (int j = 0; j < size; j++) {
// System.out.println(files.get(j));
Mat img = opencv_imgcodecs.imread(files.get(j), 0);
Mat f5 = features(img, 5);
Mat f10 = features(img, 10);
Mat f15 = features(img, 15);
Mat f20 = features(img, 20);
trainingDataf5.push_back(f5);
trainingDataf10.push_back(f10);
trainingDataf15.push_back(f15);
trainingDataf20.push_back(f20);
// System.err.println(files.get(j)); // 文件名不能包含中文
Mat f = CoreFunc.features(img, _predictsize);
trainingDataf.push_back(f);
trainingLabels.add(i); // 每一幅字符图片所对应的字符类别索引下标
}
}
path = "res/train/data/chars_recognise_ann/charsChinese/charsChinese";
// 加载汉字字符
for (int i = 0; i < strChinese.length; i++) {
System.out.println("Character: " + strChinese[i]);
String str = path + '/' + strChinese[i];
String str = DEFAULT_PATH + strChinese[i];
Vector<String> files = new Vector<String>();
FileUtil.getFiles(str, files);
int size = (int) files.size();
for (int j = 0; j < size; j++) {
// System.out.println(files.get(j));
Mat img = opencv_imgcodecs.imread(files.get(j), 0);
Mat f5 = features(img, 5);
Mat f10 = features(img, 10);
Mat f15 = features(img, 15);
Mat f20 = features(img, 20);
trainingDataf5.push_back(f5);
trainingDataf10.push_back(f10);
trainingDataf15.push_back(f15);
trainingDataf20.push_back(f20);
// System.err.println(files.get(j)); // 文件名不能包含中文
Mat f = CoreFunc.features(img, _predictsize);
trainingDataf.push_back(f);
trainingLabels.add(i + numCharacter);
}
}
trainingDataf5.convertTo(trainingDataf5, CV_32FC1);
trainingDataf10.convertTo(trainingDataf10, CV_32FC1);
trainingDataf15.convertTo(trainingDataf15, CV_32FC1);
trainingDataf20.convertTo(trainingDataf20, CV_32FC1);
// CV_32FC1 CV_32SC1 CV_32F
trainingDataf.convertTo(trainingDataf, opencv_core.CV_32F);
int[] labels = new int[trainingLabels.size()];
for (int i = 0; i < labels.length; ++i)
for (int i = 0; i < labels.length; ++i) {
labels[i] = trainingLabels.get(i).intValue();
}
new Mat(labels).copyTo(classes);
FileStorage fs = new FileStorage("res/train/ann_data.xml", FileStorage.WRITE);
fs.write("TrainingDataF5", trainingDataf5);
fs.write("TrainingDataF10", trainingDataf10);
fs.write("TrainingDataF15", trainingDataf15);
fs.write("TrainingDataF20", trainingDataf20);
FileStorage fs = new FileStorage(DATA_PATH, FileStorage.WRITE);
fs.write("TrainingDataF" + _predictsize, trainingDataf);
fs.write("classes", classes);
fs.release();
@ -205,44 +147,68 @@ public class ANNTrain {
return 0;
}
public void saveModel(int _predictsize, int _neurons) {
public void train(int _predictsize, int _neurons) {
// 读取样本文件数据
FileStorage fs = new FileStorage(DATA_PATH, FileStorage.READ);
Mat samples = new Mat(fs.get("TrainingDataF" + _predictsize));
Mat classes = new Mat(fs.get("classes"));
Mat trainClasses = new Mat(samples.rows(), numAll, CV_32F);
for (int i = 0; i < trainClasses.rows(); i++) {
for (int k = 0; k < trainClasses.cols(); k++) {
// If class of data i is same than a k class
if (k == Convert.toInt(classes.ptr(i))) {
trainClasses.ptr(i, k).put(Convert.getBytes(1f));
} else {
trainClasses.ptr(i, k).put(Convert.getBytes(0f));
}
}
}
samples.convertTo(samples, CV_32F);
// 样本文件数据已经保存到xml
FileStorage fs = new FileStorage("res/train/ann_data.xml", FileStorage.READ);
String training = "TrainingDataF" + _predictsize;
Mat TrainingData = new Mat(fs.get(training));
Mat Classes = new Mat(fs.get("classes"));
System.out.println(samples.type());
long start = getTickCount();
annTrain(TrainingData, Classes, _neurons);
long end = getTickCount();
System.out.println("完成耗时: " + Long.valueOf((end - start) / 1000).toString());
// samples.type() == CV_32F || samples.type() == CV_32S
TrainData train_data = TrainData.create(samples, ROW_SAMPLE, trainClasses);
String model_name = "res/train/ann.xml";
ann.clear();
Mat layers = new Mat(1, 3, CV_32F);
layers.ptr(0).put(Convert.getBytes(samples.cols()));
layers.ptr(1).put(Convert.getBytes(_neurons));
layers.ptr(2).put(Convert.getBytes(numAll));
FileStorage fsto = new FileStorage(model_name, FileStorage.WRITE);
ann.setLayerSizes(layers);
ann.setActivationFunction(ANN_MLP.SIGMOID_SYM, 1, 1);
ann.setTrainMethod(ANN_MLP.BACKPROP);
TermCriteria criteria = new TermCriteria(TermCriteria.MAX_ITER, 30000, 0.0001);
ann.setTermCriteria(criteria);
ann.setBackpropWeightScale(0.1);
ann.setBackpropMomentumScale(0.1);
ann.train(train_data);
System.err.println("完成 ");
FileStorage fsto = new FileStorage(MODEL_PATH, FileStorage.WRITE);
ann.write(fsto, "ann");
}
public int annMain() {
System.out.println("To be begin.");
saveTrainData();
public static void main(String[] args) {
ANNTrain annT = new ANNTrain();
// 可根据需要训练不同的predictSize或者neurons的ANN模型
/*for (int i = 2; i <= 2; i++) {
int size = i * 5;
for (int j = 5; j <= 10; j++) {
int neurons = j * 10;
saveModel(size, neurons);
}
}*/
int _predictsize = 10;
int _neurons = 40;
// annT.saveTrainData(_predictsize);
// 这里演示只训练model文件夹下的ann.xml，此模型是一个predictSize=10,neurons=40的ANN模型。
// 根据机器的不同，训练时间不一样，但一般需要10分钟左右，所以慢慢等一会吧。
saveModel(10, 40);
annT.train(_predictsize, _neurons);
System.out.println("To be end.");
return 0;
}
}
Write Preview
Loading…
Cancel
Save