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

优化ann训练算法

Browse Source
devA
yuxue 5 years ago
parent 831961b871
commit fc4c79b419
1 changed files with 130 additions and 68 deletions
Show Stats Download Patch File Download Diff File

194
src/main/java/com/yuxue/train/ANNTrain.java
Unescape View File

@ -118,38 +118,59 @@ public class ANNTrain {
/**
* 按随机数，平移或者旋转样本文件
* 进行膨胀操作
* @param inMat
* @return
*/
public Mat getSyntheticImage(Mat inMat) {
public Mat dilate(Mat inMat) {
Mat result = inMat.clone();
Mat element = Imgproc.getStructuringElement(Imgproc.MORPH_RECT, new Size(2, 2));
Imgproc.dilate(inMat, result, element);
return result;
}
/**
* 进行腐蚀操作
* @param inMat
* @return
*/
public Mat erode(Mat inMat) {
Mat result = inMat.clone();
Mat element = Imgproc.getStructuringElement(Imgproc.MORPH_RECT, new Size(2, 2));
Imgproc.erode(inMat, result, element);
return result;
}
/**
* 随机数平移
* @param inMat
* @return
*/
public Mat randTranslate(Mat inMat) {
Random rand = new Random();
int rand_type = rand.nextInt(10000);
Mat result = inMat.clone();
if (rand_type % 2 == 0) {
int ran_x = rand.nextInt(10000) % 5 - 2; // 控制在-2~3个像素范围内
int ran_y = rand.nextInt(10000) % 5 - 2;
result = translateImg(result, ran_x, ran_y); // 平移
return translateImg(result, ran_x, ran_y);
}
} else if (rand_type % 2 != 0) {
float angle = (float) (rand.nextInt(10000) % 15 - 7); // 旋转角度控制在-7~8°范围内
result = rotateImg(result, angle); // 旋转
}
/*
//进行膨胀操作
Mat element1 = Imgproc.getStructuringElement(Imgproc.MORPH_RECT, new Size(1, 1));
Mat dstImage1;
Imgproc.dilate(inMat, result, element1);
//进行腐蚀操作
Mat element2 = Imgproc.getStructuringElement(Imgproc.MORPH_RECT, new Size(1, 1));
Mat dstImage2;
Imgproc.erode(inMat, result, element2);
*/
return result;
/**
* 随机数旋转
* @param inMat
* @return
*/
public Mat randRotate(Mat inMat) {
Random rand = new Random();
Mat result = inMat.clone();
float angle = (float) (rand.nextInt(10000) % 15 - 7); // 旋转角度控制在-7~8°范围内
return rotateImg(result, angle);
}
/**
* 平移
* @param img
@ -169,6 +190,7 @@ public class ANNTrain {
return dst;
}
/**
* 旋转角度
* @param source
@ -188,32 +210,36 @@ public class ANNTrain {
public void train(int _predictsize, int _neurons) {
Mat samples = new Mat(); // 使用push_back，行数列数不能赋初始值
Vector<Integer> trainingLabels = new Vector<Integer>();
Random rand = new Random();
// 加载数字及字母字符
for (int i = 0; i < Constant.numCharacter; i++) {
String str = DEFAULT_PATH + "learn/" + Constant.strCharacters[i];
Vector<String> files = new Vector<String>();
FileUtil.getFiles(str, files); // 文件名不能包含中文
int count = 100; // 控制每个字符，最多只允许有200个样本文件
int k = 0;
// System.out.println("数字+字母：\t" + files.size());
for (String filePath : files) {
Mat img = Imgcodecs.imread(filePath, 0);
int count = 100; // 控制从训练样本中，抽取指定数量的样本
for (int j = 0; j < count; j++) {
Mat img = Imgcodecs.imread(files.get(rand.nextInt(files.size() - 1)), 0);
Mat f = features(img, _predictsize);
samples.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
// 抽取1/3样本文件，平移或者旋转变换后，加入训练样本
if (k % 3 == 0) {
samples.push_back(features(getSyntheticImage(img), _predictsize));
trainingLabels.add(i); // 每一幅字符图片所对应的字符类别索引下标
}
k++;
// 增加随机平移样本
samples.push_back(features(randTranslate(img), _predictsize));
trainingLabels.add(i);
if(count <= 0) {
break;
}
count--;
// 增加随机旋转样本
samples.push_back(features(randRotate(img), _predictsize));
trainingLabels.add(i);
// 增加膨胀样本
/*samples.push_back(features(dilate(img), _predictsize));
trainingLabels.add(i); */
// 增加腐蚀样本
samples.push_back(features(erode(img), _predictsize));
trainingLabels.add(i);
}
}
@ -223,26 +249,28 @@ public class ANNTrain {
Vector<String> files = new Vector<String>();
FileUtil.getFiles(str, files);
int count = 50; // 控制每个字符，最多只允许有100个样本文件
int k = 0;
// System.out.println("汉字：\t" + files.size());
for (String filePath : files) {
Mat img = Imgcodecs.imread(filePath, 0);
int count = 100; // 控制从训练样本中，抽取指定数量的样本
for (int j = 0; j < count; j++) {
Mat img = Imgcodecs.imread(files.get(rand.nextInt(files.size() - 1)), 0);
Mat f = features(img, _predictsize);
samples.push_back(f);
trainingLabels.add(i + Constant.numCharacter); // 每一幅字符图片所对应的字符类别索引下标
trainingLabels.add(i + Constant.numCharacter);
// 抽取1/3样本文件，平移或者旋转变换后，加入训练样本
if (k % 3 == 0) {
samples.push_back(features(getSyntheticImage(img), _predictsize));
trainingLabels.add(i); // 每一幅字符图片所对应的字符类别索引下标
}
k++;
// 增加随机平移样本
samples.push_back(features(randTranslate(img), _predictsize));
trainingLabels.add(i + Constant.numCharacter);
if(count <= 0) {
break;
}
count--;
// 增加随机旋转样本
samples.push_back(features(randRotate(img), _predictsize));
trainingLabels.add(i + Constant.numCharacter);
// 增加膨胀样本
/*samples.push_back(features(dilate(img), _predictsize));
trainingLabels.add(i + Constant.numCharacter);*/
// 增加腐蚀样本
samples.push_back(features(erode(img), _predictsize));
trainingLabels.add(i + Constant.numCharacter);
}
}
@ -262,7 +290,7 @@ public class ANNTrain {
ann.clear();
Mat layers = new Mat(1, 3, CvType.CV_32F);
layers.put(0, 0, samples.cols()); // 样本数量
layers.put(0, 0, samples.cols()); // 样本特征数
layers.put(0, 1, _neurons); //
layers.put(0, 2, classes.cols()); // 字符数
@ -285,7 +313,7 @@ public class ANNTrain {
ann.clear();
ann = ANN_MLP.load(MODEL_PATH);
Vector<String> files = new Vector<String>();
FileUtil.getFiles(DEFAULT_PATH + "test/", files);
FileUtil.getFiles(DEFAULT_PATH + "test/", files); // 获取测试文件
String plate = "";
for (String string : files) {
@ -295,25 +323,59 @@ public class ANNTrain {
int index = 0;
double maxVal = -2;
Mat output = new Mat(1, Constant.numAll, CvType.CV_32F);
ann.predict(f, output); // 预测结果 // 可以考虑将样本进行平移、旋转、腐蚀等算法，进行多次预测，取最大值--未实现
ann.predict(f, output); // 预测结果
for (int j = 0; j < Constant.numAll; j++) {
double val = output.get(0, j)[0];
if (val > maxVal) {
maxVal = val;
index = j;
// 输出预测可能的值 -- 测试用
/*String charValue = "";
if (index < Constant.numCharacter) {
charValue = String.valueOf(Constant.strCharacters[index]);
} else {
String s = Constant.strChinese[index - Constant.numCharacter];
charValue = Constant.KEY_CHINESE_MAP.get(s);
}
System.out.println(string + "==>" + j + "\t\t" + charValue + "\t" + val);*/
}
}
// 随机平移
/*f = features(randTranslate(img), Constant.predictSize);
ann.predict(f, output); // 预测结果
for (int j = 0; j < Constant.numAll; j++) {
double val = output.get(0, j)[0];
if (val > maxVal) {
maxVal = val;
index = j;
}
}*/
// 随机旋转
/*f = features(randRotate(img), Constant.predictSize);
ann.predict(f, output); // 预测结果
for (int j = 0; j < Constant.numAll; j++) {
double val = output.get(0, j)[0];
if (val > maxVal) {
maxVal = val;
index = j;
}
}*/
// 膨胀
/*f = features(dilate(img), Constant.predictSize);
ann.predict(f, output); // 预测结果
for (int j = 0; j < Constant.numAll; j++) {
double val = output.get(0, j)[0];
if (val > maxVal) {
maxVal = val;
index = j;
}
}*/
// 腐蚀 -- 识别中文字符效果会好一点，识别数字及字母效果会更差
/*f = features(erode(img), Constant.predictSize);
ann.predict(f, output); // 预测结果
for (int j = 0; j < Constant.numAll; j++) {
double val = output.get(0, j)[0];
if (val > maxVal) {
maxVal = val;
index = j;
}
}*/
if (index < Constant.numCharacter) {
plate += String.valueOf(Constant.strCharacters[index]);
} else {
@ -331,7 +393,7 @@ public class ANNTrain {
// 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
// 可根据需要训练不同的predictSize或者neurons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。
annT.train(Constant.predictSize, Constant.neurons);
// annT.train(Constant.predictSize, Constant.neurons);
annT.predict();

Write Preview
Loading…
Cancel
Save