基于SVM+HOG的花生品种识别

早在大二时，导师做过一个花生品种识别的程序，但当时用于品种识别的特征是自定义特征；而我的当初的本科毕业设计则是基于SVM + HOG的人脸识别；在后来的工作学习中，又用到了HOG特征，想着是不是之前的花生品种的识别也能用Hog特征进行识别；正所谓学以致用，也能巩固下刚刚所学的知识；

所谓的Hog特征与SVM网上资料一大堆，而且讲解得非常详细，在这里不在做过多的讲解；

本程序的开发环境，主要是依赖于VS2013 + OpenCV2.4.9，开发环境，请自行配置，那么，下面贴出代码，不足之处，请各位大侠不吝批评指正：

特征提取功能的实现：

CvMat* dataMat = NULL;
Book* book = xlCreateBook();
Sheet* sheet = book->addSheet("Sheet1");

for (int j = 1; j <= 13; ++ j)
{
CString strLoadPath = "SubImage\\";
CString strFile;
strFile.Format("s%d", j);
strLoadPath = strLoadPath + strFile + "\\";

for (int i = 1; i <= 45; ++i)
{
CString loadPath = strLoadPath;
CString stri;
stri.Format("%d.jpg", i);
loadPath = loadPath + stri;

IplImage* srcImage = cvLoadImage(loadPath);
IplImage* grayImage = cvCreateImage(cvGetSize(srcImage), IPL_DEPTH_8U, 1);
IplImage* hogImage = cvCreateImage(cvSize(96, 64), IPL_DEPTH_8U, 1);		//用于提取Hog特征的图像

cvCvtColor(srcImage, grayImage, CV_RGB2GRAY);

cvResize(grayImage, hogImage, CV_INTER_LINEAR);   //线性插值

std::vector<float> vfDescriptors;
//vfDescriptors.resize(hogImage->width * hogImage->height);
cv::Ptr<cv::HOGDescriptor> hog = new cv::HOGDescriptor(cvSize(hogImage->width, hogImage->height), cvSize(16, 16), cvSize(8, 8), cvSize(8, 8), 9);
hog->compute(hogImage, vfDescriptors, cv::Size(1, 1), cv::Size(0, 0));		//使用计算函数进行计算

//CvMat* mat = cvCreateMat(45 * 13, vfDescriptors.size(), CV_32FC1);
//cvZero(mat);

int cols = 0;

for (auto it = vfDescriptors.begin(); it != vfDescriptors.end(); ++it)
{
if (sheet)
{
sheet->writeNum(45 * (j-1) + (i - 1), cols++, *it);    //把Hog数据存储到Excel 中
}
}

//释放内存
cvReleaseImage(&srcImage);
srcImage = NULL;
cvReleaseImage(&grayImage);
grayImage = NULL;
cvReleaseImage(&hogImage);
hogImage = NULL;

}// 内层 for 循环

}

book->save("Hog.xls");
book->release();

SVM训练的实现代码：

Book* book = xlCreateBook();

if (book->load("Hog.xls"))
{
Sheet* sheet = book->getSheet(0);

if (sheet)
{

CvMat* dataMat = cvCreateMat(sheet->lastRow(), sheet->lastCol(), CV_32FC1);		//存储导入的数据

for (int i = 0; i < sheet->lastRow(); ++ i)
{
for (int j = 0; j < sheet->lastCol(); ++ j)
{
double temp = sheet->readNum(i, j);
cvSetReal2D(dataMat, i, j, temp);
}
}

MessageBox("数据导入完成");

CvMat* lableMat = cvCreateMat(sheet->lastRow(), 1, CV_32FC1);
cvZero(lableMat);

for (int i = 0; i < 13; ++ i)
{
for (int j = 0; j < 45; ++ j)
{
cvSetReal2D(lableMat, i * 45 + j, 0, i + 1);
}
}

CvSVM svm;
CvSVMParams svmParams;						//向量机参数
CvTermCriteria  svmCriteria;				//迭代中止条件
svmCriteria = cvTermCriteria(CV_TERMCRIT_EPS, 1000, FLT_EPSILON);

svmParams = CvSVMParams(CvSVM::C_SVC, CvSVM::RBF, 10.0, 8.0, 1.0, 10.0, 0.5, 0.1, NULL, svmCriteria);

//利用训练数据和确定的学习参数，进行SVM学习
svm.train(dataMat, lableMat, NULL, NULL, svmParams);
svm.save("svm.xml");

//以下代码用于SVM识别
CvMat* svmResult = cvCreateMat(sheet->lastRow(), 1, CV_32FC1);
CvMat* svmRow = NULL;

for (int i = 0; i < sheet->lastRow(); ++ i)
{
svmRow = cvCreateMat(1, sheet->lastCol(), CV_32FC1);

for (int j = 0; j < sheet->lastCol(); ++ j)
{
float temp = cvGetReal2D(dataMat, i, j);
cvSetReal2D(svmRow, 0, j, temp);
}

unsigned int ret = 0;
ret = svm.predict(svmRow);
cvSetReal2D(svmResult, i, 0, ret);

cvReleaseMat(&svmRow);
svmRow = NULL;
}

int nCount = 0;

for (int i = 0; i < 13; ++i)
{
for (int j = 0; j < 45; ++ j)
{
int ret = cvGetReal2D(svmResult, i * 45 + j, 0);
if (ret == (i + 1))
{
++nCount;
}
}
}

float recognize = 100 * nCount / 13 / 45;

CString str;
str.Format("SVM 识别率为： %f", recognize);
str = str + "%";
MessageBox(str);

}
}

测试功能的实现代码：

CvSVM svm;
svm.load("svm.xml");

CFileDialog dlg(TRUE, NULL, NULL, 0, "图片文件(*.jpg)|*.jpg||");
if (dlg.DoModal() == IDOK)
{
IplImage* testImage = cvLoadImage(dlg.GetPathName());
IplImage* grayImage = cvCreateImage(cvGetSize(testImage), IPL_DEPTH_8U, 1);
IplImage* hogImage = cvCreateImage(cvSize(96, 64), IPL_DEPTH_8U, 1);

cvCvtColor(testImage, grayImage, CV_RGB2GRAY);
cvResize(grayImage, hogImage, CV_INTER_LINEAR);

std::vector<float> vfDescriptors;
//vfDescriptors.resize(hogImage->width * hogImage->height);
cv::Ptr<cv::HOGDescriptor> hog = new cv::HOGDescriptor(cvSize(hogImage->width, hogImage->height), cvSize(16, 16), cvSize(8, 8), cvSize(8, 8), 9);
hog->compute(hogImage, vfDescriptors, cv::Size(1, 1), cv::Size(0, 0));		//使用计算函数进行计算

CvMat* mat = cvCreateMat(1, 256, CV_32FC1);

int cols = 0;

for (auto it = 0; it < 256; ++it)
{
cvSetReal2D(mat, 0, it, vfDescriptors[it]);
}

int  ret = svm.predict(mat);
CString str;
switch (ret)
{
case 1:
str = "品种1：P12";
break;
case 2:
str = "品种2：矮2";
break;
case 3:
str = "品种3：花玉22";
break;
case 4:
str = "品种4：花玉25";
break;
case 5:
str = "品种5：冀花2号";
break;
case 6:
str = "品种6：冀花4号";
break;
case 7:
str = "品种7：冀花5号";
break;
case 8:
str = "品种8：鲁花9号";
break;
case 9:
str = "品种9：青花6号";
break;
case 10:
str = "品种10：天府3号";
break;
case 11:
str = "品种11：维花8号";
break;
case 12:
str = "品种12：小白沙";
break;
case 13:
str = "品种13：中农108";
break;
}
AfxMessageBox(str);
//cvReleaseMat(&mat);
}//end if

相关的程序代码，用于测试的图像样本及程序中用到的操作 Excel 的类，已经打包上传到CSDN，请自行下载：http://download.csdn.net/detail/lingtianyulong/8377461