// μ、∑计算
private void mvButton_Click(object sender, EventArgs e)
{
ArrayList samples = FeatureHelper.GetFeaturesList();
MVHelper.SetSampleList(samples);
ResetResultView();
}
// 降维处理
private void rdButton_Click(object sender, EventArgs e)
{
//新得到的样本特征
ArrayList mdSamples = MDAHelper.GetMDSampleList();
MVHelper.SetSampleList(mdSamples);
//再用这个新样本，算出每一类的均值和协方差矩阵
ArrayList changeSamples = MVHelper.GetClassFeatureList(mdSamples);
ResetRDResultView(mdSamples);
}
// 重置相关结果视图
private void ResetResultView()
{
classChooseComboBox.Items.Clear();
classFeatureList = MVHelper.GetClassFeatureList(FeatureHelper.GetFeaturesList());
//按照类列表中的内容，生成类选择下拉列表的项目
for (int i = 0; i < classFeatureList.Count; i++)
{
Sample sample = (Sample)classFeatureList[i];
string classIDStr = "000" + sample.ClassID;
classChooseComboBox.Items.Add(classIDStr.Substring(classIDStr.Length - 3));
}
//初始选择
classChooseComboBox.SelectedIndex = 0;
//计算初始选择的结果
ShowResultView(Int32.Parse((string)this.classChooseComboBox.Items[0]));
}
// 分类运算
private void classifyButton_Click(object sender, EventArgs e)
{
//前提检验
if (SelectedPCXHelper.GetSelPCXFromLB().Count == 0 || SelectedPCXHelper.GetUnselPCXList().Count == 0)
{
MessageBox.Show(this, "您还未提取样本特征，或者还未设置测试样本集！", "提示信息", MessageBoxButtons.OK);
}
else
{
string correctRate = null; //正确率
string myfilepath = filepathText.Text.ToString();
Features feature;
#region 这里做一个文件名为unknown.pcx的判断
string filename = FeatureHelper.GetUnknownName(myfilepath);
if (filename.ToLower().Equals("unknown"))
{
feature = new Features(myfilepath);
}
else
{
int classID = Convert.ToInt32(FeatureHelper.GetUpperFoldername(myfilepath));
feature = new Features(myfilepath, classID);
}
#endregion
#region Bayes分类法
if (rbBayes.Checked)
{
#region 数据初始化
CheckInit();
double correctCount = 0.0;
#endregion
IList sampleList = FeatureHelper.GetFeaturesList(); //获取原始训练样本
//从降维器获取降维后的新样本
IList newSampleList = MDAHelper.GetMDSampleList();
MVHelper.SetSampleList((ArrayList)newSampleList);
Bayes bayes = Bayes.GetInstance();
bayes.TrainSampleList = newSampleList; //向贝叶斯分类器注入降维后的训练样本
//int classID = Convert.ToInt32(FeatureHelper.GetUpperFoldername(myfilepath));
//Features feature = new Features(myfilepath, classID);
feature = MDAHelper.MDSample(feature); //测试样本降维
int testClassID = bayes.DecisionFunction(feature); //用贝叶斯决策进行测试样本分类
//结果显示
lblunknownclassify.Text = testClassID.ToString("000");
if (feature.classID == testClassID)
{
lblerrorinfo.Text = "Bayes分类法分类正确";
lblerrorinfo.ForeColor = Color.Green;
}
//unknown.pcx处理
else if (feature.classID == -1)
{
}
else
{
lblerrorinfo.Text = "Bayes分类法分类失败";
lblerrorinfo.ForeColor = Color.Green;
}
}
#endregion
#region Kn近邻法
if (rbKn.Checked)
{
#region 相关数据初始化
CheckInit();
int testResult = -1;
double correctCount = 0.0;
int kvalue = Constant.kvalue;
#endregion
#region 效的情况下进行计算
if (KCheck(kvalue))
{
KnNear my_knearest = new KnNear();
//int classID = Convert.ToInt32(FeatureHelper.GetUpperFoldername(myfilepath));
//Features currfeature = new Features(myfilepath, classID);
testResult = my_knearest.DoK_nearest(feature, FeatureHelper.GetFeaturesList(), kvalue);
//testResult为K近邻的分类结果
// 其实testResult的结果直接就是result求的值
string result = testResult.ToString("000");
lblunknownclassify.Text = result;
result = ResultConvert(result);
int testID = Convert.ToInt32(result);
if (testID > 0 && testID == feature.classID)
{
//correctRate = "分类正确率： " + Constant.kn_Rate;
lblerrorinfo.Text = "Kn近邻法分类正确";
lblerrorinfo.ForeColor = Color.Green;
}
//unknown.pcx处理
else if (feature.classID == -1)
{
}
else
{
lblerrorinfo.Text = "Kn近邻法分类失败！";
lblerrorinfo.ForeColor = Color.Green;
}
}
#endregion
}
#endregion
#region 最近邻法
if (rbnearest.Checked)
{
#region 初始化
CheckInit();
int testResult = -1;
double correctCount = 0.0;
#endregion
#region 最近邻分类
if (NearestCheck())
{
Nearest nearest = new Nearest();
//int classID = Convert.ToInt32(FeatureHelper.GetUpperFoldername(myfilepath));
//Features currfeature = new Features(myfilepath, classID);
testResult = nearest.Do_Nearest(feature, FeatureHelper.GetFeaturesList());
string result = testResult.ToString("000");
lblunknownclassify.Text = result;
if (testResult > 0 && testResult == feature.classID)
{
lblerrorinfo.Text = "最近邻法分类正确";
lblerrorinfo.ForeColor = Color.Green;
}
//unknown.pcx处理
else if (feature.classID == -1)
{
}
else
{
lblerrorinfo.Text = "最近邻法分类失败！";
lblerrorinfo.ForeColor = Color.Green;
}
}
#endregion
}
#endregion
}
}
private void testButton_Click(object sender, EventArgs e)
{
#region 数据初始化
double correctCount = 0.0;
double correctRate = 0.0;
IList sampleList = FeatureHelper.GetFeaturesList(); //获取原始训练样本
//从降维器获取降维后的新样本
IList newSampleList = MDAHelper.GetMDSampleList();
MVHelper.SetSampleList((ArrayList)newSampleList);
Bayes bayes = Bayes.GetInstance();
bayes.TrainSampleList = newSampleList; //向贝叶斯分类器注入降维后的训练样本
IList testSampleList = FeatureHelper.GetTestFeaturesList(); //获取测试样本
#endregion
#region DataGridView操作
bayesDataGridView.DataSource = null;
bayesDataGridView.Rows.Clear();
bayesDataGridView.Refresh();
DataSet ds = new DataSet();
DataTable dt = new DataTable();
// 或者直接将arr作为参数传入
//FeatureHelper.GetSamplesFeatures(); //初始化训练和测试样本
dt.Columns.Add("文件夹", typeof(string));
dt.Columns.Add("所属类别", typeof(string));
dt.Columns.Add("测试类别", typeof(string));
dt.Columns.Add("正误判断", typeof(string));
for (int i = 0; i < testSampleList.Count; i++)
{
DataRow row = dt.NewRow();
string rightOrWrong = "×";
Features feature = (Features)testSampleList[i];
row[0] = feature.Filepath;
row[1] = feature.classID;
feature = MDAHelper.MDSample(feature); //测试样本降维
int testClassID = bayes.DecisionFunction(feature); //用贝叶斯决策进行测试样本分类
// 用StringBuilder加快字符串处理速度。【值类型和堆类型】
StringBuilder sb = new StringBuilder();
sb.Append("类");
sb.Append(testClassID.ToString());
sb.ToString();
row[2] = sb;
if (feature.classID == testClassID)
{
correctCount++;
row[3] = " ";
}
else
{
row[3] = rightOrWrong;
//this.bayesDataGridView.DefaultCellStyle.ForeColor = Color.Red;
}
dt.Rows.Add(row);
}
ds.Tables.Add(dt);
this.bayesDataGridView.DataSource = ds.Tables[0];
#endregion
#region Bayes分类性能显示
correctRate = (correctCount / Convert.ToDouble(testSampleList.Count)) * 100.0;
Constant.bayes_Rate = correctRate.ToString("0.000") + "%";
dataShowLabel.Text = "测试样本总数 " + testSampleList.Count + " ，Bayes判断正确 " + correctCount + " 个，正确率为：" + Constant.bayes_Rate;
#endregion
}
//进行最近邻分量
//输入参数testsamples为要进行测试的样本集,trainingSamples为训练样本集
//k_value是设置的K参数
public int Do_Nearest(Features myTestSample, ArrayList trainingSamples)
{
#region 数据Data
int i, j = 0;
int index = 0; //index用于最近邻
int result = 0; // class分类结果
Features myTrainingSample;
#endregion
#region trainingCount 初始化
ArrayList trainlist = MVHelper.GetClassFeatureList(trainingSamples);
int trainingCount = 0;
if (trainlist.Count == 1) //只选择一类的话
{
trainingCount = 1; //((Sample)trainlist[0]).ClassSampleList.Count;
}
else
{
trainingCount = trainlist.Count;
}
#endregion
if (trainingCount == 0 || myTestSample == null)
{
return(-1);
}
#region 最近邻分类算法
//建立用来记录当前测试样本到每个训练样本的距离以及对应的训练样本类别
//sampleDistance[] myDistance = new sampleDistance[trainingCount];
sampleDistance[] myDimetion = new sampleDistance[Constant.classnumber];
//依次计算当前tmpsample样本与训练样本集中Wi类Ni个样本的欧式距离
for (i = 0; i < trainingCount; i++)
{
Sample sample = (Sample)trainlist[i];
int n_SampleNum = sample.ClassSampleList.Count;
// 这里myDistance的初始化应该与n_SampleNum具体值进行动态改变
sampleDistance[] myDistance = new sampleDistance[n_SampleNum];
for (int k = 0; k < n_SampleNum; k++)
{
myTrainingSample = (Features)sample.ClassSampleList[k];
myDistance[k].classID = myTrainingSample.classID;
myDistance[k].distance = MeasureDistance(myTestSample.feature_vector, myTrainingSample.feature_vector);
}
//当前tmpsample样本与训练样本集中Wi类Ni个样本的欧式距离
Array.Sort(myDistance, CompareByDistance);
//myDimetion[i] = myDistance[0]; //不能这么直接赋值
myDimetion[i].classID = myDistance[0].classID;
myDimetion[i].distance = myDistance[0].distance;
//index += n_SampleNum;
}
#endregion
#region myDimetion排序时机问题选择
//上面已经得到了当前tmpsample样本与训练样本集中c个类之间的距离,下面对其排序（升序）
// 考虑到如果myDimetion长度只为1的话，就会报错
if (trainlist.Count == 1)
{
result = myDimetion[0].classID;
return(result);
}
else
{
Array.Sort(myDimetion, CompareByDistance);
result = myDimetion[0].classID;
return(result);
}
#endregion
}
