childview.cpp

基于bp人工智能的图像识别的技术

//初始的权值和偏置存放于init
void CChildView::OnPrewb() 
{   mark_wb=1;//表示已经产生了随机初始权值和偏置
	W1=MatrixRand(s1,r);  B1=MatrixRand(s1,1);
	W2=MatrixRand(s2,s1); B2=MatrixRand(s2,1);	
	int i,j;
	for(i=0;i<s1;i++) 
	{   B1_init[i]=B1[i];
		for(j=0;j<r;j++) 
			W1_init[i*r+j]=W1[i*r+j];
	}
    for(i=0;i<s2;i++) 
	{   B2_init[i]=B2[i];
		for(j=0;j<s1;j++) 
			W2_init[i*s1+j]=W2[i*s1+j];
	}
	free(W1);
	free(W2);
	free(B1);
	free(B2);
    CClientDC dc(this);
	CString str;
	CFont font;
	font.CreatePointFont (300,_TEXT("新宋体"));
	dc.SelectObject (&font);
	dc.Rectangle (rect);
	dc.DrawText (_T("正在随机产生初始权值和偏置.....OK！"),-1,&rect,DT_SINGLELINE|DT_CENTER|DT_VCENTER);
}


//显示标准的0到9样本数据（是未加噪声的标准样本数据）
void CChildView::OnDispSamples() 
{
	CClientDC dc(this);	
	dc.Rectangle (rect);
	CPen pen(PS_SOLID,3,RGB(105,90,170));
	dc.SelectObject(&pen);
	CPoint pcenter(20,20);
	int i;
	CFont font1;
	font1.CreatePointFont (250,_T("新宋体"));
	dc.SelectObject (&font1);
	dc.TextOut (30,80,"以下依次为数字0到9的标准不含噪声的样本数据：");
	for(i=0;i<10;i++)
	{   pcenter.x =30+i*90;
	    pcenter.y =150;
	    DrawNumber(P_samples[i],pcenter);
	}

}


//对已经训练完毕的BP网络应用于：加噪声的未知数字的识别
void CChildView::OnBpnRun() 
{   int targetindex,i=5,j,k,sbl=0,*np;
    double* temp1[40];	double* temp2[40];	double* temp3[40];
	double* temp4[40];	double* temp5[40];	double* temp6[40];
	CClientDC dc(this);
	dc.Rectangle (rect);
	CString str;
	CPoint pcenter(40,40);
	CFont font;
	font.CreatePointFont (120,_TEXT("新宋体"));
	dc.SelectObject (&font);
	dc.Rectangle (rect);
	dc.TextOut(129,20,"--------------应用BP网络进行数字识别（识别对象为随机加噪的40组未知数据）：-------------------");	
	for(i=0;i<40;i++)
	   for(j=0;j<N_P;j++)	
		   P_withnoise[i][j]=P_samples[i%10][j];
	for(i=0;i<10;i++)
	   for(k=0;k<=4;k++)
		{	np=Rand2(N_P,k+2);
	        for(j=0;j<k+2;j++)
	           P_withnoise[i+k*10][np[j]]=(P_withnoise[i+k*10][np[j]]==1)?0:1;
		}

	for(i=0;i<40;i++)//测试40组数据
	{   P=P_withnoise[i];
		temp2[i]=N1=MatrixAdd((temp1[i]=MatrixMul(W1_final,P,s1,r,1)),B1_final,s1,1);  //N1: s1×1
		temp3[i]=A1=f1(N1,s1);
		temp5[i]=N2=MatrixAdd((temp4[i]=MatrixMul(W2_final,A1,s2,s1,1)),B2_final,s2,1);//N2: s2×1
		temp6[i]=A2=f2(N2,s2);	
		targetindex=TargetTest(A2);    
	    pcenter.x =40+(i%(Ns/Nsext))*98;
	    pcenter.y =80+95*int(i/(Ns/Nsext));
		DrawNumber(P,pcenter);
		dc.TextOut(pcenter.x-15,pcenter.y+50,"识别结果:");
	    str.Format("%d ",targetindex);
	    dc.TextOut (pcenter.x+60,pcenter.y+50,str);
		if(i%10==targetindex) sbl++;
		//TRACE("It is number %d ! %s!",targetindex,(i==targetindex)?"Right":"Wrong");
		//if(i==targetindex) {sbl++;TRACE("\n");}
	    //else TRACE(" 应该是 %d \n",i);
	}
    dc.TextOut (45,pcenter.y+160,"当前BP网络对未知数字（含3～15%的随机噪声）的识别率达到：");	
	str.Format ("%3.2f",100*(double)sbl/40);
	dc.TextOut (490,pcenter.y+160,str);
	dc.TextOut (537,pcenter.y+160,"%");
    for(i=0;i<40;i++)
	{ free(temp1[i]);free(temp2[i]);free(temp3[i]);free(temp4[i]);free(temp5[i]);free(temp6[i]);}
	//TRACE("网络应用测试：识别率为: 百分之%2.2f\n ",Nsext*100*(double)sbl/(double)Ns);
}


//BP网络的校验：使用已知样本(包括不带噪、带噪)进行分类看是否正确
void CChildView::OnBPNVerify() 
{   int targetindex,i,sbl=0,k;
    double* temp1[Ns];	double* temp2[Ns];	double* temp3[Ns];
	double* temp4[Ns];	double* temp5[Ns];	double* temp6[Ns];
	CClientDC dc(this);
	dc.Rectangle (rect);
	CString str;
	CPoint pcenter(40,40);
	CFont font;
	font.CreatePointFont (120,_TEXT("新宋体"));
	dc.SelectObject (&font);
	dc.Rectangle (rect);
	dc.TextOut(30,20,"------------校验结果-----------------");	
	dc.TextOut(40,50,"第一行不含有噪声的数字，后五行分别加入了3～15%的随机噪声：");
	for(i=0;i<Ns;i++)
	{  k=i-10*int(i/10);
       P=P_samples[i];
	   temp2[i]=N1=MatrixAdd((temp1[i]=MatrixMul(W1_final,P,s1,r,1)),B1_final,s1,1);  //N1: s1×1
	   temp3[i]=A1=f1(N1,s1);
       temp5[i]=N2=MatrixAdd((temp4[i]=MatrixMul(W2_final,A1,s2,s1,1)),B2_final,s2,1);//N2: s2×1
	   temp6[i]=A2=f2(N2,s2);	
	   targetindex=TargetTest(A2);    
	   TRACE("It is number %d ! %s!",targetindex,(k==targetindex)?"Right":"Wrong");
	   if(k==targetindex) {sbl++;TRACE("\n");}
	   else TRACE(" 应该是 %d \n",k);
	   pcenter.x =40+(i%10)*98;
	   pcenter.y =80+90*int(i/10);
	   DrawNumber(P,pcenter);
	   dc.TextOut(pcenter.x-15,pcenter.y+50,"识别结果:");
	   str.Format("%d ",k);
	   dc.TextOut (pcenter.x+60,pcenter.y+50,str);
	}	
	TRACE("识别率为: 百分之%2.2f\n ",100*(double)sbl/(double)Ns);
    dc.TextOut (60,pcenter.y+100,"当前网络对已知数据的识别率已达到：");
	str.Format ("%3.2f",100*(double)sbl/(double)Ns);
	dc.TextOut (360,pcenter.y+100,str);
	dc.TextOut (400,pcenter.y+100," %");
	for(i=0;i<Ns;i++)
	{ free(temp1[i]);free(temp2[i]);free(temp3[i]);free(temp4[i]);free(temp5[i]);free(temp6[i]);}
}

//引入动量因子的训练过程
void CChildView::OnBPTRAINOptimal1() 
{   dongliangmark=1;
	errorfilemark++;
	if(mark_wb==0) OnPrewb();//随机产生初始权值和偏置存放于init
	int i,j,current,loops,trainok;
	int mark[Ns];//训练成功标志
	//numE=(double*)malloc(N_Err*sizeof(double));//存放最多N_Err点的均方误差
	/////////////////////////////////////////////////////////////////////
	current=loops=trainok=0;
	for(i=0;i<N_Err;i++) numE[i]=0;
	for(i=0;i<Ns;i++) mark[i]=0;
	initwb();//初始化权值和偏置（将init赋值给final）
	/////////////////////////////////////////////////////////////////////
	CClientDC dc(this);
	dc.Rectangle (rect);
	CString str;
	CFont font;
	font.CreatePointFont (350,_TEXT("新宋体"));
	dc.SelectObject (&font);
	dc.DrawText (_T("正在训练网络...请稍等...."),-1,&rect,DT_SINGLELINE|DT_CENTER|DT_VCENTER);
	double* temp[23];
	double* dw1=(double*)malloc(s1*r*sizeof(double));
	double* dw2=(double*)malloc(s2*s1*sizeof(double));
	double* db1=(double*)malloc(s1*1*sizeof(double));
	double* db2=(double*)malloc(s2*1*sizeof(double));
	for(i=0;i<s1;i++)
	{   db1[i]=0.1;
	   for(j=0;j<r;j++)
		  dw1[i*r+j]=0.1;
	}
	for(i=0;i<s2;i++)
	{   db2[i]=0.1;
	   for(j=0;j<s1;j++)
		  dw2[i*s1+j]=0.1;
	}
	CTime startTime = CTime::GetCurrentTime();
	struct _timeb timebuffer;
	int mills1,mills2;
    _ftime(&timebuffer);
	mills1=timebuffer.millitm;
step2:
	loadwb();//将final赋值给新分配内存空间的W1,W2,B1,B2
	current=loops%(Nsext*10);
	P=P_samples[current];//current为0到9，以循环指向数字0～9的标准输入样本
	T=T_output[current];
	/******************************计算正向传播信息********************************/
	N1=MatrixAdd((temp[0]=MatrixMul(W1,P,s1,r,1)),B1,s1,1);  //N1: s1×1
	A1=f1(N1,s1);//f1是自定义的函数，具备对矢量进行运算的功能
	N2=MatrixAdd((temp[1]=MatrixMul(W2,A1,s2,s1,1)),B2,s2,1);//N2: s2×1
	A2=f2(N2,s2);
	/******************************反向传播误差的计算******************************/
	E=MatrixSub(T,A2,s2,1);//T、A2、E为s2×1
	numE[loops]=0;	//注意，此时的误差不是一个数了！是个向量
	for(i=0;i<s2;i++)
        numE[loops]=numE[loops]+E[i]*E[i];
	//-------计算S2、S1（注意顺序）
	S2=MatrixMulNum((temp[3]=MatrixMul((temp[2]=F2(N2,s2)),E,s2,s2,1)),s2,1,-2.0);
	S1=MatrixMul(temp[6]=(MatrixMul((temp[4]=F1(N1,s1)),(temp[5]=MatrixInver(W2,s2,s1)),s1,s1,s2)),S2,s1,s2,1);
	//调整输出层神经元的权值和偏置
	dW2=MatrixMul(S2,(temp[7]=MatrixInver(A1,s1,1)),s2,1,s1);//dW2: s2×s1
	dW2=MatrixMulNum((temp[8]=dW2),s2,s1,(rmobp-1)*learnspeed);
	temp[15]=MatrixMulNum(dw2,s2,s1,rmobp);
	temp[16]=MatrixAdd(temp[15],dW2,s2,s1);
	W2=MatrixAdd((temp[9]=W2),temp[16],s2,s1);
	dB2=MatrixMulNum(S2,s2,1,(rmobp-1)*learnspeed);
	temp[17]=MatrixMulNum(db2,s2,1,rmobp);
	temp[18]=MatrixAdd(temp[17],dB2,s2,1);
	B2=MatrixAdd((temp[10]=B2),temp[18],s2,1);
	//调整隐层神经元权值和偏置
	dW1=MatrixMul(S1,(temp[11]=MatrixInver(P,r,1)),s1,1,r);//dW1: s1×r
    dW1=MatrixMulNum((temp[12]=dW1),s1,r,(rmobp-1)*learnspeed);
	temp[19]=MatrixMulNum(dw1,s1,r,rmobp);
	temp[20]=MatrixAdd(temp[19],dW1,s1,r);
	W1=MatrixAdd((temp[13]=W1),temp[20],s1,r);
	dB1=MatrixMulNum(S1,s1,1,(rmobp-1)*learnspeed);
	temp[21]=MatrixMulNum(db1,s1,1,rmobp);
	temp[22]=MatrixAdd(temp[21],dB1,s1,1);
	B1=MatrixAdd((temp[14]=B1),temp[22],s1,1);
    dc.DrawText (_T("正在训练网络...请稍等...."),-1,&rect,DT_SINGLELINE|DT_CENTER|DT_VCENTER);
	TRACE("%d loops, error is: %f  \n",loops,numE[loops]);
	savewb();//将临时空间W、B中的数据存于final
	for(i=0;i<s1;i++)
	{   db1[i]=dB1[i];
	   for(j=0;j<r;j++)
		  dw1[i*r+j]=dW1[i*r+j];
	}
	for(i=0;i<s2;i++)
	{   db2[i]=dB2[i];
	   for(j=0;j<s1;j++)
		  dw2[i*s1+j]=dW2[i*s1+j];
	}
	//释放动态分配的内存区
	free(W1);free(B1);free(N1);free(A1);free(dB1);free(S1);free(dW1);
	free(W2);free(B2);free(N2);free(A2);free(dB2);free(S2);free(dW2);
	for(i=0;i<23;i++) free(temp[i]);
	if(loops>N_Err) goto step3;
	if(numE[loops]>error_accept)  //说明误差不能满足要求。继续调节
	{   mark[current]=0; //验证有效位重新计数
		++loops;
	    goto step2;
	}
	else mark[current]=1;//执行到此步说明在该次迭代中，误差符合要求，记功一次
	trainok=0;
	for(i=0;i<Ns;i++)
		trainok+=mark[i];
	if(trainok>=Ns)
	{   //训练可以结束了;
	    goto step3;
	}
	else {  ++loops; goto step2;}
step3:
	learncount=loops-1;//计算总的学习次数并存放于全局变量
    dc.Rectangle (rect);
	dc.TextOut (300,150,_T("训练完成！"));
	str.Format ("%d",learncount);
	dc.TextOut (80,400,"总共训练了：");
	dc.TextOut (360,400,str);
	dc.TextOut (490,400,"次");
	str.Format ("%f",numE[loops-1]);
	dc.TextOut (80,500,"最终误差：");
	dc.TextOut (340,500,str);
	stre=str;
	//计算训练所用时间
	CTime endTime = CTime::GetCurrentTime();
	CTimeSpan elapsedTime = endTime - startTime;
    _ftime(&timebuffer);
	mills2=timebuffer.millitm;
	long int timeused;
	timeused=elapsedTime.GetTotalSeconds();
	if (mills1<=mills2)
		timeused=timeused*1000+(mills2-mills1);//将计算训练所用时间换算成毫秒
	else
		timeused=(timeused-1)*1000+(mills1-mills2);
	strt.Format ("加入动量因子后，训练网络所用时间为：%d 毫秒    (初始权值不变的情况下)",timeused);
    ::AfxMessageBox(strt); 
}

void CChildView::OnSaveerror() 
{
	//写输出文件：训练过程中的误差
	int i;
	//errorfilemark++;
	CString filename;
	CString str;
	filename.Format ("训练误差记录%d.txt",errorfilemark);
	CFile file (filename, CFile::modeReadWrite|CFile::modeCreate);
	CString str_file;
	char buffer[2]={13,10};
	if(dongliangmark==0) str_file.Format ("训练过程误差记录（没有加入动量因子）：");
	else str_file.Format ("训练过程误差记录（使用了动量因子）：");
	file.Write (str_file, str_file.GetLength());
    file.Write (buffer,sizeof(buffer));
	for(i=0;i<learncount;i++)
	{
		str_file.Format ("loos%5d,误差为: %4.4f",i+1,numE[i]);
		//str_file.GetLength();
		file.Write (str_file, str_file.GetLength());
		file.Write (buffer,sizeof(buffer));
	}
	str_file.Format ("训练完毕!");
	file.Write (str_file,str_file.GetLength());	
	file.Write (buffer,sizeof(buffer));
	file.Write (strt,strt.GetLength ());
	file.Write (buffer,sizeof(buffer));
	str_file.Format ("最终误差： ");
	file.Write (str_file,str_file.GetLength ());
	file.Write (stre,stre.GetLength ());
	file.Close ();
	str.Format ("训练过程误差信息输出到: 训练误差记录%d.txt 中",errorfilemark);
	::AfxMessageBox(str);
}

void CChildView::OnSaveweight() 
{
	int i,j;
	//errorfilemark++;
	CString filename;
	CString str;
	filename.Format ("网络最终权值%d.txt",errorfilemark);
	CFile file (filename, CFile::modeReadWrite|CFile::modeCreate);
	CString str_file;
	char buffer[2]={13,10};
	str_file.Format ("网络训练最终结果－隐层和输出层的权值与偏置：");
	file.Write (str_file, str_file.GetLength());
    file.Write (buffer,sizeof(buffer));
	str_file.Format ("*************************隐层的：*****************************");
	file.Write (str_file, str_file.GetLength());
    file.Write (buffer,sizeof(buffer));
	for(i=0;i<s1;i++)
	   for(j=0;j<r;j++)
	   {   str_file.Format ("W1(%2d,%2d): %3.4f",i,j,W1_final[i*r+j]);
		   file.Write (str_file, str_file.GetLength());
		   file.Write (buffer,sizeof(buffer));
	   }
    for(i=0;i<s1;i++)
	   {   str_file.Format ("B1(%d): %3.4f",i,B1_final[i]);
		   file.Write (str_file, str_file.GetLength());
		   file.Write (buffer,sizeof(buffer));
	   }
    file.Write (buffer,sizeof(buffer));    file.Write (buffer,sizeof(buffer));
	str_file.Format ("*************************输出层的：*****************************");
	file.Write (str_file, str_file.GetLength());
    file.Write (buffer,sizeof(buffer));
    for(i=0;i<s2;i++)
	   for(j=0;j<s1;j++)
	   {   str_file.Format ("W2(%2d,%2d): %3.4f",i,j,W2_final[i*s1+j]);
		   file.Write (str_file, str_file.GetLength());
		   file.Write (buffer,sizeof(buffer));
	   }
	for(i=0;i<s2;i++)
	   {   str_file.Format ("B2(%d): %3.4f",i,B2_final[i]);
		   file.Write (str_file, str_file.GetLength());
		   file.Write (buffer,sizeof(buffer));
	   }
	file.Write (buffer,sizeof(buffer));
	str_file.Format ("欲求此程序源代码，请与作者联系：[e-mail] ");
	file.Write (str_file, str_file.GetLength());
    file.Write (buffer,sizeof(buffer));
	str.Format ("训练后的网络最终权值、偏置输出到: 网络最终权值%d.txt 中 ",errorfilemark);
	::AfxMessageBox(str);	
}