卷积神经网络(CNN)的简单实现(MNIST)

卷积神经网络(CNN)的基础介绍见 http://xilinx.eetrend.com/article/11731 ,这里主要以代码实现为主。

CNN是一个多层的神经网络,每层由多个二维平面组成,而每个平面由多个独立神经元组成。

以MNIST作为数据库,仿照LeNet-5和tiny-cnn( http://blog.csdn.net/fengbingchun/article/details/50573841 ) 设计一个简单的7层CNN结构如下:
输入层Input:神经元数量32*32=1024;

C1层:卷积窗大小5*5,输出特征图数量6,卷积窗种类6,输出特征图大小28*28,可训练参数(权值+阈值(偏置))5*5*6+6=150+6,神经元数量28*28*6=4704;

S2层:卷积窗大小2*2,输出下采样图数量6,卷积窗种类6,输出下采样图大小14*14,可训练参数1*6+6=6+6,神经元数量14*14*6=1176;

C3层:卷积窗大小5*5,输出特征图数量16,卷积窗种类6*16=96,输出特征图大小10*10,可训练参数5*5*(6*16)+16=2400+16,神经元数量10*10*16=1600;

S4层:卷积窗大小2*2,输出下采样图数量16,卷积窗种类16,输出下采样图大小5*5,可训练参数1*16+16=16+16,神经元数量5*5*16=400;

C5层:卷积窗大小5*5,输出特征图数量120,卷积窗种类16*120=1920,输出特征图大小1*1,可训练参数5*5*(16*120)+120=48000+120,神经元数量1*1*120=120;

输出层Output:卷积窗大小1*1,输出特征图数量10,卷积窗种类120*10=1200,输出特征图大小1*1,可训练参数1*(120*10)+10=1200+10,神经元数量1*1*10=10。

下面对实现执行过程进行描述说明:

1. 从MNIST数据库中分别获取训练样本和测试样本数据:

(1)、原有MNIST库中图像大小为28*28,这里缩放为32*32,数据值范围为[-1,1],扩充值均取-1;总共60000个32*32训练样本,10000个32*32测试样本;

(2)、输出层有10个输出节点,在训练阶段,对应位置的节点值设为0.8,其它节点设为-0.8.

2. 初始化权值和阈值(偏置):权值就是卷积图像,每一个特征图上的神经元共享相同的权值和阈值,特征图的数量等于阈值的个数

(1)、权值采用uniform rand的方法初始化;

(2)、阈值均初始化为0.

3. 前向传播:根据权值和阈值,主要计算每层神经元的值

(1)、输入层:每次输入一个32*32数据。

(2)、C1层:分别用每一个5*5的卷积图像去乘以32*32的图像,获得一个28*28的图像,即对应位置相加再求和,stride长度为1;一共6个5*5的卷积图像,然后对每一个神经元加上一个阈值,最后再通过tanh激活函数对每一神经元进行运算得到最终每一个神经元的结果。

(3)、S2层:对C1中6个28*28的特征图生成6个14*14的下采样图,相邻四个神经元分别进行相加求和,然后乘以一个权值,再求均值即除以4,然后再加上一个阈值,最后再通过tanh激活函数对每一神经元进行运算得到最终每一个神经元的结果。

(4)、C3层:由S2中的6个14*14下采样图生成16个10*10特征图,对于生成的每一个10*10的特征图,是由6个5*5的卷积图像去乘以6个14*14的下采样图,然后对应位置相加求和,然后对每一个神经元加上一个阈值,最后再通过tanh激活函数对每一神经元进行运算得到最终每一个神经元的结果。

(5)、S4层:由C3中16个10*10的特征图生成16个5*5下采样图,相邻四个神经元分别进行相加求和,然后乘以一个权值,再求均值即除以4,然后再加上一个阈值,最后再通过tanh激活函数对每一神经元进行运算得到最终每一个神经元的结果。

(6)、C5层:由S4中16个5*5下采样图生成120个1*1特征图,对于生成的每一个1*1的特征图,是由16个5*5的卷积图像去乘以16个5*5的下采用图,然后相加求和,然后对每一个神经元加上一个阈值,最后再通过tanh激活函数对每一神经元进行运算得到最终每一个神经元的结果。

(7)、输出层:即全连接层,输出层中的每一个神经元均是由C5层中的120个神经元乘以相对应的权值,然后相加求和;然后对每一个神经元加上一个阈值,最后再通过tanh激活函数对每一神经元进行运算得到最终每一个神经元的结果。

4. 反向传播:主要计算每层神经元、权值和阈值的误差,以用来更新权值和阈值

(1)、输出层:计算输出层神经元误差;通过mse损失函数的导数函数和tanh激活函数的导数函数来计算输出层神经元误差。

(2)、C5层:计算C5层神经元误差、输出层权值误差、输出层阈值误差;通过输出层神经元误差乘以输出层权值,求和,结果再乘以C5层神经元的tanh激活函数的导数,获得C5层每一个神经元误差;通过输出层神经元误差乘以C5层神经元获得输出层权值误差;输出层误差即为输出层阈值误差。

(3)、S4层:计算S4层神经元误差、C5层权值误差、C5层阈值误差;通过C5层权值乘以C5层神经元误差,求和,结果再乘以S4层神经元的tanh激活函数的导数,获得S4层每一个神经元误差;通过S4层神经元乘以C5层神经元误差,求和,获得C5层权值误差;C5层神经元误差即为C5层阈值误差。

(4)、C3层:计算C3层神经元误差、S4层权值误差、S4层阈值误差;

(5)、S2层:计算S2层神经元误差、C3层权值误差、C3层阈值误差;

(6)、C1层:计算C1层神经元误差、S2层权值误差、S2层阈值误差;

(7)、输入层:计算C1层权值误差、C1层阈值误差.

代码文件:

CNN.hpp:
#ifndef _CNN_HPP_
#define _CNN_HPP_

#include
#include

namespace ANN {

#define width_image_input_CNN 32 //归一化图像宽
#define height_image_input_CNN 32 //归一化图像高
#define width_image_C1_CNN 28
#define height_image_C1_CNN 28
#define width_image_S2_CNN 14
#define height_image_S2_CNN 14
#define width_image_C3_CNN 10
#define height_image_C3_CNN 10
#define width_image_S4_CNN 5
#define height_image_S4_CNN 5
#define width_image_C5_CNN 1
#define height_image_C5_CNN 1
#define width_image_output_CNN 1
#define height_image_output_CNN 1

#define width_kernel_conv_CNN 5 //卷积核大小
#define height_kernel_conv_CNN 5
#define width_kernel_pooling_CNN 2
#define height_kernel_pooling_CNN 2
#define size_pooling_CNN 2

#define num_map_input_CNN 1 //输入层map个数
#define num_map_C1_CNN 6 //C1层map个数
#define num_map_S2_CNN 6 //S2层map个数
#define num_map_C3_CNN 16 //C3层map个数
#define num_map_S4_CNN 16 //S4层map个数
#define num_map_C5_CNN 120 //C5层map个数
#define num_map_output_CNN 10 //输出层map个数

#define num_patterns_train_CNN 60000 //训练模式对数(总数)
#define num_patterns_test_CNN 10000 //测试模式对数(总数)
#define num_epochs_CNN 100 //最大迭代次数
#define accuracy_rate_CNN 0.985 //要求达到的准确率
#define learning_rate_CNN 0.01 //学习率
#define eps_CNN 1e-8

#define len_weight_C1_CNN 150 //C1层权值数,5*5*6*1=150
#define len_bias_C1_CNN 6 //C1层阈值数,6
#define len_weight_S2_CNN 6 //S2层权值数,1*6=6
#define len_bias_S2_CNN 6 //S2层阈值数,6
#define len_weight_C3_CNN 2400 //C3层权值数,5*5*16*6=2400
#define len_bias_C3_CNN 16 //C3层阈值数,16
#define len_weight_S4_CNN 16 //S4层权值数,1*16=16
#define len_bias_S4_CNN 16 //S4层阈值数,16
#define len_weight_C5_CNN 48000 //C5层权值数,5*5*16*120=48000
#define len_bias_C5_CNN 120 //C5层阈值数,120
#define len_weight_output_CNN 1200 //输出层权值数,120*10=1200
#define len_bias_output_CNN 10 //输出层阈值数,10

#define num_neuron_input_CNN 1024 //输入层神经元数,32*32=1024
#define num_neuron_C1_CNN 4704 //C1层神经元数,28*28*6=4704
#define num_neuron_S2_CNN 1176 //S2层神经元数,14*14*6=1176
#define num_neuron_C3_CNN 1600 //C3层神经元数,10*10*16=1600
#define num_neuron_S4_CNN 400 //S4层神经元数,5*5*16=400
#define num_neuron_C5_CNN 120 //C5层神经元数,1*120=120
#define num_neuron_output_CNN 10 //输出层神经元数,1*10=10

class CNN {
public:
CNN();
~CNN();

void init(); //初始化,分配空间
bool train(); //训练
int predict(const unsigned char* data, int width, int height); //预测
bool readModelFile(const char* name); //读取已训练好的BP model

protected:
typedef std::vector > wi_connections;
typedef std::vector > wo_connections;
typedef std::vector > io_connections;

void release(); //释放申请的空间
bool saveModelFile(const char* name); //将训练好的model保存起来,包括各层的节点数,权值和阈值
bool initWeightThreshold(); //初始化,产生[-1, 1]之间的随机小数
bool getSrcData(); //读取MNIST数据
double test(); //训练完一次计算一次准确率
double activation_function_tanh(double x); //激活函数:tanh
double activation_function_tanh_derivative(double x); //激活函数tanh的导数
double activation_function_identity(double x);
double activation_function_identity_derivative(double x);
double loss_function_mse(double y, double t); //损失函数:mean squared error
double loss_function_mse_derivative(double y, double t);
void loss_function_gradient(const double* y, const double* t, double* dst, int len);
double dot_product(const double* s1, const double* s2, int len); //点乘
bool muladd(const double* src, double c, int len, double* dst); //dst[i] += c * src[i]
void init_variable(double* val, double c, int len);
bool uniform_rand(double* src, int len, double min, double max);
double uniform_rand(double min, double max);
int get_index(int x, int y, int channel, int width, int height, int depth);
void calc_out2wi(int width_in, int height_in, int width_out, int height_out, int depth_out, std::vector& out2wi);
void calc_out2bias(int width, int height, int depth, std::vector& out2bias);
void calc_in2wo(int width_in, int height_in, int width_out, int height_out, int depth_in, int depth_out, std::vector& in2wo);
void calc_weight2io(int width_in, int height_in, int width_out, int height_out, int depth_in, int depth_out, std::vector& weight2io);
void calc_bias2out(int width_in, int height_in, int width_out, int height_out, int depth_in, int depth_out, std::vector >& bias2out);

bool Forward_C1(); //前向传播
bool Forward_S2();
bool Forward_C3();
bool Forward_S4();
bool Forward_C5();
bool Forward_output();
bool Backward_output();
bool Backward_C5(); //反向传播
bool Backward_S4();
bool Backward_C3();
bool Backward_S2();
bool Backward_C1();
bool Backward_input();
bool UpdateWeights(); //更新权值、阈值
void update_weights_bias(const double* delta, double* e_weight, double* weight, int len);

private:
double* data_input_train; //原始标准输入数据,训练,范围:[-1, 1]
double* data_output_train; //原始标准期望结果,训练,取值:-0.8/0.8
double* data_input_test; //原始标准输入数据,测试,范围:[-1, 1]
double* data_output_test; //原始标准期望结果,测试,取值:-0.8/0.8
double* data_single_image;
double* data_single_label;

double weight_C1[len_weight_C1_CNN];
double bias_C1[len_bias_C1_CNN];
double weight_S2[len_weight_S2_CNN];
double bias_S2[len_bias_S2_CNN];
double weight_C3[len_weight_C3_CNN];
double bias_C3[len_bias_C3_CNN];
double weight_S4[len_weight_S4_CNN];
double bias_S4[len_bias_S4_CNN];
double weight_C5[len_weight_C5_CNN];
double bias_C5[len_bias_C5_CNN];
double weight_output[len_weight_output_CNN];
double bias_output[len_bias_output_CNN];

double E_weight_C1[len_weight_C1_CNN];
double E_bias_C1[len_bias_C1_CNN];
double E_weight_S2[len_weight_S2_CNN];
double E_bias_S2[len_bias_S2_CNN];
double E_weight_C3[len_weight_C3_CNN];
double E_bias_C3[len_bias_C3_CNN];
double E_weight_S4[len_weight_S4_CNN];
double E_bias_S4[len_bias_S4_CNN];
double* E_weight_C5;
double* E_bias_C5;
double* E_weight_output;
double* E_bias_output;

double neuron_input[num_neuron_input_CNN]; //data_single_image
double neuron_C1[num_neuron_C1_CNN];
double neuron_S2[num_neuron_S2_CNN];
double neuron_C3[num_neuron_C3_CNN];
double neuron_S4[num_neuron_S4_CNN];
double neuron_C5[num_neuron_C5_CNN];
double neuron_output[num_neuron_output_CNN];

double delta_neuron_output[num_neuron_output_CNN]; //神经元误差
double delta_neuron_C5[num_neuron_C5_CNN];
double delta_neuron_S4[num_neuron_S4_CNN];
double delta_neuron_C3[num_neuron_C3_CNN];
double delta_neuron_S2[num_neuron_S2_CNN];
double delta_neuron_C1[num_neuron_C1_CNN];
double delta_neuron_input[num_neuron_input_CNN];

double delta_weight_C1[len_weight_C1_CNN]; //权值、阈值误差
double delta_bias_C1[len_bias_C1_CNN];
double delta_weight_S2[len_weight_S2_CNN];
double delta_bias_S2[len_bias_S2_CNN];
double delta_weight_C3[len_weight_C3_CNN];
double delta_bias_C3[len_bias_C3_CNN];
double delta_weight_S4[len_weight_S4_CNN];
double delta_bias_S4[len_bias_S4_CNN];
double delta_weight_C5[len_weight_C5_CNN];
double delta_bias_C5[len_bias_C5_CNN];
double delta_weight_output[len_weight_output_CNN];
double delta_bias_output[len_bias_output_CNN];

std::vector out2wi_S2; // out_id -> [(weight_id, in_id)]
std::vector out2bias_S2;
std::vector out2wi_S4;
std::vector out2bias_S4;
std::vector in2wo_C3; // in_id -> [(weight_id, out_id)]
std::vector weight2io_C3; // weight_id -> [(in_id, out_id)]
std::vector > bias2out_C3;
std::vector in2wo_C1;
std::vector weight2io_C1;
std::vector > bias2out_C1;
};

}

#endif //_CNN_HPP_

CNN.cpp:
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include

namespace ANN {

CNN::CNN()
{
data_input_train = NULL;
data_output_train = NULL;
data_input_test = NULL;
data_output_test = NULL;
data_single_image = NULL;
data_single_label = NULL;
E_weight_C5 = NULL;
E_bias_C5 = NULL;
E_weight_output = NULL;
E_bias_output = NULL;
}

CNN::~CNN()
{
release();
}

void CNN::release()
{
if (data_input_train) {
delete[] data_input_train;
data_input_train = NULL;
}
if (data_output_train) {
delete[] data_output_train;
data_output_train = NULL;
}
if (data_input_test) {
delete[] data_input_test;
data_input_test = NULL;
}
if (data_output_test) {
delete[] data_output_test;
data_output_test = NULL;
}

if (E_weight_C5) {
delete[] E_weight_C5;
E_weight_C5 = NULL;
}
if (E_bias_C5) {
delete[] E_bias_C5;
E_bias_C5 = NULL;
}
if (E_weight_output) {
delete[] E_weight_output;
E_weight_output = NULL;
}
if (E_bias_output) {
delete[] E_bias_output;
E_bias_output = NULL;
}
}

// connection table [Y.Lecun, 1998 Table.1]
#define O true
#define X false
static const bool tbl[6][16] = {
O, X, X, X, O, O, O, X, X, O, O, O, O, X, O, O,
O, O, X, X, X, O, O, O, X, X, O, O, O, O, X, O,
O, O, O, X, X, X, O, O, O, X, X, O, X, O, O, O,
X, O, O, O, X, X, O, O, O, O, X, X, O, X, O, O,
X, X, O, O, O, X, X, O, O, O, O, X, O, O, X, O,
X, X, X, O, O, O, X, X, O, O, O, O, X, O, O, O
};
#undef O
#undef X

void CNN::init_variable(double* val, double c, int len)
{
for (int i = 0; i < len; i++) {
val[i] = c;
}
}

void CNN::init()
{
int len1 = width_image_input_CNN * height_image_input_CNN * num_patterns_train_CNN;
data_input_train = new double[len1];
init_variable(data_input_train, -1.0, len1);

int len2 = num_map_output_CNN * num_patterns_train_CNN;
data_output_train = new double[len2];
init_variable(data_output_train, -0.8, len2);

int len3 = width_image_input_CNN * height_image_input_CNN * num_patterns_test_CNN;
data_input_test = new double[len3];
init_variable(data_input_test, -1.0, len3);

int len4 = num_map_output_CNN * num_patterns_test_CNN;
data_output_test = new double[len4];
init_variable(data_output_test, -0.8, len4);

std::fill(E_weight_C1, E_weight_C1 + len_weight_C1_CNN, 0.0);
std::fill(E_bias_C1, E_bias_C1 + len_bias_C1_CNN, 0.0);
std::fill(E_weight_S2, E_weight_S2 + len_weight_S2_CNN, 0.0);
std::fill(E_bias_S2, E_bias_S2 + len_bias_S2_CNN, 0.0);
std::fill(E_weight_C3, E_weight_C3 + len_weight_C3_CNN, 0.0);
std::fill(E_bias_C3, E_bias_C3 + len_bias_C3_CNN, 0.0);
std::fill(E_weight_S4, E_weight_S4 + len_weight_S4_CNN, 0.0);
std::fill(E_bias_S4, E_bias_S4 + len_bias_S4_CNN, 0.0);
E_weight_C5 = new double[len_weight_C5_CNN];
std::fill(E_weight_C5, E_weight_C5 + len_weight_C5_CNN, 0.0);
E_bias_C5 = new double[len_bias_C5_CNN];
std::fill(E_bias_C5, E_bias_C5 + len_bias_C5_CNN, 0.0);
E_weight_output = new double[len_weight_output_CNN];
std::fill(E_weight_output, E_weight_output + len_weight_output_CNN, 0.0);
E_bias_output = new double[len_bias_output_CNN];
std::fill(E_bias_output, E_bias_output + len_bias_output_CNN, 0.0);

initWeightThreshold();
getSrcData();
}

double CNN::uniform_rand(double min, double max)
{
static std::mt19937 gen(1);
std::uniform_real_distribution dst(min, max);
return dst(gen);
}

bool CNN::uniform_rand(double* src, int len, double min, double max)
{
for (int i = 0; i < len; i++) {
src[i] = uniform_rand(min, max);
}

return true;
}

bool CNN::initWeightThreshold()
{
srand(time(0) + rand());
const double scale = 6.0;

double min_ = -std::sqrt(scale / (25.0 + 150.0));
double max_ = std::sqrt(scale / (25.0 + 150.0));
uniform_rand(weight_C1, len_weight_C1_CNN, min_, max_);
for (int i = 0; i < len_bias_C1_CNN; i++) {
bias_C1[i] = 0.0;
}

min_ = -std::sqrt(scale / (4.0 + 1.0));
max_ = std::sqrt(scale / (4.0 + 1.0));
uniform_rand(weight_S2, len_weight_S2_CNN, min_, max_);
for (int i = 0; i < len_bias_S2_CNN; i++) {
bias_S2[i] = 0.0;
}

min_ = -std::sqrt(scale / (150.0 + 400.0));
max_ = std::sqrt(scale / (150.0 + 400.0));
uniform_rand(weight_C3, len_weight_C3_CNN, min_, max_);
for (int i = 0; i < len_bias_C3_CNN; i++) {
bias_C3[i] = 0.0;
}

min_ = -std::sqrt(scale / (4.0 + 1.0));
max_ = std::sqrt(scale / (4.0 + 1.0));
uniform_rand(weight_S4, len_weight_S4_CNN, min_, max_);
for (int i = 0; i < len_bias_S4_CNN; i++) {
bias_S4[i] = 0.0;
}

min_ = -std::sqrt(scale / (400.0 + 3000.0));
max_ = std::sqrt(scale / (400.0 + 3000.0));
uniform_rand(weight_C5, len_weight_C5_CNN, min_, max_);
for (int i = 0; i < len_bias_C5_CNN; i++) {
bias_C5[i] = 0.0;
}

min_ = -std::sqrt(scale / (120.0 + 10.0));
max_ = std::sqrt(scale / (120.0 + 10.0));
uniform_rand(weight_output, len_weight_output_CNN, min_, max_);
for (int i = 0; i < len_bias_output_CNN; i++) {
bias_output[i] = 0.0;
}

return true;
}

static int reverseInt(int i)
{
unsigned char ch1, ch2, ch3, ch4;
ch1 = i & 255;
ch2 = (i >> 8) & 255;
ch3 = (i >> 16) & 255;
ch4 = (i >> 24) & 255;
return((int)ch1 << 24) + ((int)ch2 << 16) + ((int)ch3 << 8) + ch4;
}

static void readMnistImages(std::string filename, double* data_dst, int num_image)
{
const int width_src_image = 28;
const int height_src_image = 28;
const int x_padding = 2;
const int y_padding = 2;
const double scale_min = -1;
const double scale_max = 1;

std::ifstream file(filename, std::ios::binary);
assert(file.is_open());

int magic_number = 0;
int number_of_images = 0;
int n_rows = 0;
int n_cols = 0;
file.read((char*)&magic_number, sizeof(magic_number));
magic_number = reverseInt(magic_number);
file.read((char*)&number_of_images, sizeof(number_of_images));
number_of_images = reverseInt(number_of_images);
assert(number_of_images == num_image);
file.read((char*)&n_rows, sizeof(n_rows));
n_rows = reverseInt(n_rows);
file.read((char*)&n_cols, sizeof(n_cols));
n_cols = reverseInt(n_cols);
assert(n_rows == height_src_image && n_cols == width_src_image);

int size_single_image = width_image_input_CNN * height_image_input_CNN;

for (int i = 0; i < number_of_images; ++i) {
int addr = size_single_image * i;

for (int r = 0; r < n_rows; ++r) {
for (int c = 0; c < n_cols; ++c) {
unsigned char temp = 0;
file.read((char*)&temp, sizeof(temp));
data_dst[addr + width_image_input_CNN * (r + y_padding) + c + x_padding] = (temp / 255.0) * (scale_max - scale_min) + scale_min;
}
}
}
}

static void readMnistLabels(std::string filename, double* data_dst, int num_image)
{
const double scale_max = 0.8;

std::ifstream file(filename, std::ios::binary);
assert(file.is_open());

int magic_number = 0;
int number_of_images = 0;
file.read((char*)&magic_number, sizeof(magic_number));
magic_number = reverseInt(magic_number);
file.read((char*)&number_of_images, sizeof(number_of_images));
number_of_images = reverseInt(number_of_images);
assert(number_of_images == num_image);

for (int i = 0; i < number_of_images; ++i) {
unsigned char temp = 0;
file.read((char*)&temp, sizeof(temp));
data_dst[i * num_map_output_CNN + temp] = scale_max;
}
}

bool CNN::getSrcData()
{
assert(data_input_train && data_output_train && data_input_test && data_output_test);

std::string filename_train_images = "E:/GitCode/NN_Test/data/train-images.idx3-ubyte";
std::string filename_train_labels = "E:/GitCode/NN_Test/data/train-labels.idx1-ubyte";
readMnistImages(filename_train_images, data_input_train, num_patterns_train_CNN);
readMnistLabels(filename_train_labels, data_output_train, num_patterns_train_CNN);

std::string filename_test_images = "E:/GitCode/NN_Test/data/t10k-images.idx3-ubyte";
std::string filename_test_labels = "E:/GitCode/NN_Test/data/t10k-labels.idx1-ubyte";
readMnistImages(filename_test_images, data_input_test, num_patterns_test_CNN);
readMnistLabels(filename_test_labels, data_output_test, num_patterns_test_CNN);

return true;
}

bool CNN::train()
{
out2wi_S2.clear();
out2bias_S2.clear();
out2wi_S4.clear();
out2bias_S4.clear();
in2wo_C3.clear();
weight2io_C3.clear();
bias2out_C3.clear();
in2wo_C1.clear();
weight2io_C1.clear();
bias2out_C1.clear();

calc_out2wi(width_image_C1_CNN, height_image_C1_CNN, width_image_S2_CNN, height_image_S2_CNN, num_map_S2_CNN, out2wi_S2);
calc_out2bias(width_image_S2_CNN, height_image_S2_CNN, num_map_S2_CNN, out2bias_S2);
calc_out2wi(width_image_C3_CNN, height_image_C3_CNN, width_image_S4_CNN, height_image_S4_CNN, num_map_S4_CNN, out2wi_S4);
calc_out2bias(width_image_S4_CNN, height_image_S4_CNN, num_map_S4_CNN, out2bias_S4);
calc_in2wo(width_image_C3_CNN, height_image_C3_CNN, width_image_S4_CNN, height_image_S4_CNN, num_map_C3_CNN, num_map_S4_CNN, in2wo_C3);
calc_weight2io(width_image_C3_CNN, height_image_C3_CNN, width_image_S4_CNN, height_image_S4_CNN, num_map_C3_CNN, num_map_S4_CNN, weight2io_C3);
calc_bias2out(width_image_C3_CNN, height_image_C3_CNN, width_image_S4_CNN, height_image_S4_CNN, num_map_C3_CNN, num_map_S4_CNN, bias2out_C3);
calc_in2wo(width_image_C1_CNN, height_image_C1_CNN, width_image_S2_CNN, height_image_S2_CNN, num_map_C1_CNN, num_map_C3_CNN, in2wo_C1);
calc_weight2io(width_image_C1_CNN, height_image_C1_CNN, width_image_S2_CNN, height_image_S2_CNN, num_map_C1_CNN, num_map_C3_CNN, weight2io_C1);
calc_bias2out(width_image_C1_CNN, height_image_C1_CNN, width_image_S2_CNN, height_image_S2_CNN, num_map_C1_CNN, num_map_C3_CNN, bias2out_C1);

int iter = 0;
for (iter = 0; iter < num_epochs_CNN; iter++) {
std::cout << "epoch: " << iter + 1;

for (int i = 0; i < num_patterns_train_CNN; i++) {
data_single_image = data_input_train + i * num_neuron_input_CNN;
data_single_label = data_output_train + i * num_neuron_output_CNN;

Forward_C1();
Forward_S2();
Forward_C3();
Forward_S4();
Forward_C5();
Forward_output();

Backward_output();
Backward_C5();
Backward_S4();
Backward_C3();
Backward_S2();
Backward_C1();
Backward_input();

UpdateWeights();
}

double accuracyRate = test();
std::cout << ", accuray rate: " << accuracyRate << std::endl;
if (accuracyRate > accuracy_rate_CNN) {
saveModelFile("E:/GitCode/NN_Test/data/cnn.model");
std::cout << "generate cnn model" << std::endl;
break;
}
}

if (iter == num_epochs_CNN) {
saveModelFile("E:/GitCode/NN_Test/data/cnn.model");
std::cout << "generate cnn model" << std::endl;
}

return true;
}

double CNN::activation_function_tanh(double x)
{
double ep = std::exp(x);
double em = std::exp(-x);

return (ep - em) / (ep + em);
}

double CNN::activation_function_tanh_derivative(double x)
{
return (1.0 - x * x);
}

double CNN::activation_function_identity(double x)
{
return x;
}

double CNN::activation_function_identity_derivative(double x)
{
return 1;
}

double CNN::loss_function_mse(double y, double t)
{
return (y - t) * (y - t) / 2;
}

double CNN::loss_function_mse_derivative(double y, double t)
{
return (y - t);
}

void CNN::loss_function_gradient(const double* y, const double* t, double* dst, int len)
{
for (int i = 0; i < len; i++) {
dst[i] = loss_function_mse_derivative(y[i], t[i]);
}
}

double CNN::dot_product(const double* s1, const double* s2, int len)
{
double result = 0.0;

for (int i = 0; i < len; i++) {
result += s1[i] * s2[i];
}

return result;
}

bool CNN::muladd(const double* src, double c, int len, double* dst)
{
for (int i = 0; i < len; i++) {
dst[i] += (src[i] * c);
}

return true;
}

int CNN::get_index(int x, int y, int channel, int width, int height, int depth)
{
assert(x >= 0 && x < width);
assert(y >= 0 && y < height);
assert(channel >= 0 && channel < depth);
return (height * channel + y) * width + x;
}

void CNN::calc_out2wi(int width_in, int height_in, int width_out, int height_out, int depth_out, std::vector& out2wi)
{
for (int i = 0; i < depth_out; i++) {
int block = width_in * height_in * i;

for (int y = 0; y < height_out; y++) {
for (int x = 0; x < width_out; x++) {
int rows = y * width_kernel_pooling_CNN;
int cols = x * height_kernel_pooling_CNN;

wi_connections wi_connections_;
std::pair pair_;

for (int m = 0; m < width_kernel_pooling_CNN; m++) {
for (int n = 0; n < height_kernel_pooling_CNN; n++) {
pair_.first = i;
pair_.second = (rows + m) * width_in + cols + n + block;
wi_connections_.push_back(pair_);
}
}
out2wi.push_back(wi_connections_);
}
}
}
}

void CNN::calc_out2bias(int width, int height, int depth, std::vector& out2bias)
{
for (int i = 0; i < depth; i++) {
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++) {
out2bias.push_back(i);
}
}
}
}

void CNN::calc_in2wo(int width_in, int height_in, int width_out, int height_out, int depth_in, int depth_out, std::vector& in2wo)
{
int len = width_in * height_in * depth_in;
in2wo.resize(len);

for (int c = 0; c < depth_in; c++) {
for (int y = 0; y < height_in; y += height_kernel_pooling_CNN) {
for (int x = 0; x < width_in; x += width_kernel_pooling_CNN) {
int dymax = min(size_pooling_CNN, height_in - y);
int dxmax = min(size_pooling_CNN, width_in - x);
int dstx = x / width_kernel_pooling_CNN;
int dsty = y / height_kernel_pooling_CNN;

for (int dy = 0; dy < dymax; dy++) {
for (int dx = 0; dx < dxmax; dx++) {
int index_in = get_index(x + dx, y + dy, c, width_in, height_in, depth_in);
int index_out = get_index(dstx, dsty, c, width_out, height_out, depth_out);

wo_connections wo_connections_;
std::pair pair_;
pair_.first = c;
pair_.second = index_out;
wo_connections_.push_back(pair_);

in2wo[index_in] = wo_connections_;
}
}
}
}
}
}

void CNN::calc_weight2io(int width_in, int height_in, int width_out, int height_out, int depth_in, int depth_out, std::vector& weight2io)
{
int len = depth_in;
weight2io.resize(len);

for (int c = 0; c < depth_in; c++) {
for (int y = 0; y < height_in; y += height_kernel_pooling_CNN) {
for (int x = 0; x < width_in; x += width_kernel_pooling_CNN) {
int dymax = min(size_pooling_CNN, height_in - y);
int dxmax = min(size_pooling_CNN, width_in - x);
int dstx = x / width_kernel_pooling_CNN;
int dsty = y / height_kernel_pooling_CNN;

for (int dy = 0; dy < dymax; dy++) {
for (int dx = 0; dx < dxmax; dx++) {
int index_in = get_index(x + dx, y + dy, c, width_in, height_in, depth_in);
int index_out = get_index(dstx, dsty, c, width_out, height_out, depth_out);

std::pair pair_;
pair_.first = index_in;
pair_.second = index_out;

weight2io[c].push_back(pair_);
}
}
}
}
}
}

void CNN::calc_bias2out(int width_in, int height_in, int width_out, int height_out, int depth_in, int depth_out, std::vector >& bias2out)
{
int len = depth_in;
bias2out.resize(len);

for (int c = 0; c < depth_in; c++) {
for (int y = 0; y < height_out; y++) {
for (int x = 0; x < width_out; x++) {
int index_out = get_index(x, y, c, width_out, height_out, depth_out);
bias2out[c].push_back(index_out);
}
}
}
}

bool CNN::Forward_C1()
{
init_variable(neuron_C1, 0.0, num_neuron_C1_CNN);

for (int o = 0; o < num_map_C1_CNN; o++) {
for (int inc = 0; inc < num_map_input_CNN; inc++) {
int addr1 = get_index(0, 0, num_map_input_CNN * o + inc, width_kernel_conv_CNN, height_kernel_conv_CNN, num_map_C1_CNN * num_map_input_CNN);
int addr2 = get_index(0, 0, inc, width_image_input_CNN, height_image_input_CNN, num_map_input_CNN);
int addr3 = get_index(0, 0, o, width_image_C1_CNN, height_image_C1_CNN, num_map_C1_CNN);

const double* pw = &weight_C1[0] + addr1;
const double* pi = data_single_image + addr2;
double* pa = &neuron_C1[0] + addr3;

for (int y = 0; y < height_image_C1_CNN; y++) {
for (int x = 0; x < width_image_C1_CNN; x++) {
const double* ppw = pw;
const double* ppi = pi + y * width_image_input_CNN + x;
double sum = 0.0;

for (int wy = 0; wy < height_kernel_conv_CNN; wy++) {
for (int wx = 0; wx < width_kernel_conv_CNN; wx++) {
sum += *ppw++ * ppi[wy * width_image_input_CNN + wx];
}
}

pa[y * width_image_C1_CNN + x] += sum;
}
}
}

int addr3 = get_index(0, 0, o, width_image_C1_CNN, height_image_C1_CNN, num_map_C1_CNN);
double* pa = &neuron_C1[0] + addr3;
double b = bias_C1[o];
for (int y = 0; y < height_image_C1_CNN; y++) {
for (int x = 0; x < width_image_C1_CNN; x++) {
pa[y * width_image_C1_CNN + x] += b;
}
}
}

for (int i = 0; i < num_neuron_C1_CNN; i++) {
neuron_C1[i] = activation_function_tanh(neuron_C1[i]);
}

return true;
}

bool CNN::Forward_S2()
{
init_variable(neuron_S2, 0.0, num_neuron_S2_CNN);
double scale_factor = 1.0 / (width_kernel_pooling_CNN * height_kernel_pooling_CNN);

assert(out2wi_S2.size() == num_neuron_S2_CNN);
assert(out2bias_S2.size() == num_neuron_S2_CNN);

for (int i = 0; i < num_neuron_S2_CNN; i++) {
const wi_connections& connections = out2wi_S2[i];
neuron_S2[i] = 0;

for (int index = 0; index < connections.size(); index++) {
neuron_S2[i] += weight_S2[connections[index].first] * neuron_C1[connections[index].second];
}

neuron_S2[i] *= scale_factor;
neuron_S2[i] += bias_S2[out2bias_S2[i]];
}

for (int i = 0; i < num_neuron_S2_CNN; i++) {
neuron_S2[i] = activation_function_tanh(neuron_S2[i]);
}

return true;
}

bool CNN::Forward_C3()
{
init_variable(neuron_C3, 0.0, num_neuron_C3_CNN);

for (int o = 0; o < num_map_C3_CNN; o++) {
for (int inc = 0; inc < num_map_S2_CNN; inc++) {
if (!tbl[inc][o]) continue;

int addr1 = get_index(0, 0, num_map_S2_CNN * o + inc, width_kernel_conv_CNN, height_kernel_conv_CNN, num_map_C3_CNN * num_map_S2_CNN);
int addr2 = get_index(0, 0, inc, width_image_S2_CNN, height_image_S2_CNN, num_map_S2_CNN);
int addr3 = get_index(0, 0, o, width_image_C3_CNN, height_image_C3_CNN, num_map_C3_CNN);

const double* pw = &weight_C3[0] + addr1;
const double* pi = &neuron_S2[0] + addr2;
double* pa = &neuron_C3[0] + addr3;

for (int y = 0; y < height_image_C3_CNN; y++) {
for (int x = 0; x < width_image_C3_CNN; x++) {
const double* ppw = pw;
const double* ppi = pi + y * width_image_S2_CNN + x;
double sum = 0.0;

for (int wy = 0; wy < height_kernel_conv_CNN; wy++) {
for (int wx = 0; wx < width_kernel_conv_CNN; wx++) {
sum += *ppw++ * ppi[wy * width_image_S2_CNN + wx];
}
}

pa[y * width_image_C3_CNN + x] += sum;
}
}
}

int addr3 = get_index(0, 0, o, width_image_C3_CNN, height_image_C3_CNN, num_map_C3_CNN);
double* pa = &neuron_C3[0] + addr3;
double b = bias_C3[o];
for (int y = 0; y < height_image_C3_CNN; y++) {
for (int x = 0; x < width_image_C3_CNN; x++) {
pa[y * width_image_C3_CNN + x] += b;
}
}
}

for (int i = 0; i < num_neuron_C3_CNN; i++) {
neuron_C3[i] = activation_function_tanh(neuron_C3[i]);
}

return true;
}

bool CNN::Forward_S4()
{
double scale_factor = 1.0 / (width_kernel_pooling_CNN * height_kernel_pooling_CNN);
init_variable(neuron_S4, 0.0, num_neuron_S4_CNN);

assert(out2wi_S4.size() == num_neuron_S4_CNN);
assert(out2bias_S4.size() == num_neuron_S4_CNN);

for (int i = 0; i < num_neuron_S4_CNN; i++) {
const wi_connections& connections = out2wi_S4[i];
neuron_S4[i] = 0.0;

for (int index = 0; index < connections.size(); index++) {
neuron_S4[i] += weight_S4[connections[index].first] * neuron_C3[connections[index].second];
}

neuron_S4[i] *= scale_factor;
neuron_S4[i] += bias_S4[out2bias_S4[i]];
}

for (int i = 0; i < num_neuron_S4_CNN; i++) {
neuron_S4[i] = activation_function_tanh(neuron_S4[i]);
}

return true;
}

bool CNN::Forward_C5()
{
init_variable(neuron_C5, 0.0, num_neuron_C5_CNN);

for (int o = 0; o < num_map_C5_CNN; o++) {
for (int inc = 0; inc < num_map_S4_CNN; inc++) {
int addr1 = get_index(0, 0, num_map_S4_CNN * o + inc, width_kernel_conv_CNN, height_kernel_conv_CNN, num_map_C5_CNN * num_map_S4_CNN);
int addr2 = get_index(0, 0, inc, width_image_S4_CNN, height_image_S4_CNN, num_map_S4_CNN);
int addr3 = get_index(0, 0, o, width_image_C5_CNN, height_image_C5_CNN, num_map_C5_CNN);

const double *pw = &weight_C5[0] + addr1;
const double *pi = &neuron_S4[0] + addr2;
double *pa = &neuron_C5[0] + addr3;

for (int y = 0; y < height_image_C5_CNN; y++) {
for (int x = 0; x < width_image_C5_CNN; x++) {
const double *ppw = pw;
const double *ppi = pi + y * width_image_S4_CNN + x;
double sum = 0.0;

for (int wy = 0; wy < height_kernel_conv_CNN; wy++) {
for (int wx = 0; wx < width_kernel_conv_CNN; wx++) {
sum += *ppw++ * ppi[wy * width_image_S4_CNN + wx];
}
}

pa[y * width_image_C5_CNN + x] += sum;
}
}
}

int addr3 = get_index(0, 0, o, width_image_C5_CNN, height_image_C5_CNN, num_map_C5_CNN);
double *pa = &neuron_C5[0] + addr3;
double b = bias_C5[o];
for (int y = 0; y < height_image_C5_CNN; y++) {
for (int x = 0; x < width_image_C5_CNN; x++) {
pa[y * width_image_C5_CNN + x] += b;
}
}
}

for (int i = 0; i < num_neuron_C5_CNN; i++) {
neuron_C5[i] = activation_function_tanh(neuron_C5[i]);
}

return true;
}

bool CNN::Forward_output()
{
init_variable(neuron_output, 0.0, num_neuron_output_CNN);

for (int i = 0; i < num_neuron_output_CNN; i++) {
neuron_output[i] = 0.0;

for (int c = 0; c < num_neuron_C5_CNN; c++) {
neuron_output[i] += weight_output[c * num_neuron_output_CNN + i] * neuron_C5[c];
}

neuron_output[i] += bias_output[i];
}

for (int i = 0; i < num_neuron_output_CNN; i++) {
neuron_output[i] = activation_function_tanh(neuron_output[i]);
}

return true;
}

bool CNN::Backward_output()
{
init_variable(delta_neuron_output, 0.0, num_neuron_output_CNN);

double dE_dy[num_neuron_output_CNN];
init_variable(dE_dy, 0.0, num_neuron_output_CNN);
loss_function_gradient(neuron_output, data_single_label, dE_dy, num_neuron_output_CNN); // 损失函数: mean squared error(均方差)

// delta = dE/da = (dE/dy) * (dy/da)
for (int i = 0; i < num_neuron_output_CNN; i++) {
double dy_da[num_neuron_output_CNN];
init_variable(dy_da, 0.0, num_neuron_output_CNN);

dy_da[i] = activation_function_tanh_derivative(neuron_output[i]);
delta_neuron_output[i] = dot_product(dE_dy, dy_da, num_neuron_output_CNN);
}

return true;
}

bool CNN::Backward_C5()
{
init_variable(delta_neuron_C5, 0.0, num_neuron_C5_CNN);
init_variable(delta_weight_output, 0.0, len_weight_output_CNN);
init_variable(delta_bias_output, 0.0, len_bias_output_CNN);

for (int c = 0; c < num_neuron_C5_CNN; c++) {
// propagate delta to previous layer
// prev_delta[c] += current_delta[r] * W_[c * out_size_ + r]
delta_neuron_C5[c] = dot_product(&delta_neuron_output[0], &weight_output[c * num_neuron_output_CNN], num_neuron_output_CNN);
delta_neuron_C5[c] *= activation_function_tanh_derivative(neuron_C5[c]);
}

// accumulate weight-step using delta
// dW[c * out_size + i] += current_delta[i] * prev_out[c]
for (int c = 0; c < num_neuron_C5_CNN; c++) {
muladd(&delta_neuron_output[0], neuron_C5[c], num_neuron_output_CNN, &delta_weight_output[0] + c * num_neuron_output_CNN);
}

for (int i = 0; i < len_bias_output_CNN; i++) {
delta_bias_output[i] += delta_neuron_output[i];
}

return true;
}

bool CNN::Backward_S4()
{
init_variable(delta_neuron_S4, 0.0, num_neuron_S4_CNN);
init_variable(delta_weight_C5, 0.0, len_weight_C5_CNN);
init_variable(delta_bias_C5, 0.0, len_bias_C5_CNN);

// propagate delta to previous layer
for (int inc = 0; inc < num_map_S4_CNN; inc++) {
for (int outc = 0; outc < num_map_C5_CNN; outc++) {
int addr1 = get_index(0, 0, num_map_S4_CNN * outc + inc, width_kernel_conv_CNN, height_kernel_conv_CNN, num_map_S4_CNN * num_map_C5_CNN);
int addr2 = get_index(0, 0, outc, width_image_C5_CNN, height_image_C5_CNN, num_map_C5_CNN);
int addr3 = get_index(0, 0, inc, width_image_S4_CNN, height_image_S4_CNN, num_map_S4_CNN);

const double* pw = &weight_C5[0] + addr1;
const double* pdelta_src = &delta_neuron_C5[0] + addr2;
double* pdelta_dst = &delta_neuron_S4[0] + addr3;

for (int y = 0; y < height_image_C5_CNN; y++) {
for (int x = 0; x < width_image_C5_CNN; x++) {
const double* ppw = pw;
const double ppdelta_src = pdelta_src[y * width_image_C5_CNN + x];
double* ppdelta_dst = pdelta_dst + y * width_image_S4_CNN + x;

for (int wy = 0; wy < height_kernel_conv_CNN; wy++) {
for (int wx = 0; wx < width_kernel_conv_CNN; wx++) {
ppdelta_dst[wy * width_image_S4_CNN + wx] += *ppw++ * ppdelta_src;
}
}
}
}
}
}

for (int i = 0; i < num_neuron_S4_CNN; i++) {
delta_neuron_S4[i] *= activation_function_tanh_derivative(neuron_S4[i]);
}

// accumulate dw
for (int inc = 0; inc < num_map_S4_CNN; inc++) {
for (int outc = 0; outc < num_map_C5_CNN; outc++) {
for (int wy = 0; wy < height_kernel_conv_CNN; wy++) {
for (int wx = 0; wx < width_kernel_conv_CNN; wx++) {
int addr1 = get_index(wx, wy, inc, width_image_S4_CNN, height_image_S4_CNN, num_map_S4_CNN);
int addr2 = get_index(0, 0, outc, width_image_C5_CNN, height_image_C5_CNN, num_map_C5_CNN);
int addr3 = get_index(wx, wy, num_map_S4_CNN * outc + inc, width_kernel_conv_CNN, height_kernel_conv_CNN, num_map_S4_CNN * num_map_C5_CNN);

double dst = 0.0;
const double* prevo = &neuron_S4[0] + addr1;
const double* delta = &delta_neuron_C5[0] + addr2;

for (int y = 0; y < height_image_C5_CNN; y++) {
dst += dot_product(prevo + y * width_image_S4_CNN, delta + y * width_image_C5_CNN, width_image_C5_CNN);
}

delta_weight_C5[addr3] += dst;
}
}
}
}

// accumulate db
for (int outc = 0; outc < num_map_C5_CNN; outc++) {
int addr2 = get_index(0, 0, outc, width_image_C5_CNN, height_image_C5_CNN, num_map_C5_CNN);
const double* delta = &delta_neuron_C5[0] + addr2;

for (int y = 0; y < height_image_C5_CNN; y++) {
for (int x = 0; x < width_image_C5_CNN; x++) {
delta_bias_C5[outc] += delta[y * width_image_C5_CNN + x];
}
}
}

return true;
}

bool CNN::Backward_C3()
{
init_variable(delta_neuron_C3, 0.0, num_neuron_C3_CNN);
init_variable(delta_weight_S4, 0.0, len_weight_S4_CNN);
init_variable(delta_bias_S4, 0.0, len_bias_S4_CNN);

double scale_factor = 1.0 / (width_kernel_pooling_CNN * height_kernel_pooling_CNN);

assert(in2wo_C3.size() == num_neuron_C3_CNN);
assert(weight2io_C3.size() == len_weight_S4_CNN);
assert(bias2out_C3.size() == len_bias_S4_CNN);

for (int i = 0; i < num_neuron_C3_CNN; i++) {
const wo_connections& connections = in2wo_C3[i];
double delta = 0.0;

for (int j = 0; j < connections.size(); j++) {
delta += weight_S4[connections[j].first] * delta_neuron_S4[connections[j].second];
}

delta_neuron_C3[i] = delta * scale_factor * activation_function_tanh_derivative(neuron_C3[i]);
}

for (int i = 0; i < len_weight_S4_CNN; i++) {
const io_connections& connections = weight2io_C3[i];
double diff = 0;

for (int j = 0; j < connections.size(); j++) {
diff += neuron_C3[connections[j].first] * delta_neuron_S4[connections[j].second];
}

delta_weight_S4[i] += diff * scale_factor;
}

for (int i = 0; i < len_bias_S4_CNN; i++) {
const std::vector& outs = bias2out_C3[i];
double diff = 0;

for (int o = 0; o < outs.size(); o++) {
diff += delta_neuron_S4[outs[o]];
}

delta_bias_S4[i] += diff;
}

return true;
}

bool CNN::Backward_S2()
{
init_variable(delta_neuron_S2, 0.0, num_neuron_S2_CNN);
init_variable(delta_weight_C3, 0.0, len_weight_C3_CNN);
init_variable(delta_bias_C3, 0.0, len_bias_C3_CNN);

// propagate delta to previous layer
for (int inc = 0; inc < num_map_S2_CNN; inc++) {
for (int outc = 0; outc < num_map_C3_CNN; outc++) {
if (!tbl[inc][outc]) continue;

int addr1 = get_index(0, 0, num_map_S2_CNN * outc + inc, width_kernel_conv_CNN, height_kernel_conv_CNN, num_map_S2_CNN * num_map_C3_CNN);
int addr2 = get_index(0, 0, outc, width_image_C3_CNN, height_image_C3_CNN, num_map_C3_CNN);
int addr3 = get_index(0, 0, inc, width_image_S2_CNN, height_image_S2_CNN, num_map_S2_CNN);

const double *pw = &weight_C3[0] + addr1;
const double *pdelta_src = &delta_neuron_C3[0] + addr2;;
double* pdelta_dst = &delta_neuron_S2[0] + addr3;

for (int y = 0; y < height_image_C3_CNN; y++) {
for (int x = 0; x < width_image_C3_CNN; x++) {
const double* ppw = pw;
const double ppdelta_src = pdelta_src[y * width_image_C3_CNN + x];
double* ppdelta_dst = pdelta_dst + y * width_image_S2_CNN + x;

for (int wy = 0; wy < height_kernel_conv_CNN; wy++) {
for (int wx = 0; wx < width_kernel_conv_CNN; wx++) {
ppdelta_dst[wy * width_image_S2_CNN + wx] += *ppw++ * ppdelta_src;
}
}
}
}
}
}

for (int i = 0; i < num_neuron_S2_CNN; i++) {
delta_neuron_S2[i] *= activation_function_tanh_derivative(neuron_S2[i]);
}

// accumulate dw
for (int inc = 0; inc < num_map_S2_CNN; inc++) {
for (int outc = 0; outc < num_map_C3_CNN; outc++) {
if (!tbl[inc][outc]) continue;

for (int wy = 0; wy < height_kernel_conv_CNN; wy++) {
for (int wx = 0; wx < width_kernel_conv_CNN; wx++) {
int addr1 = get_index(wx, wy, inc, width_image_S2_CNN, height_image_S2_CNN, num_map_S2_CNN);
int addr2 = get_index(0, 0, outc, width_image_C3_CNN, height_image_C3_CNN, num_map_C3_CNN);
int addr3 = get_index(wx, wy, num_map_S2_CNN * outc + inc, width_kernel_conv_CNN, height_kernel_conv_CNN, num_map_S2_CNN * num_map_C3_CNN);

double dst = 0.0;
const double* prevo = &neuron_S2[0] + addr1;
const double* delta = &delta_neuron_C3[0] + addr2;

for (int y = 0; y < height_image_C3_CNN; y++) {
dst += dot_product(prevo + y * width_image_S2_CNN, delta + y * width_image_C3_CNN, width_image_C3_CNN);
}

delta_weight_C3[addr3] += dst;
}
}
}
}

// accumulate db
for (int outc = 0; outc < len_bias_C3_CNN; outc++) {
int addr1 = get_index(0, 0, outc, width_image_C3_CNN, height_image_C3_CNN, num_map_C3_CNN);
const double* delta = &delta_neuron_C3[0] + addr1;

for (int y = 0; y < height_image_C3_CNN; y++) {
for (int x = 0; x < width_image_C3_CNN; x++) {
delta_bias_C3[outc] += delta[y * width_image_C3_CNN + x];
}
}
}

return true;
}

bool CNN::Backward_C1()
{
init_variable(delta_neuron_C1, 0.0, num_neuron_C1_CNN);
init_variable(delta_weight_S2, 0.0, len_weight_S2_CNN);
init_variable(delta_bias_S2, 0.0, len_bias_S2_CNN);

double scale_factor = 1.0 / (width_kernel_pooling_CNN * height_kernel_pooling_CNN);

assert(in2wo_C1.size() == num_neuron_C1_CNN);
assert(weight2io_C1.size() == len_weight_S2_CNN);
assert(bias2out_C1.size() == len_bias_S2_CNN);

for (int i = 0; i < num_neuron_C1_CNN; i++) {
const wo_connections& connections = in2wo_C1[i];
double delta = 0.0;

for (int j = 0; j < connections.size(); j++) {
delta += weight_S2[connections[j].first] * delta_neuron_S2[connections[j].second];
}

delta_neuron_C1[i] = delta * scale_factor * activation_function_tanh_derivative(neuron_C1[i]);
}

for (int i = 0; i < len_weight_S2_CNN; i++) {
const io_connections& connections = weight2io_C1[i];
double diff = 0.0;

for (int j = 0; j < connections.size(); j++) {
diff += neuron_C1[connections[j].first] * delta_neuron_S2[connections[j].second];
}

delta_weight_S2[i] += diff * scale_factor;
}

for (int i = 0; i < len_bias_S2_CNN; i++) {
const std::vector& outs = bias2out_C1[i];
double diff = 0;

for (int o = 0; o < outs.size(); o++) {
diff += delta_neuron_S2[outs[o]];
}

delta_bias_S2[i] += diff;
}

return true;
}

bool CNN::Backward_input()
{
init_variable(delta_neuron_input, 0.0, num_neuron_input_CNN);
init_variable(delta_weight_C1, 0.0, len_weight_C1_CNN);
init_variable(delta_bias_C1, 0.0, len_bias_C1_CNN);

// propagate delta to previous layer
for (int inc = 0; inc < num_map_input_CNN; inc++) {
for (int outc = 0; outc < num_map_C1_CNN; outc++) {
int addr1 = get_index(0, 0, num_map_input_CNN * outc + inc, width_kernel_conv_CNN, height_kernel_conv_CNN, num_map_C1_CNN);
int addr2 = get_index(0, 0, outc, width_image_C1_CNN, height_image_C1_CNN, num_map_C1_CNN);
int addr3 = get_index(0, 0, inc, width_image_input_CNN, height_image_input_CNN, num_map_input_CNN);

const double* pw = &weight_C1[0] + addr1;
const double* pdelta_src = &delta_neuron_C1[0] + addr2;
double* pdelta_dst = &delta_neuron_input[0] + addr3;

for (int y = 0; y < height_image_C1_CNN; y++) {
for (int x = 0; x < width_image_C1_CNN; x++) {
const double* ppw = pw;
const double ppdelta_src = pdelta_src[y * width_image_C1_CNN + x];
double* ppdelta_dst = pdelta_dst + y * width_image_input_CNN + x;

for (int wy = 0; wy < height_kernel_conv_CNN; wy++) {
for (int wx = 0; wx < width_kernel_conv_CNN; wx++) {
ppdelta_dst[wy * width_image_input_CNN + wx] += *ppw++ * ppdelta_src;
}
}
}
}
}
}

for (int i = 0; i < num_neuron_input_CNN; i++) {
delta_neuron_input[i] *= activation_function_identity_derivative(data_single_image[i]/*neuron_input[i]*/);
}

// accumulate dw
for (int inc = 0; inc < num_map_input_CNN; inc++) {
for (int outc = 0; outc < num_map_C1_CNN; outc++) {
for (int wy = 0; wy < height_kernel_conv_CNN; wy++) {
for (int wx = 0; wx < width_kernel_conv_CNN; wx++) {
int addr1 = get_index(wx, wy, inc, width_image_input_CNN, height_image_input_CNN, num_map_input_CNN);
int addr2 = get_index(0, 0, outc, width_image_C1_CNN, height_image_C1_CNN, num_map_C1_CNN);
int addr3 = get_index(wx, wy, num_map_input_CNN * outc + inc, width_kernel_conv_CNN, height_kernel_conv_CNN, num_map_C1_CNN);

double dst = 0.0;
const double* prevo = data_single_image + addr1;//&neuron_input[0]
const double* delta = &delta_neuron_C1[0] + addr2;

for (int y = 0; y < height_image_C1_CNN; y++) {
dst += dot_product(prevo + y * width_image_input_CNN, delta + y * width_image_C1_CNN, width_image_C1_CNN);
}

delta_weight_C1[addr3] += dst;
}
}
}
}

// accumulate db
for (int outc = 0; outc < len_bias_C1_CNN; outc++) {
int addr1 = get_index(0, 0, outc, width_image_C1_CNN, height_image_C1_CNN, num_map_C1_CNN);
const double* delta = &delta_neuron_C1[0] + addr1;

for (int y = 0; y < height_image_C1_CNN; y++) {
for (int x = 0; x < width_image_C1_CNN; x++) {
delta_bias_C1[outc] += delta[y * width_image_C1_CNN + x];
}
}
}

return true;
}

void CNN::update_weights_bias(const double* delta, double* e_weight, double* weight, int len)
{
for (int i = 0; i < len; i++) {
e_weight[i] += delta[i] * delta[i];
weight[i] -= learning_rate_CNN * delta[i] / (std::sqrt(e_weight[i]) + eps_CNN);
}
}

bool CNN::UpdateWeights()
{
update_weights_bias(delta_weight_C1, E_weight_C1, weight_C1, len_weight_C1_CNN);
update_weights_bias(delta_bias_C1, E_bias_C1, bias_C1, len_bias_C1_CNN);

update_weights_bias(delta_weight_S2, E_weight_S2, weight_S2, len_weight_S2_CNN);
update_weights_bias(delta_bias_S2, E_bias_S2, bias_S2, len_bias_S2_CNN);

update_weights_bias(delta_weight_C3, E_weight_C3, weight_C3, len_weight_C3_CNN);
update_weights_bias(delta_bias_C3, E_bias_C3, bias_C3, len_bias_C3_CNN);

update_weights_bias(delta_weight_S4, E_weight_S4, weight_S4, len_weight_S4_CNN);
update_weights_bias(delta_bias_S4, E_bias_S4, bias_S4, len_bias_S4_CNN);

update_weights_bias(delta_weight_C5, E_weight_C5, weight_C5, len_weight_C5_CNN);
update_weights_bias(delta_bias_C5, E_bias_C5, bias_C5, len_bias_C5_CNN);

update_weights_bias(delta_weight_output, E_weight_output, weight_output, len_weight_output_CNN);
update_weights_bias(delta_bias_output, E_bias_output, bias_output, len_bias_output_CNN);

return true;
}

int CNN::predict(const unsigned char* data, int width, int height)
{
assert(data && width == width_image_input_CNN && height == height_image_input_CNN);

const double scale_min = -1;
const double scale_max = 1;

double tmp[width_image_input_CNN * height_image_input_CNN];
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++) {
tmp[y * width + x] = (data[y * width + x] / 255.0) * (scale_max - scale_min) + scale_min;
}
}

data_single_image = &tmp[0];

Forward_C1();
Forward_S2();
Forward_C3();
Forward_S4();
Forward_C5();
Forward_output();

int pos = -1;
double max_value = -9999.0;

for (int i = 0; i < num_neuron_output_CNN; i++) {
if (neuron_output[i] > max_value) {
max_value = neuron_output[i];
pos = i;
}
}

return pos;
}

bool CNN::readModelFile(const char* name)
{
FILE* fp = fopen(name, "rb");
if (fp == NULL) {
return false;
}

int width_image_input =0;
int height_image_input = 0;
int width_image_C1 = 0;
int height_image_C1 = 0;
int width_image_S2 = 0;
int height_image_S2 = 0;
int width_image_C3 = 0;
int height_image_C3 = 0;
int width_image_S4 = 0;
int height_image_S4 = 0;
int width_image_C5 = 0;
int height_image_C5 = 0;
int width_image_output = 0;
int height_image_output = 0;

int width_kernel_conv = 0;
int height_kernel_conv = 0;
int width_kernel_pooling = 0;
int height_kernel_pooling = 0;

int num_map_input = 0;
int num_map_C1 = 0;
int num_map_S2 = 0;
int num_map_C3 = 0;
int num_map_S4 = 0;
int num_map_C5 = 0;
int num_map_output = 0;

int len_weight_C1 = 0;
int len_bias_C1 = 0;
int len_weight_S2 = 0;
int len_bias_S2 = 0;
int len_weight_C3 = 0;
int len_bias_C3 = 0;
int len_weight_S4 = 0;
int len_bias_S4 = 0;
int len_weight_C5 = 0;
int len_bias_C5 = 0;
int len_weight_output = 0;
int len_bias_output = 0;

int num_neuron_input = 0;
int num_neuron_C1 = 0;
int num_neuron_S2 = 0;
int num_neuron_C3 = 0;
int num_neuron_S4 = 0;
int num_neuron_C5 = 0;
int num_neuron_output = 0;

fread(&width_image_input, sizeof(int), 1, fp);
fread(&height_image_input, sizeof(int), 1, fp);
fread(&width_image_C1, sizeof(int), 1, fp);
fread(&height_image_C1, sizeof(int), 1, fp);
fread(&width_image_S2, sizeof(int), 1, fp);
fread(&height_image_S2, sizeof(int), 1, fp);
fread(&width_image_C3, sizeof(int), 1, fp);
fread(&height_image_C3, sizeof(int), 1, fp);
fread(&width_image_S4, sizeof(int), 1, fp);
fread(&height_image_S4, sizeof(int), 1, fp);
fread(&width_image_C5, sizeof(int), 1, fp);
fread(&height_image_C5, sizeof(int), 1, fp);
fread(&width_image_output, sizeof(int), 1, fp);
fread(&height_image_output, sizeof(int), 1, fp);

fread(&width_kernel_conv, sizeof(int), 1, fp);
fread(&height_kernel_conv, sizeof(int), 1, fp);
fread(&width_kernel_pooling, sizeof(int), 1, fp);
fread(&height_kernel_pooling, sizeof(int), 1, fp);

fread(&num_map_input, sizeof(int), 1, fp);
fread(&num_map_C1, sizeof(int), 1, fp);
fread(&num_map_S2, sizeof(int), 1, fp);
fread(&num_map_C3, sizeof(int), 1, fp);
fread(&num_map_S4, sizeof(int), 1, fp);
fread(&num_map_C5, sizeof(int), 1, fp);
fread(&num_map_output, sizeof(int), 1, fp);

fread(&len_weight_C1, sizeof(int), 1, fp);
fread(&len_bias_C1, sizeof(int), 1, fp);
fread(&len_weight_S2, sizeof(int), 1, fp);
fread(&len_bias_S2, sizeof(int), 1, fp);
fread(&len_weight_C3, sizeof(int), 1, fp);
fread(&len_bias_C3, sizeof(int), 1, fp);
fread(&len_weight_S4, sizeof(int), 1, fp);
fread(&len_bias_S4, sizeof(int), 1, fp);
fread(&len_weight_C5, sizeof(int), 1, fp);
fread(&len_bias_C5, sizeof(int), 1, fp);
fread(&len_weight_output, sizeof(int), 1, fp);
fread(&len_bias_output, sizeof(int), 1, fp);

fread(&num_neuron_input, sizeof(int), 1, fp);
fread(&num_neuron_C1, sizeof(int), 1, fp);
fread(&num_neuron_S2, sizeof(int), 1, fp);
fread(&num_neuron_C3, sizeof(int), 1, fp);
fread(&num_neuron_S4, sizeof(int), 1, fp);
fread(&num_neuron_C5, sizeof(int), 1, fp);
fread(&num_neuron_output, sizeof(int), 1, fp);

fread(weight_C1, sizeof(weight_C1), 1, fp);
fread(bias_C1, sizeof(bias_C1), 1, fp);
fread(weight_S2, sizeof(weight_S2), 1, fp);
fread(bias_S2, sizeof(bias_S2), 1, fp);
fread(weight_C3, sizeof(weight_C3), 1, fp);
fread(bias_C3, sizeof(bias_C3), 1, fp);
fread(weight_S4, sizeof(weight_S4), 1, fp);
fread(bias_S4, sizeof(bias_S4), 1, fp);
fread(weight_C5, sizeof(weight_C5), 1, fp);
fread(bias_C5, sizeof(bias_C5), 1, fp);
fread(weight_output, sizeof(weight_output), 1, fp);
fread(bias_output, sizeof(bias_output), 1, fp);

fflush(fp);
fclose(fp);

out2wi_S2.clear();
out2bias_S2.clear();
out2wi_S4.clear();
out2bias_S4.clear();

calc_out2wi(width_image_C1_CNN, height_image_C1_CNN, width_image_S2_CNN, height_image_S2_CNN, num_map_S2_CNN, out2wi_S2);
calc_out2bias(width_image_S2_CNN, height_image_S2_CNN, num_map_S2_CNN, out2bias_S2);
calc_out2wi(width_image_C3_CNN, height_image_C3_CNN, width_image_S4_CNN, height_image_S4_CNN, num_map_S4_CNN, out2wi_S4);
calc_out2bias(width_image_S4_CNN, height_image_S4_CNN, num_map_S4_CNN, out2bias_S4);

return true;
}

bool CNN::saveModelFile(const char* name)
{
FILE* fp = fopen(name, "wb");
if (fp == NULL) {
return false;
}

int width_image_input = width_image_input_CNN;
int height_image_input = height_image_input_CNN;
int width_image_C1 = width_image_C1_CNN;
int height_image_C1 = height_image_C1_CNN;
int width_image_S2 = width_image_S2_CNN;
int height_image_S2 = height_image_S2_CNN;
int width_image_C3 = width_image_C3_CNN;
int height_image_C3 = height_image_C3_CNN;
int width_image_S4 = width_image_S4_CNN;
int height_image_S4 = height_image_S4_CNN;
int width_image_C5 = width_image_C5_CNN;
int height_image_C5 = height_image_C5_CNN;
int width_image_output = width_image_output_CNN;
int height_image_output = height_image_output_CNN;

int width_kernel_conv = width_kernel_conv_CNN;
int height_kernel_conv = height_kernel_conv_CNN;
int width_kernel_pooling = width_kernel_pooling_CNN;
int height_kernel_pooling = height_kernel_pooling_CNN;

int num_map_input = num_map_input_CNN;
int num_map_C1 = num_map_C1_CNN;
int num_map_S2 = num_map_S2_CNN;
int num_map_C3 = num_map_C3_CNN;
int num_map_S4 = num_map_S4_CNN;
int num_map_C5 = num_map_C5_CNN;
int num_map_output = num_map_output_CNN;

int len_weight_C1 = len_weight_C1_CNN;
int len_bias_C1 = len_bias_C1_CNN;
int len_weight_S2 = len_weight_S2_CNN;
int len_bias_S2 = len_bias_S2_CNN;
int len_weight_C3 = len_weight_C3_CNN;
int len_bias_C3 = len_bias_C3_CNN;
int len_weight_S4 = len_weight_S4_CNN;
int len_bias_S4 = len_bias_S4_CNN;
int len_weight_C5 = len_weight_C5_CNN;
int len_bias_C5 = len_bias_C5_CNN;
int len_weight_output = len_weight_output_CNN;
int len_bias_output = len_bias_output_CNN;

int num_neuron_input = num_neuron_input_CNN;
int num_neuron_C1 = num_neuron_C1_CNN;
int num_neuron_S2 = num_neuron_S2_CNN;
int num_neuron_C3 = num_neuron_C3_CNN;
int num_neuron_S4 = num_neuron_S4_CNN;
int num_neuron_C5 = num_neuron_C5_CNN;
int num_neuron_output = num_neuron_output_CNN;

fwrite(&width_image_input, sizeof(int), 1, fp);
fwrite(&height_image_input, sizeof(int), 1, fp);
fwrite(&width_image_C1, sizeof(int), 1, fp);
fwrite(&height_image_C1, sizeof(int), 1, fp);
fwrite(&width_image_S2, sizeof(int), 1, fp);
fwrite(&height_image_S2, sizeof(int), 1, fp);
fwrite(&width_image_C3, sizeof(int), 1, fp);
fwrite(&height_image_C3, sizeof(int), 1, fp);
fwrite(&width_image_S4, sizeof(int), 1, fp);
fwrite(&height_image_S4, sizeof(int), 1, fp);
fwrite(&width_image_C5, sizeof(int), 1, fp);
fwrite(&height_image_C5, sizeof(int), 1, fp);
fwrite(&width_image_output, sizeof(int), 1, fp);
fwrite(&height_image_output, sizeof(int), 1, fp);

fwrite(&width_kernel_conv, sizeof(int), 1, fp);
fwrite(&height_kernel_conv, sizeof(int), 1, fp);
fwrite(&width_kernel_pooling, sizeof(int), 1, fp);
fwrite(&height_kernel_pooling, sizeof(int), 1, fp);

fwrite(&num_map_input, sizeof(int), 1, fp);
fwrite(&num_map_C1, sizeof(int), 1, fp);
fwrite(&num_map_S2, sizeof(int), 1, fp);
fwrite(&num_map_C3, sizeof(int), 1, fp);
fwrite(&num_map_S4, sizeof(int), 1, fp);
fwrite(&num_map_C5, sizeof(int), 1, fp);
fwrite(&num_map_output, sizeof(int), 1, fp);

fwrite(&len_weight_C1, sizeof(int), 1, fp);
fwrite(&len_bias_C1, sizeof(int), 1, fp);
fwrite(&len_weight_S2, sizeof(int), 1, fp);
fwrite(&len_bias_S2, sizeof(int), 1, fp);
fwrite(&len_weight_C3, sizeof(int), 1, fp);
fwrite(&len_bias_C3, sizeof(int), 1, fp);
fwrite(&len_weight_S4, sizeof(int), 1, fp);
fwrite(&len_bias_S4, sizeof(int), 1, fp);
fwrite(&len_weight_C5, sizeof(int), 1, fp);
fwrite(&len_bias_C5, sizeof(int), 1, fp);
fwrite(&len_weight_output, sizeof(int), 1, fp);
fwrite(&len_bias_output, sizeof(int), 1, fp);

fwrite(&num_neuron_input, sizeof(int), 1, fp);
fwrite(&num_neuron_C1, sizeof(int), 1, fp);
fwrite(&num_neuron_S2, sizeof(int), 1, fp);
fwrite(&num_neuron_C3, sizeof(int), 1, fp);
fwrite(&num_neuron_S4, sizeof(int), 1, fp);
fwrite(&num_neuron_C5, sizeof(int), 1, fp);
fwrite(&num_neuron_output, sizeof(int), 1, fp);

fwrite(weight_C1, sizeof(weight_C1), 1, fp);
fwrite(bias_C1, sizeof(bias_C1), 1, fp);
fwrite(weight_S2, sizeof(weight_S2), 1, fp);
fwrite(bias_S2, sizeof(bias_S2), 1, fp);
fwrite(weight_C3, sizeof(weight_C3), 1, fp);
fwrite(bias_C3, sizeof(bias_C3), 1, fp);
fwrite(weight_S4, sizeof(weight_S4), 1, fp);
fwrite(bias_S4, sizeof(bias_S4), 1, fp);
fwrite(weight_C5, sizeof(weight_C5), 1, fp);
fwrite(bias_C5, sizeof(bias_C5), 1, fp);
fwrite(weight_output, sizeof(weight_output), 1, fp);
fwrite(bias_output, sizeof(bias_output), 1, fp);

fflush(fp);
fclose(fp);

return true;
}

double CNN::test()
{
int count_accuracy = 0;

for (int num = 0; num < num_patterns_test_CNN; num++) {
data_single_image = data_input_test + num * num_neuron_input_CNN;
data_single_label = data_output_test + num * num_neuron_output_CNN;

Forward_C1();
Forward_S2();
Forward_C3();
Forward_S4();
Forward_C5();
Forward_output();

int pos_t = -1;
int pos_y = -2;
double max_value_t = -9999.0;
double max_value_y = -9999.0;

for (int i = 0; i < num_neuron_output_CNN; i++) {
if (neuron_output[i] > max_value_y) {
max_value_y = neuron_output[i];
pos_y = i;
}

if (data_single_label[i] > max_value_t) {
max_value_t = data_single_label[i];
pos_t = i;
}
}

if (pos_y == pos_t) {
++count_accuracy;
}

Sleep(1);
}

return (count_accuracy * 1.0 / num_patterns_test_CNN);
}

}

测试代码如下:
int test_CNN_train()
{
ANN::CNN cnn1;
cnn1.init();
cnn1.train();

return 0;
}

int test_CNN_predict()
{
ANN::CNN cnn2;
bool flag = cnn2.readModelFile("E:/GitCode/NN_Test/data/cnn.model");
if (!flag) {
std::cout << "read cnn model error" << std::endl;
return -1;
}

int width{ 32 }, height{ 32 };
std::vector target{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };
std::string image_path{ "E:/GitCode/NN_Test/data/images/" };

for (auto i : target) {
std::string str = std::to_string(i);
str += ".png";
str = image_path + str;

cv::Mat src = cv::imread(str, 0);
if (src.data == nullptr) {
fprintf(stderr, "read image error: %s\n", str.c_str());
return -1;
}

cv::Mat tmp(src.rows, src.cols, CV_8UC1, cv::Scalar::all(255));
cv::subtract(tmp, src, tmp);

cv::resize(tmp, tmp, cv::Size(width, height));

auto ret = cnn2.predict(tmp.data, width, height);

fprintf(stdout, "the actual digit is: %d, correct digit is: %d\n", ret, i);
}

return 0;
}

通过执行test_CNN_train()函数可生成cnn model文件,执行结果如下:

通过执行test_CNN_predict()函数来测试CNN的准确率,通过画图工具,每个数字生成一张图像,共10幅,如下图:

测试结果如下:

代码实现解析见: http://blog.csdn.net/fengbingchun/article/details/53445209
GitHub:https://github.com/fengbingchun/NN

文章转载自:fengbingchun的博客