基于C++利用神经网络实现mnist手写数据集识别（源码+数据）.rar

#include <iostream> #include <fstream> #include <random> using namespace std; #include <Windows.h> #include <windowsx.h> double LEARNING_RATE; #include "Neuron.h" #include "Layer.h" #include "Connections.h" struct Instance { unsigned char image[28 * 28]; unsigned char label; void display()const { for (unsigned i = 0; i < 28; ++i) { for (unsigned j = 0; j < 28; ++j) if (image[28 * i + j] > 0b10000000ui8) cout << "▇"; else cout << ' '; cout << '\n'; } cout << (int)label << endl; } }trainingSet[60000], testSet[10000]; void readFile(const char* const& imagePath, const char* const& labelPath, Instance* const& instances, const unsigned& n) { ifstream imageFile(imagePath, ios::in | ios::binary); ifstream lableFile(labelPath, ios::in | ios::binary); imageFile.seekg(16); lableFile.seekg(8); for (unsigned i = 0; i < n; ++i) { imageFile.read((char*)instances[i].image, 28 * 28); lableFile.get((char&)instances[i].label); } } struct functions { static inline double 激活函数(double const& x) { return tanh(x); } static inline double 激活函数的导数(double const& x) { register double y = tanh(x); return 1 - y * y; } }; unsigned number; Layer* pLayer; Connections* pConnections; void menu(unsigned& n) { system("cls"); cout << "================================================================\n"; cout << " 1 - 学习 2 - 测试\n"; cout << " 3 - 手写识别 0 - 结束\n"; cin >> n; } void 学习() { unsigned begin, end, batch; cout << "起始序号(e.g.0): "; cin >> begin; cout << "终止序号(e.g.60000): "; cin >> end; cout << "每组个数(e.g.1): "; cin >> batch; cout << "学习速率(e.g.0.01): "; cin >> LEARNING_RATE; cout << " ]\r["; unsigned i = 0; unsigned progress = 0; unsigned const step = (end - begin) / 33 + 1; for (; begin < end; ++begin) { if (++progress == step) { progress = 0; cout << '#'; } pLayer[0].setOutput(trainingSet[begin].image); for (unsigned j = 0; j < number; ++j) { pConnections[j].transmit(); pLayer[j + 1].calculate<functions>(); } pLayer[number].expectedOutput(trainingSet[begin].label); for (unsigned j = number - 1; j > 0; --j) { pLayer[j + 1].adjust<functions>(); pConnections[j].adjust(); pLayer[j].clearOutput(); pConnections[j].pass(); } pLayer[1].adjust<functions>(); pConnections[0].adjust(); for (unsigned j = 1; j < number + 1; ++j) pLayer[j].clearInput(); if (++i == batch) { i = 0; for (unsigned j = 0; j < number; ++j) { pConnections[j].learn(batch, LEARNING_RATE); pLayer[j + 1].learn(batch, LEARNING_RATE); } } } } void 测试() { cout << " ]\r["; unsigned sum = 0; unsigned progress = 0; unsigned const step = 10000 / 33 + 1; for (unsigned i = 0; i < 10000; ++i) { if (++progress == step) { progress = 0; cout << '#'; } pLayer[0].setOutput(testSet[i].image); for (unsigned j = 0; j < number; ++j) { pConnections[j].transmit(); pLayer[j + 1].calculate<functions>(); } unsigned iMax; double max = -1; unsigned j = 0; for (Neuron const& a : pLayer[number]) { if (a.output() > max) { max = a.output(); iMax = j; } ++j; } if (iMax == testSet[i].label) ++sum; for (unsigned j = 1; j < number + 1; ++j) pLayer[j].clearInput(); } cout << "\n正确率" << sum << "/10000\n"; } LRESULT CALLBACK WndProc(HWND hWindow, UINT message, WPARAM wParam, LPARAM lParam) { static HPEN hPen; static HBRUSH hBrush[17]; static unsigned char pixels[28 * 28]; bool (*changeColor)(int const& x, int const& y, int const& X, int const& Y) = [](int const& x, int const& y, int const& X, int const& Y)->bool { float const dx = (X * 20 + 10 - x) / 20.f; float const dy = (Y * 20 + 10 - y) / 20.f; float const distance = dx * dx + dy * dy; if (distance >= 2.25f) return false; unsigned char const color = distance <= 1 ? 0xffui8 : unsigned(-0xff * 0.8f * (distance - 2.25f)); if (pixels[Y * 28 + X] < color) { pixels[Y * 28 + X] = color; return true; } return false; }; void (*paintColor)(unsigned const& X, unsigned const& Y, HWND const& hWindow) = [](unsigned const& X, unsigned const& Y, HWND const& hWindow) { HDC hdc = GetDC(hWindow); SelectObject(hdc, hPen); SelectObject(hdc, hBrush[16 - (pixels[Y * 28 + X] + 8) / 16]); Rectangle(hdc, X * 20, Y * 20, (X + 1) * 20, (Y + 1) * 20); ReleaseDC(hWindow, hdc); }; switch (message) { case WM_CREATE: hPen = CreatePen(PS_SOLID, 0, 0x00FFFFFFui32); for (unsigned i = 0; i < 16; ++i) hBrush[i] = CreateSolidBrush(0x00101010ui32 * i); hBrush[16] = CreateSolidBrush(0x00FFFFFFui32); memset(pixels, 0, 28 * 28); return 0; case WM_DESTROY: DeleteObject(hPen); for (unsigned i = 0; i < 17; ++i) DeleteObject(hBrush[i]); PostQuitMessage(0); return 0; case WM_LBUTTONDOWN: { HDC hdc = GetDC(hWindow); SelectObject(hdc, hPen); for (int Y = 0; Y < 28; ++Y) for (int X = 0; X < 28; ++X) { SelectObject(hdc, hBrush[16 - (pixels[Y * 28 + X] + 8) / 16]); Rectangle(hdc, X * 20, Y * 20, (X + 1) * 20, (Y + 1) * 20); } ReleaseDC(hWindow, hdc); } case WM_MOUSEMOVE: if (wParam & MK_LBUTTON) { int const x = GET_X_LPARAM(lParam), y = GET_Y_LPARAM(lParam); int const X = x / 20, Y = y / 20; if (changeColor(x, y, X, Y)) paintColor(X, Y, hWindow); if (X > 0) if (changeColor(x, y, X - 1, Y)) paintColor(X - 1, Y, hWindow); if (Y > 0) if (changeColor(x, y, X, Y - 1)) paintColor(X, Y - 1, hWindow); if (X < 27) if (changeColor(x, y, X + 1, Y)) paintColor(X + 1, Y, hWindow); if (Y < 27) if (changeColor(x, y, X, Y + 1)) paintColor(X, Y + 1, hWindow); if (X > 0 && Y > 0) if (changeColor(x, y, X - 1, Y - 1)) paintColor(X - 1, Y - 1, hWindow); if (X > 0 && Y < 27) if (changeColor(x, y, X - 1, Y + 1)) paintColor(X - 1, Y + 1, hWindow); if (X < 27 && Y > 0) if (changeColor(x, y, X + 1, Y - 1)) paintColor(X + 1, Y - 1, hWindow); if (X < 27 && Y < 27) if (changeColor(x, y, X + 1, Y + 1)) paintColor(X + 1, Y + 1, hWindow); } return 0; case WM_LBUTTONUP: { pLayer[0].setOutput(pixels); for (unsigned i = 0; i < number; ++i) { pConnections[i].transmit(); pLayer[i + 1].calculate<functions>(); } double sum = 10; for (Neuron const& a : pLayer[number]) sum += a.output(); unsigned values[10]; unsigned i = 0; for (Neuron const& a : pLayer[number]) values[i++] = (a.output() + 1) / sum * 100; HDC hdc = GetDC(hWindow); SelectObject(hdc, hPen); SelectObject(hdc, hBrush[16]); Rectangle(hdc, 0, 0, 90, 160); WCHAR w[256]; wsprintf(w, L"0的概率%hu%%\n1的概率%hu%%\n2的概率%hu%%\n3的概率%hu%%\n4的概率%hu%%\n\ 5的概率%hu%%\n6的概率%hu%%\n7的概率%hu%%\n8的概率%hu%%\n9的概率%hu%%\n", values[0], values[1], values[2], values[3], values[4], values[5], values[6], values[7], values[8], values[9]); RECT rect = { 0,0,560,560 }; DrawText(hdc, w, -1, &rect, DT_LEFT | DT_NOCLIP | DT_NOPREFIX); ReleaseDC(hWindow, hdc); for (i = 1; i < number + 1; ++i) pLayer[i].clearInput(); return 0; } } return DefWindowProc(hWindow, message, wParam, lParam); } void 手写识别() { HINSTANCE hInstance = GetModuleHandle(nullptr); WNDCLASSEX window = { sizeof(WNDCLASSEX), CS_HREDRAW | CS_VREDRAW, WndProc,0,0,hInstance, LoadIcon(nullptr, IDI_APPLICATION), LoadCursor(nullptr, IDC_ARROW), (HBRUSH)(COLOR_WINDOW + 1), nullptr,L"ClassName", LoadIcon(nullptr, IDI_APPLICATION) }; RegisterClassEx(&window); RECT rectangle = { 0,0,560,560 }; AdjustWindowRect(&rectangle, WS_OVERLAPPEDWINDOW, false); HWND handleWindow = CreateWindow(window.lpszClassName, L"左键滑动", WS_CAPTION | WS_SYSMENU, CW_USEDEFAULT, 0, rectangle.right - rectangle.left, rectangle.bottom - rectangle.top, nullptr, nullptr, hInstance, nullptr); UpdateWindow(handleWindow); ShowWind