深度学习实战 | 卷积神经网络LeNet手写数字识别（带手写板GUI界面）

引言

在深度学习领域，卷积神经网络（Convolutional Neural Network, CNN）是一种广泛应用于图像识别任务的神经网络结构。LeNet是一种经典的CNN结构，被广泛应用于基础的图像分类任务。本文将介绍如何使用LeNet卷积神经网络实现手写数字识别，并使用Pytorch实现LeNet手写数字识别，使用PyQt5实现手写板GUI界面，使用户能够通过手写板输入数字并进行识别。

请添加图片描述

完整代码下载：Python手写数字识别带手写板GUI界面 Pytorch代码含训练模型（付费资源，如果你觉得这篇博客对你有帮助，欢迎购买支持~）

1. LeNet卷积神经网络

LeNet是由Yann LeCun等人于1998年提出的卷积神经网络结构，主要用于手写字符识别。在本文中，我们将使用LeNet结构构建一个用于手写数字识别的神经网络模型。以下是LeNet的基本结构：

请添加图片描述

Layer 1: Convolutional Layer
    - Input: 28x28x1 (灰度图像)
    - Filter: 5x5, Stride: 1, Depth: 6
    - Activation: Sigmoid
    - Output: 28x28x6

Layer 2: Average Pooling Layer
    - Input: 28x28x6
    - Pooling: 2x2, Stride: 2
    - Output: 14x14x6

Layer 3: Convolutional Layer
    - Input: 14x14x6
    - Filter: 5x5, Stride: 1, Depth: 16
    - Activation: Sigmoid
    - Output: 10x10x16

Layer 4: Average Pooling Layer
    - Input: 10x10x16
    - Pooling: 2x2, Stride: 2
    - Output: 5x5x16

Layer 5: Fully Connected Layer
    - Input: 5x5x16
    - Output: 120
    - Activation: Sigmoid

Layer 6: Fully Connected Layer
    - Input: 120
    - Output: 84
    - Activation: Sigmoid

Layer 7: Output Layer
    - Input: 84
    - Output: 10 (对应0-9的数字)
    - Activation: Softmax

2. 手写数字识别实现

使用深度学习框架（例如Pytorch）构建LeNet模型：

import torch
import torch.nn as nn
import torch.nn.functional as F

class LeNet(nn.Module):
    def __init__(self):
        super(LeNet, self).__init__()
        self.conv1 = nn.Conv2d(1, 6, kernel_size=5)
        self.pool1 = nn.AvgPool2d(kernel_size=2, stride=2)
        self.conv2 = nn.Conv2d(6, 16, kernel_size=5)
        self.pool2 = nn.AvgPool2d(kernel_size=2, stride=2)
        self.fc1 = nn.Linear(16 * 5 * 5, 120)
        self.fc2 = nn.Linear(120, 84)
        self.fc3 = nn.Linear(84, 10)

    def forward(self, x):
        x = F.sigmoid(self.conv1(x))
        x = self.pool1(x)
        x = F.sigmoid(self.conv2(x))
        x = self.pool2(x)
        x = x.view(-1, 16 * 5 * 5)
        x = F.sigmoid(self.fc1(x))
        x = F.sigmoid(self.fc2(x))
        x = self.fc3(x)
        return F.log_softmax(x, dim=1)

并使用手写数字数据集MNIST进行训练。确保正确实现数据预处理和模型训练过程：

import torch
import torch.nn as nn
import torch.optim as optim
from torchvision import datasets, transforms
from torch.autograd import Variable
from net import Net

if __name__ == "__main__":
    # 设置训练参数
    batch_size = 64
    epochs = 140
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

    # 数据集
    transform = transforms.Compose([transforms.ToTensor(),
                                    transforms.Normalize((0.5,), (0.5,))])
    trainset = datasets.MNIST('~/.pytorch/MNIST_data/', download=True, train=True, transform=transform)
    trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size, shuffle=True)

    # 输出提示信息
    print("batch_size:", batch_size)
    print("data_batches:", len(trainloader))
    print("epochs:", epochs)

    # 神经网络
    net = Net().to(device)
    net.load_state_dict(torch.load('model.pth'))

    # 损失函数和优化器
    criterion = nn.NLLLoss()
    optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)

    # 训练网络
    for epoch in range(epochs):
        running_loss = 0.0
        for i, data in enumerate(trainloader, 0):
            inputs, labels = data
            inputs, labels = Variable(inputs).to(device), Variable(labels).to(device)

            # 反向传播优化参数
            optimizer.zero_grad()
            outputs = net(inputs)
            loss = criterion(outputs, labels)
            loss.backward()
            optimizer.step()
            running_loss += loss.item()

            if i % 938 == 937:    # 每轮输出损失值
                print('[epoch: %d, batches: %d] loss: %.5f' %
                      (epoch + 1, i + 1, running_loss / 2000))
                running_loss = 0.0
        torch.save(net.state_dict(), './model.pth')  # 每轮保存模型参数

    print('Finished Training')

3. 手写板GUI界面开发

模型训练完成后，为了让用户通过手写板输入数字，我们将开发一个简单直观的GUI界面。使用GUI库（例如PyQt5），创建一个窗口，包含一个手写板区域，用户可以在上面写数字。添加一个识别按钮，点击后将手写板上的数字送入LeNet模型进行识别，并在界面上显示识别结果。

以下是PyQt5代码示例：

from PyQt5.QtWidgets import *
from PyQt5.QtGui import *
from PyQt5.QtCore import *
import sys

import torch
from utils import *
from net import Net

class MainWindow(QMainWindow):
    def __init__(self):
        super().__init__()
        self.title = '手写数字识别'
        self.initUI()

    def initUI(self):
        self.setWindowTitle(self.title)
        self.setMinimumSize(500, 400)
        self.main_widget = QWidget()
        self.main_layout = QGridLayout()
        self.main_widget.setLayout(self.main_layout)
        self.setCentralWidget(self.main_widget)
        self.canvas = Canvas()
        self.canvas.setFixedSize(300,300)
        self.label = QLabel()
        self.label.setFixedSize(100,100)
        self.label.setText('识别结果')
        self.label.setStyleSheet("font-size:15px;color:red") 
        self.clear_button = QPushButton('清除')
        self.clear_button.setFixedSize(100,50)
        self.clear_button.clicked.connect(self.canvas.clear)
        self.recognize_button = QPushButton('识别')
        self.recognize_button.setFixedSize(100,50)
        self.recognize_button.clicked.connect(self.recognize)
        self.main_layout.addWidget(self.canvas,0,0,3,1)
        self.main_layout.addWidget(self.label,0,1)
        self.main_layout.addWidget(self.clear_button,1,1)
        self.main_layout.addWidget(self.recognize_button,2,1)
    
    def recognize(self):
        self.canvas.recognize()
        self.label.setText('识别结果: ' + str(self.canvas.recognize()))

class Canvas(QLabel):
    x0=-10; y0=-10; x1=-10; y1=-10
    def __init__(self):
        super(Canvas,self).__init__()
        self.pixmap = QPixmap(300, 300)
        self.pixmap.fill(Qt.white)
        self.Color=Qt.blue
        self.penwidth=10

    def paintEvent(self,event):
        painter=QPainter(self.pixmap)
        painter.setPen(QPen(self.Color,self.penwidth,Qt.SolidLine))
        painter.drawLine(self.x0,self.y0,self.x1,self.y1)

        Label_painter=QPainter(self)
        Label_painter.drawPixmap(2,2,self.pixmap)

    def mousePressEvent(self, event):
        self.x1=event.x()
        self.y1=event.y()

    def mouseMoveEvent(self, event):
        self.x0 = self.x1
        self.y0 = self.y1
        self.x1 = event.x()
        self.y1 = event.y()
        self.update()

    def clear(self):
        self.x0=-10; self.y0=-10; self.x1=-10; self.y1=-10
        self.pixmap.fill(Qt.white)
        self.update()
    
    def recognize(self):
        arr = pixmap2np(self.pixmap)
        arr = 255 - arr[:,:,2]
        arr = clip_image(arr)
        arr = resize_image(arr)
        arr = np.expand_dims(arr, axis=0)
        arr_batch = np.expand_dims(arr, axis=0)
        tensor = torch.FloatTensor(arr_batch)
        tensor = (tensor/255 - 0.5) * 2
        possibles = net(tensor).detach().numpy()
        result = np.argmax(possibles)
        return result

if __name__ == '__main__':
    net = Net()
    net.load_state_dict(torch.load('model.pth'))
    app = QApplication(sys.argv)
    win = MainWindow()
    win.show()
    sys.exit(app.exec_())