【深度学习】LeNet网络架构

---

什么是LeNet

LeNet是一种经典的卷积神经网络，由Yann LeCun等人在1998年提出。它是深度学习中第一个成功应用于手写数字识别的卷积神经网络，并且被认为是现代卷积神经网络的基础。

LeNet模型包含了多个卷积层和池化层，以及最后的全连接层用于分类。其中，每个卷积层都包含了一个卷积操作和一个非线性激活函数，用于提取输入图像的特征。池化层则用于缩小特征图的尺寸，减少模型参数和计算量。全连接层则将特征向量映射到类别概率上。

---

代码实现网络架构

如何搭建网络模型参考博客：Pytorch学习笔记（模型训练）
我们采用CIFAR-10数据集进行训练测试，上面网络模型是1个channel的32x32，而我们的数据集是3个channel的32x32，模型结构不变，改变一下输入输出大小。
model.py:

import torch
from torch import nn


# 搭建网络模型
class LeNet(nn.Module):
    def __init__(self):
        super(LeNet, self).__init__()
        self.model = nn.Sequential(
            nn.Conv2d(3, 16, kernel_size=5, stride=1, padding=0),
            nn.ReLU(),
            nn.MaxPool2d(kernel_size=2),
            nn.Conv2d(16, 32, kernel_size=5, stride=1, padding=0),
            nn.ReLU(),
            nn.MaxPool2d(kernel_size=2),
            nn.Flatten(),
            nn.Linear(32 * 5 * 5, 120),
            nn.ReLU(),
            nn.Linear(120, 84),
            nn.ReLU(),
            nn.Linear(84, 10),
        )

    def forward(self, x):
        x = self.model(x)
        return x


# 测试
if __name__ == '__main__':
    leNet = LeNet()
    input = torch.ones((64, 3, 32, 32))
    output = leNet(input)
    print(output.shape)

train.py

import torch.optim
import torchvision.datasets
from torch import nn
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter

from learning.lenet.model import LeNet

# 1. 数据集
dataset_train = torchvision.datasets.CIFAR10("./data", train=True, transform=torchvision.transforms.ToTensor(),
                                             download=True)
dataset_test = torchvision.datasets.CIFAR10("./data", train=True, transform=torchvision.transforms.ToTensor(),

                                            download=True)
train_data_size = len(dataset_train)
test_data_size = len(dataset_test)
# 2. 加载数据集
dataloader_train = DataLoader(dataset_train, batch_size=64)
dataloader_test = DataLoader(dataset_test, batch_size=64)

# 3. 搭建model
leNet = LeNet()
if torch.cuda.is_available():
    leNet = leNet.cuda()

# 4. 创建损失函数
loss_fn = nn.CrossEntropyLoss()
if torch.cuda.is_available():
    loss_fn = loss_fn.cuda()

# 5. 优化器
learning_rate = 0.1
optimizer = torch.optim.SGD(leNet.parameters(), lr=learning_rate)  # 随机梯度下降

# 6. 设置训练网络的一些参数
total_train_step = 0  # 记录训练次数
total_test_step = 0  # 训练测试次数
epoch = 5  # 训练轮数

# 补充tensorboard
writer = SummaryWriter("../../logs")

# 开始训练
for i in range(epoch):
    print(f"--------第{i+1}轮训练开始--------")
    # 训练
    leNet.train()
    for data in dataloader_train:
        imgs, targets = data
        if torch.cuda.is_available():
            imgs = imgs.cuda()
            targets = targets.cuda()
        outputs = leNet(imgs)
        loss = loss_fn(outputs, targets)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        total_train_step += 1
        if total_train_step % 100 == 0:
            print(f"训练次数：{total_train_step}---loss：{loss.item()}")
            writer.add_scalar("train_loss", loss.item(), total_train_step)

    # 测试
    leNet.eval()
    total_test_loss = 0  # 总体的误差
    total_accuracy = 0  # 总体的正确率
    with torch.no_grad():
        for data in dataloader_test:
            imgs, targets = data
            if torch.cuda.is_available():
                imgs = imgs.cuda()
                targets = targets.cuda()
            outputs = leNet(imgs)
            loss = loss_fn(outputs, targets)
            total_test_loss += loss.item()
            accuracy = (outputs.argmax(1) == targets).sum()
            total_accuracy += accuracy
    print(f"整体测试集上的loss：{total_test_loss}")
    print(f"整体测试集上的准确率：{total_accuracy/test_data_size}")
    writer.add_scalar("test_loss", total_test_loss, total_test_step)
    writer.add_scalar("total_accuracy", total_accuracy/test_data_size, total_test_step)
    total_test_step += 1

    # 保存每一轮训练的模型
    torch.save(leNet, f"leNet_{i+1}.pth")
    print("模式已保存")


writer.close()

5轮训练中，第5轮的准确率是最高的，采用第5轮的模型进行测试：

test.py

import torch
import torchvision.transforms
from PIL import Image

from learning.lenet.model import LeNet


# 需要测试的图片
image_path = "../../imgs/airplane.png"
image = Image.open(image_path)
image = image.convert('RGB')  # png图片多了一个透明度通道，修改成rgb三个通道
transform = torchvision.transforms.Compose([torchvision.transforms.Resize((32, 32)),
                                            torchvision.transforms.ToTensor()])
image = transform(image)
print(image.shape)


# 引入网络架构



# 读取网络模型  如果保存的模型是通过gpu训练出来的，需要添加 map_location=torch.device("cpu")
model_load = torch.load("leNet_5.pth", map_location=torch.device("cpu"))
# 原有的图片是没有bitch-size的，而我们的输入是需要的
image = torch.reshape(image, (1, 3, 32, 32))
model_load.eval()
with torch.no_grad():
    outputs = model_load(image)
print(outputs)

classes = ('plane', 'car', 'bird', 'cat',
               'deer', 'dog', 'frog', 'horse', 'ship', 'truck')

print(classes[outputs.argmax(1)])

---