python实战（十五）——中文手写体数字图像CNN分类

一、任务背景

        本次python实战，我们使用来自Kaggle的数据集《Chinese MNIST》进行CNN分类建模，不同于经典的MNIST数据集，我们这次使用的数据集是汉字手写体数字。除了常规的汉字“零”到“九”之外还多了“十”、“百”、“千”、“万”、“亿”，共15种汉字数字。

二、python建模

1、数据读取

        首先，读取jpg数据文件，可以看到总共有15000张图像数据。

import pandas as pd
import ospath = '/kaggle/input/chinese-mnist/data/data/'
files = os.listdir(path)
print('数据总量：', len(files))

        我们也可以打印一张图片出来看看。

import matplotlib.pyplot as plt
import matplotlib.image as mpimg# 定义图片路径
image_path = path+files[3]# 加载图片
image = mpimg.imread(image_path)# 绘制图片
plt.figure(figsize=(3, 3))
plt.imshow(image)
plt.axis('off')  # 关闭坐标轴
plt.show()

2、数据集构建

        加载必要的库以便后续使用，再定义一些超参数。

import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torchvision import transforms
from PIL import Image
from torch.utils.data import Dataset, DataLoader, random_split
from sklearn.metrics import precision_score, recall_score, f1_score# 超参数
batch_size = 64
learning_rate = 0.01
num_epochs = 5# 数据预处理
transform = transforms.Compose([transforms.ToTensor(),transforms.Normalize((0.5,), (0.5,))
])

        这里，我们看一看数据集介绍就会知道图片名称及其含义，需要从chinese_mnist.csv文件中根据图片名称中的几个数字来确定图片对应的标签。

# 获取所有图片文件的路径
all_images = [os.path.join(path, img) for img in os.listdir(path) if img.endswith('.jpg')]# 读取索引-标签对应关系csv文件，并将'suite_id', 'sample_id', 'code'设置为索引列便于查找
index_df = pd.read_csv('/kaggle/input/chinese-mnist/chinese_mnist.csv')
index_df.set_index(['suite_id', 'sample_id', 'code'], inplace=True)# 定义函数，根据各索引取值定位图片对应的数值标签value
def get_label_from_index(filename, index_df):suite_id, sample_id, code = map(int, filename.split('.')[0].split('_')[1:])return index_df.loc[(suite_id, sample_id, code), 'value']# 构建value值对应的标签序号，用于模型训练
label_dic = {0:0, 1:1, 2:2, 3:3, 4:4, 5:5, 6:6, 7:7, 8:8, 9:9, 10:10, 100:11, 1000:12, 10000:13, 100000000:14}
# 获取所有图片的标签并转化为标签序号
all_labels = [get_label_from_index(os.path.basename(img), index_df) for img in all_images]
all_labels = [label_dic[li] for li in all_labels]# 将图片路径和标签分成训练集和测试集
train_images, test_images, train_labels, test_labels = train_test_split(all_images, all_labels, test_size=0.2, random_state=2024)

        下面定义数据集类并完成数据的加载。

# 自定义数据集类
class CustomDataset(Dataset):def __init__(self, image_paths, labels, transform=None):self.image_paths = image_pathsself.labels = labelsself.transform = transformdef __len__(self):return len(self.image_paths)def __getitem__(self, idx):image = Image.open(self.image_paths[idx]).convert('L')  # 转换为灰度图像label = self.labels[idx]if self.transform:image = self.transform(image)return image, label# 创建训练集和测试集数据集
train_dataset = CustomDataset(train_images, train_labels, transform=transform)
test_dataset = CustomDataset(test_images, test_labels, transform=transform)# 创建数据加载器
train_loader = DataLoader(dataset=train_dataset, batch_size=64, shuffle=True)
test_loader = DataLoader(dataset=test_dataset, batch_size=64, shuffle=False)# 打印一些信息
print(f'训练集样本数: {len(train_dataset)}')
print(f'测试集样本数: {len(test_dataset)}')

3、模型构建

        我们构建一个包含两层卷积层和池化层的CNN并且在池化层中使用最大池化的方式。

# 定义CNN模型
class CNN(nn.Module):def __init__(self):super(CNN, self).__init__()self.conv1 = nn.Conv2d(1, 32, kernel_size=3, padding=1)self.conv2 = nn.Conv2d(32, 64, kernel_size=3, padding=1)self.pool = nn.MaxPool2d(kernel_size=2, stride=2, padding=0)self.fc1 = nn.Linear(64 * 16 * 16, 128)self.fc2 = nn.Linear(128, 15)def forward(self, x):x = self.pool(F.relu(self.conv1(x)))x = self.pool(F.relu(self.conv2(x)))x = x.view(-1, 64 * 16 * 16)x = F.relu(self.fc1(x))x = self.fc2(x)return x

4、模型实例化及训练

        下面我们对模型进行实例化并定义criterion和optimizer。

# 初始化模型、损失函数和优化器
model = CNN()
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=learning_rate, momentum=0.9)

        定义训练的代码并调用代码训练模型。

from tqdm import tqdm
# 训练模型
def train(model, train_loader, criterion, optimizer, epochs):model.train()running_loss = 0.0for epoch in range(epochs):for data, target in tqdm(train_loader):optimizer.zero_grad()output = model(data)loss = criterion(output, target)loss.backward()optimizer.step()running_loss += loss.item()print(f'Epoch [{epoch + 1}], Loss: {running_loss / len(train_loader):.4f}')running_loss = 0.0train(model, train_loader, criterion, optimizer, num_epochs)

5、测试模型

        定义模型测试代码，调用代码看指标可知我们所构建的CNN模型表现还不错。

# 测试模型
def test(model, test_loader, criterion):model.eval()test_loss = 0correct = 0all_preds = []all_targets = []with torch.no_grad():for data, target in test_loader:output = model(data)test_loss += criterion(output, target).item()pred = output.argmax(dim=1, keepdim=True)correct += pred.eq(target.view_as(pred)).sum().item()all_preds.extend(pred.cpu().numpy())all_targets.extend(target.cpu().numpy())test_loss /= len(test_loader.dataset)accuracy = 100. * correct / len(test_loader.dataset)precision = precision_score(all_targets, all_preds, average='macro')recall = recall_score(all_targets, all_preds, average='macro')f1 = f1_score(all_targets, all_preds, average='macro')print(f'Test Loss: {test_loss:.4f}, Accuracy: {accuracy:.2f}%, Precision: {precision:.4f}, Recall: {recall:.4f}, F1 Score: {f1:.4f}')test(model, test_loader, criterion)

三、完整代码

import pandas as pd
import os
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torchvision import transforms
from PIL import Image
from tqdm import tqdm
from torch.utils.data import Dataset, DataLoader, random_split
from sklearn.metrics import precision_score, recall_score, f1_scorepath = '/kaggle/input/chinese-mnist/data/data/'
files = os.listdir(path)
print('数据总量：', len(files))# 超参数
batch_size = 64
learning_rate = 0.01
num_epochs = 5# 数据预处理
transform = transforms.Compose([transforms.ToTensor(),transforms.Normalize((0.5,), (0.5,))
])# 获取所有图片文件的路径
all_images = [os.path.join(path, img) for img in os.listdir(path) if img.endswith('.jpg')]# 读取索引-标签对应关系csv文件，并将'suite_id', 'sample_id', 'code'设置为索引列便于查找
index_df = pd.read_csv('/kaggle/input/chinese-mnist/chinese_mnist.csv')
index_df.set_index(['suite_id', 'sample_id', 'code'], inplace=True)# 定义函数，根据各索引取值定位图片对应的数值标签value
def get_label_from_index(filename, index_df):suite_id, sample_id, code = map(int, filename.split('.')[0].split('_')[1:])return index_df.loc[(suite_id, sample_id, code), 'value']# 构建value值对应的标签序号，用于模型训练
label_dic = {0:0, 1:1, 2:2, 3:3, 4:4, 5:5, 6:6, 7:7, 8:8, 9:9, 10:10, 100:11, 1000:12, 10000:13, 100000000:14}# 获取所有图片的标签并转化为标签序号
all_labels = [get_label_from_index(os.path.basename(img), index_df) for img in all_images]
all_labels = [label_dic[li] for li in all_labels]# 将图片路径和标签分成训练集和测试集
train_images, test_images, train_labels, test_labels = train_test_split(all_images, all_labels, test_size=0.2, random_state=2024)# 自定义数据集类
class CustomDataset(Dataset):def __init__(self, image_paths, labels, transform=None):self.image_paths = image_pathsself.labels = labelsself.transform = transformdef __len__(self):return len(self.image_paths)def __getitem__(self, idx):image = Image.open(self.image_paths[idx]).convert('L')  # 转换为灰度图像label = self.labels[idx]if self.transform:image = self.transform(image)return image, label# 创建训练集和测试集数据集
train_dataset = CustomDataset(train_images, train_labels, transform=transform)
test_dataset = CustomDataset(test_images, test_labels, transform=transform)# 创建数据加载器
train_loader = DataLoader(dataset=train_dataset, batch_size=64, shuffle=True)
test_loader = DataLoader(dataset=test_dataset, batch_size=64, shuffle=False)# 打印信息
print(f'训练集样本数: {len(train_dataset)}')
print(f'测试集样本数: {len(test_dataset)}')# 定义CNN模型
class CNN(nn.Module):def __init__(self):super(CNN, self).__init__()self.conv1 = nn.Conv2d(1, 32, kernel_size=3, padding=1)self.conv2 = nn.Conv2d(32, 64, kernel_size=3, padding=1)self.pool = nn.MaxPool2d(kernel_size=2, stride=2, padding=0)self.fc1 = nn.Linear(64 * 16 * 16, 128)self.fc2 = nn.Linear(128, 15)def forward(self, x):x = self.pool(F.relu(self.conv1(x)))x = self.pool(F.relu(self.conv2(x)))x = x.view(-1, 64 * 16 * 16)x = F.relu(self.fc1(x))x = self.fc2(x)return x# 初始化模型、损失函数和优化器
model = CNN()
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=learning_rate, momentum=0.9)# 训练模型
def train(model, train_loader, criterion, optimizer, epochs):model.train()running_loss = 0.0for epoch in range(epochs):for data, target in tqdm(train_loader):optimizer.zero_grad()output = model(data)loss = criterion(output, target)loss.backward()optimizer.step()running_loss += loss.item()print(f'Epoch [{epoch + 1}], Loss: {running_loss / len(train_loader):.4f}')running_loss = 0.0train(model, train_loader, criterion, optimizer, num_epochs)# 测试模型
def test(model, test_loader, criterion):model.eval()test_loss = 0correct = 0all_preds = []all_targets = []with torch.no_grad():for data, target in test_loader:output = model(data)test_loss += criterion(output, target).item()pred = output.argmax(dim=1, keepdim=True)correct += pred.eq(target.view_as(pred)).sum().item()all_preds.extend(pred.cpu().numpy())all_targets.extend(target.cpu().numpy())test_loss /= len(test_loader.dataset)accuracy = 100. * correct / len(test_loader.dataset)precision = precision_score(all_targets, all_preds, average='macro')recall = recall_score(all_targets, all_preds, average='macro')f1 = f1_score(all_targets, all_preds, average='macro')print(f'Test Loss: {test_loss:.4f}, Accuracy: {accuracy:.2f}%, Precision: {precision:.4f}, Recall: {recall:.4f}, F1 Score: {f1:.4f}')test(model, test_loader, criterion)

四、总结

        本文基于汉字手写体数字图像进行了CNN分类实战，CNN作为图像处理的经典模型，展现出了它强大的图像特征提取能力，结合更加复杂的模型框架CNN还可用于高精度人脸识别、物体识别等任务中。