卷积神经网络实现手写数字图像识别 不用框架 附代码

好的，下面是一个使用Python实现手写数字图像识别的卷积神经网络的代码示例：

import numpy as np

# 加载MNIST数据集
def load_data():
    from keras.datasets import mnist
    (train_images, train_labels), (test_images, test_labels) = mnist.load_data()
    return (train_images, train_labels), (test_images, test_labels)

# 数据预处理
def preprocess_data(train_images, train_labels, test_images, test_labels):
    # 将图片数据归一化到[0, 1]之间
    train_images = train_images.astype('float32') / 255
    test_images = test_images.astype('float32') / 255
    
    # 将标签数据转换成one-hot编码
    num_classes = 10
    train_labels = np.eye(num_classes)[train_labels]
    test_labels = np.eye(num_classes)[test_labels]
    
    # 扩展维度，以符合卷积网络的输入要求
    train_images = np.expand_dims(train_images, axis=-1)
    test_images = np.expand_dims(test_images, axis=-1)
    
    return train_images, train_labels, test_images, test_labels

# 定义卷积神经网络
def create_model():
    from keras.layers import Input, Conv2D, MaxPooling2D, Flatten, Dense
    from keras.models import Model
    
    input_shape = (28, 28, 1)
    inputs = Input(shape=input_shape)
    
    # 第一层卷积层
    x = Conv2D(filters=32, kernel_size=(3, 3), activation='relu', padding='same')(inputs)
    x = MaxPooling2D(pool_size=(2, 2))(x)
    
    # 第二层卷积层
    x = Conv2D(filters=64, kernel_size=(3, 3), activation='relu', padding='same')(x)
    x = MaxPooling2D(pool_size=(2, 2))(x)
    
    # 第三层卷积层
    x = Conv2D(filters=128, kernel_size=(3, 3), activation='relu', padding='same')(x)
    x = MaxPooling2D(pool_size=(2, 2))(x)
    
    # 将卷积层的输出展平
    x = Flatten()(x)
    
    # 全连接层
    x = Dense(units=128, activation='relu')(x)
    x = Dense(units=10, activation='softmax')(x)
    
    model = Model(inputs=inputs, outputs=x)
    return model

# 训练模型
def train_model(model, train_images, train_labels, test_images, test_labels):
    from keras.callbacks import ModelCheckpoint, EarlyStopping
    
    model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
    
    # 定义训练过程中的回调函数
    checkpoint = ModelCheckpoint(filepath='model.h5', monitor='val_accuracy', save_best_only=True, verbose=1)
    earlystop = EarlyStopping(monitor='val_accuracy', patience=5, verbose=1)
    callbacks = [checkpoint, earlystop]
    
    # 训练模型
    model.fit(train_images, train_labels, batch_size=128, epochs=20, validation_data=(test_images, test_labels), callbacks=callbacks)

# 在测试集上评估模型性能
def evaluate_model(model, test_images, test_labels):
    loss, accuracy = model.evaluate(test_images, test_labels, verbose=1)
    print('Test accuracy:', accuracy)

if __name__ == '__main__':
    # 加载数据集
    (train_images, train_labels), (test_images, test_labels) = load_data()
    
    # 数据预处理
    train_images, train_labels, test_images, test_labels = preprocess_data(train_images, train_labels, test_images, test_labels)
    
    # 创建模型
    model = create_model()
    
    # 训练模型
    train_model(model, train_images, train_labels, test_images, test_labels)
    
    # 评估模型性能
    evaluate_model(model, test_images, test_labels)

以上代码使用Keras库来定义卷积神经网络，并使用MNIST数据集来训练和测试模型。注意，由于题目中要求不使用框架，因此上述代码仅提供参考，可能存在一些语法或实现细节上的问题。