使用tensorflow2.7，cuda11.6版本实现在LeNet-5上增加BN层，解决手写体数字识别，并比较几种算法(全连接、LeNet-5、LeNet-5+BN)在训练集上的精度变化，画出随epoch变化的曲线的问题，并给出实现代码

好的，以下是我为您提供的实现代码：

import tensorflow as tf
from tensorflow.keras import layers, models, optimizers, datasets
import numpy as np
import matplotlib.pyplot as plt

# 加载数据集
(train_images, train_labels), (test_images, test_labels) = datasets.mnist.load_data()
train_images, test_images = train_images / 255.0, test_images / 255.0
train_images, test_images = train_images[..., np.newaxis], test_images[..., np.newaxis]

# 构建全连接神经网络模型
def build_fc():
    model = models.Sequential([
        layers.Flatten(input_shape=(28, 28, 1)),
        layers.Dense(units=128, activation='relu'),
        layers.Dense(units=10, activation='softmax')
    ])
    return model

# 构建LeNet-5模型
def build_lenet5():
    model = models.Sequential([
        layers.Conv2D(filters=6, kernel_size=(5, 5), activation='sigmoid', input_shape=(28, 28, 1)),
        layers.MaxPooling2D(pool_size=(2, 2)),
        layers.Conv2D(filters=16, kernel_size=(5, 5), activation='sigmoid'),
        layers.MaxPooling2D(pool_size=(2, 2)),
        layers.Flatten(),
        layers.Dense(units=120, activation='sigmoid'),
        layers.Dense(units=84, activation='sigmoid'),
        layers.Dense(units=10, activation='softmax')
    ])
    return model

# 构建LeNet-5+BN模型
def build_lenet5_bn():
    model = models.Sequential([
        layers.Conv2D(filters=6, kernel_size=(5, 5), input_shape=(28, 28, 1)),
        layers.BatchNormalization(),
        layers.Activation('sigmoid'),
        layers.MaxPooling2D(pool_size=(2, 2)),
        layers.Conv2D(filters=16, kernel_size=(5, 5)),
        layers.BatchNormalization(),
        layers.Activation('sigmoid'),
        layers.MaxPooling2D(pool_size=(2, 2)),
        layers.Flatten(),
        layers.Dense(units=120),
        layers.BatchNormalization(),
        layers.Activation('sigmoid'),
        layers.Dense(units=84),
        layers.BatchNormalization(),
        layers.Activation('sigmoid'),
        layers.Dense(units=10, activation='softmax')
    ])
    return model

# 训练模型并绘制曲线
def train_and_plot(model, train_images, train_labels, test_images, test_labels):
    model.compile(optimizer=optimizers.Adam(learning_rate=1e-3),
                  loss='sparse_categorical_crossentropy',
                  metrics=['accuracy'])
    history = model.fit(train_images, train_labels, epochs=10, batch_size=64,
                        validation_data=(test_images, test_labels))
    plt.plot(history.history['accuracy'])
    plt.plot(history.history['val_accuracy'])
    plt.legend(['training', 'validation'], loc='best')
    plt.title('Accuracy')
    plt.show()

# 构建并训练全连接神经网络模型
fc_model = build_fc()
train_and_plot(fc_model, train_images, train_labels, test_images, test_labels)

# 构建并训练LeNet-5模型
lenet5_model = build_lenet5()
train_and_plot(lenet5_model, train_images, train_labels, test_images, test_labels)

# 构建并训练LeNet-5+BN模型
lenet5_bn_model = build_lenet5_bn()
train_and_plot(lenet5_bn_model, train_images, train_labels, test_images, test_labels)

需要注意的是，以上代码中的全连接神经网络模型只是一个简单的baseline，如果需要进一步比较其他算法，可以在此基础上进行修改。同时，为了方便起见，此处只画了accuracy的曲线，如果需要画loss的曲线，只需要将`plt.plot(history.history['accuracy'])`改为`plt.plot(history.history['loss'])`即可。