tensorflow resnet50 复现网络结构 不使用预训练模型 二分类 训练自己的数据集 添加防止过拟合操作（例如冻结全连接层等）保存训练结果 预测 代码

好的，以下是基于 TensorFlow 复现 ResNet50 网络的代码。代码中包含自己的数据集的二分类训练，以及防止过拟合的操作（冻结全连接层）。

import tensorflow as tf
from tensorflow.keras import layers, models, optimizers
from tensorflow.keras.preprocessing.image import ImageDataGenerator

# 定义 ResNet50 网络结构
def resnet50_model(input_shape=(224, 224, 3), num_classes=2):
    input_tensor = layers.Input(shape=input_shape)
    # Stage 1
    x = layers.Conv2D(filters=64, kernel_size=(7, 7), strides=(2, 2), padding='same', name='conv1')(input_tensor)
    x = layers.BatchNormalization(name='bn_conv1')(x)
    x = layers.Activation('relu')(x)
    x = layers.MaxPool2D(pool_size=(3, 3), strides=(2, 2), padding='same')(x)
    # Stage 2
    x = conv_block(x, filters=[64, 64, 256], stage=2, block='a', strides=(1, 1))
    x = identity_block(x, filters=[64, 64, 256], stage=2, block='b')
    x = identity_block(x, filters=[64, 64, 256], stage=2, block='c')
    # Stage 3
    x = conv_block(x, filters=[128, 128, 512], stage=3, block='a')
    x = identity_block(x, filters=[128, 128, 512], stage=3, block='b')
    x = identity_block(x, filters=[128, 128, 512], stage=3, block='c')
    x = identity_block(x, filters=[128, 128, 512], stage=3, block='d')
    # Stage 4
    x = conv_block(x, filters=[256, 256, 1024], stage=4, block='a')
    x = identity_block(x, filters=[256, 256, 1024], stage=4, block='b')
    x = identity_block(x, filters=[256, 256, 1024], stage=4, block='c')
    x = identity_block(x, filters=[256, 256, 1024], stage=4, block='d')
    x = identity_block(x, filters=[256, 256, 1024], stage=4, block='e')
    x = identity_block(x, filters=[256, 256, 1024], stage=4, block='f')
    # Stage 5
    x = conv_block(x, filters=[512, 512, 2048], stage=5, block='a')
    x = identity_block(x, filters=[512, 512, 2048], stage=5, block='b')
    x = identity_block(x, filters=[512, 512, 2048], stage=5, block='c')
    x = layers.GlobalAveragePooling2D()(x)
    x = layers.Dense(units=num_classes, activation='softmax', name='fc' + str(num_classes))(x)
    # Create model.
    model = models.Model(inputs=input_tensor, outputs=x, name='resnet50')
    return model

# 定义 identity block
def identity_block(input_tensor, filters, stage, block):
    filters1, filters2, filters3 = filters
    block_name = 'res' + str(stage) + block + '_branch'
    x = layers.Conv2D(filters=filters1, kernel_size=(1, 1), name=block_name + '2a')(input_tensor)
    x = layers.BatchNormalization(name=block_name + '2b')(x)
    x = layers.Activation('relu')(x)
    x = layers.Conv2D(filters=filters2, kernel_size=(3, 3), padding='same', name=block_name + '2c')(x)
    x = layers.BatchNormalization(name=block_name + '2d')(x)
    x = layers.Activation('relu')(x)
    x = layers.Conv2D(filters=filters3, kernel_size=(1, 1), name=block_name + '2e')(x)
    x = layers.BatchNormalization(name=block_name + '2f')(x)
    x = layers.add([input_tensor, x])
    x = layers.Activation('relu')(x)
    return x

# 定义 conv block
def conv_block(input_tensor, filters, stage, block, strides=(2, 2)):
    filters1, filters2, filters3 = filters
    block_name = 'res' + str(stage) + block + '_branch'
    x = layers.Conv2D(filters=filters1, kernel_size=(1, 1), strides=strides, name=block_name + '2a')(input_tensor)
    x = layers.BatchNormalization(name=block_name + '2b')(x)
    x = layers.Activation('relu')(x)
    x = layers.Conv2D(filters=filters2, kernel_size=(3, 3), padding='same', name=block_name + '2c')(x)
    x = layers.BatchNormalization(name=block_name + '2d')(x)
    x = layers.Activation('relu')(x)
    x = layers.Conv2D(filters=filters3, kernel_size=(1, 1), name=block_name + '2e')(x)
    x = layers.BatchNormalization(name=block_name + '2f')(x)
    shortcut = layers.Conv2D(filters=filters3, kernel_size=(1, 1), strides=strides, name=block_name + '1')(input_tensor)
    shortcut = layers.BatchNormalization(name=block_name + '2')(shortcut)
    x = layers.add([shortcut, x])
    x = layers.Activation('relu')(x)
    return x

# 加载数据
train_datagen = ImageDataGenerator(
        rescale=1./255,
        shear_range=0.2,
        zoom_range=0.2,
        horizontal_flip=True)

test_datagen = ImageDataGenerator(rescale=1./255)

train_generator = train_datagen.flow_from_directory(
        'train',
        target_size=(224, 224),
        batch_size=32,
        class_mode='categorical')

validation_generator = test_datagen.flow_from_directory(
        'test',
        target_size=(224, 224),
        batch_size=32,
        class_mode='categorical')

# 创建 ResNet50 模型
model = resnet50_model(input_shape=(224, 224, 3), num_classes=2)

# 冻结全连接层
for layer in model.layers[:-1]:
    layer.trainable = False

# 编译模型
model.compile(optimizer=optimizers.Adam(lr=0.001), loss='categorical_crossentropy', metrics=['accuracy'])

# 训练模型
history = model.fit_generator(
        train_generator,
        steps_per_epoch=train_generator.samples // train_generator.batch_size,
        epochs=10,
        validation_data=validation_generator,
        validation_steps=validation_generator.samples // validation_generator.batch_size)

# 保存模型
model.save('resnet50_model.h5')

# 预测
from tensorflow.keras.preprocessing import image
import numpy as np

img_path = 'example.jpg'
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = x / 255.0
preds = model.predict(x)
print('Predicted:', preds[0])

使用该代码，你可以复现 ResNet50 网络结构，并对自己的数据集进行二分类训练。此外，你还可以添加防止过拟合的操作，如冻结全连接层，以提高模型的泛化能力。最后，你还可以保存训练结果和进行预测。