ResNet神经网络训练

ResNet训练

导包

import tensorflow as tf

from tensorflow.keras import layers, models, datasets, optimizers

加载Fashion-MNIST数据集

(train_images, train_labels), (test_images, test_labels) = datasets.fashion_mnist.load_data()

归一化像素值到[0, 1]区间

train_images = train_images.astype("float32") / 255

test_images = test_images.astype("float32") / 255

由于Fashion-MNIST的图像是灰度图像，需要增加一个颜色通道

train_images = train_images.reshape(-1, 28, 28, 1)

test_images = test_images.reshape(-1, 28, 28, 1)

对标签进行分类编码

train_labels = tf.keras.utils.to_categorical(train_labels, 10)

test_labels = tf.keras.utils.to_categorical(test_labels, 10)

定义一个简单的卷积层来增加图像的尺寸，使其适合ResNet50的输入

preprocessing = models.Sequential([

 layers.Conv2D(3, (1, 1), activation='relu', input_shape=(28, 28, 1)),

 layers.UpSampling2D((2, 2)),  # 将图像尺寸增加到56x56

])

应用预处理

train_images = preprocessing(train_images)

test_images = preprocessing(test_images)

加载ResNet50模型并冻结所有层

base_model = tf.keras.applications.ResNet50(weights='imagenet', include_top=False, input_shape=(56, 56, 3))

base_model.trainable = False

添加自定义分类器

model = models.Sequential([

    base_model,

    layers.GlobalAveragePooling2D(),

    layers.Dense(10, activation='softmax')

])

编译模型

model.compile(optimizer=optimizers.Adam(),

              loss='categorical_crossentropy',

              metrics=['accuracy'])

训练模型

model.fit(train_images, train_labels, epochs=10, batch_size=64,validation_data=(test_images, test_labels))

评估模型

test_loss, test_acc = model.evaluate(test_images, test_labels)

print(f'Test Accuracy: {test_acc:.4f}')

训练模型保存

save_path = r'D:\\图像处理、深度学习\\训练保存\\ResNet.h5'

model.save(save_path)