数据增强方法

最新推荐文章于 2024-10-16 17:30:30 发布

小叮当爱咖啡

最新推荐文章于 2024-10-16 17:30:30 发布

阅读量788

点赞数 22

文章标签：深度学习人工智能

本文链接：https://blog.csdn.net/2301_76606951/article/details/142219045

版权

本文为为🔗365天深度学习训练营内部文章

原作者：K同学啊

当原数据量很小的时候，进行数据增强，并通过数据增强用少量数据达到非常非常棒的识别准确率

一加载数据

import matplotlib.pyplot as plt
import numpy as np
#隐藏警告
import warnings
warnings.filterwarnings('ignore')

from tensorflow.keras import layers
import tensorflow as tf

data_dir   = "./data/"
img_height = 224
img_width  = 224
batch_size = 32

train_ds = tf.keras.preprocessing.image_dataset_from_directory(
    data_dir,
    validation_split=0.3,
    subset="training",
    seed=12,
    image_size=(img_height, img_width),
    batch_size=batch_size)

Found 600 files belonging to 2 classes.
Using 420 files for training.

val_ds = tf.keras.preprocessing.image_dataset_from_directory(
    data_dir,
    validation_split=0.3,
    subset="validation",
    seed=12,
    image_size=(img_height, img_width),
    batch_size=batch_size)

Found 600 files belonging to 2 classes.
Using 180 files for validation.

由于原始数据集不包含测试集，因此需要创建一个。使用 tf.data.experimental.cardinality 确定验证集中有多少批次的数据，然后将其中的 20% 移至测试集。

val_batches = tf.data.experimental.cardinality(val_ds)
test_ds     = val_ds.take(val_batches // 5)
val_ds      = val_ds.skip(val_batches // 5)

print('Number of validation batches: %d' % tf.data.experimental.cardinality(val_ds))
print('Number of test batches: %d' % tf.data.experimental.cardinality(test_ds))

Number of validation batches: 12
Number of test batches: 2

一共有猫、狗两类

二数据预处理

这里展示数据的归一化

AUTOTUNE = tf.data.AUTOTUNE

def preprocess_image(image,label):
    return (image/255.0,label)

# 归一化处理
train_ds = train_ds.map(preprocess_image, num_parallel_calls=AUTOTUNE)
val_ds   = val_ds.map(preprocess_image, num_parallel_calls=AUTOTUNE)
test_ds  = test_ds.map(preprocess_image, num_parallel_calls=AUTOTUNE)

train_ds = train_ds.cache().prefetch(buffer_size=AUTOTUNE)
val_ds   = val_ds.cache().prefetch(buffer_size=AUTOTUNE)

plt.figure(figsize=(15, 10))  # 图形的宽为15高为10

for images, labels in train_ds.take(1):
    for i in range(8):
        
        ax = plt.subplot(5, 8, i + 1) 
        plt.imshow(images[i])
        plt.title(class_names[labels[i]])
        
        plt.axis("off")

三、数据增强

我们可以使用 tf.keras.layers.experimental.preprocessing.RandomFlip 与 tf.keras.layers.experimental.preprocessing.RandomRotation 进行数据增强

tf.keras.layers.experimental.preprocessing.RandomFlip：水平和垂直随机翻转每个图像。
tf.keras.layers.experimental.preprocessing.RandomRotation：随机旋转每个图像

data_augmentation = tf.keras.Sequential([
  tf.keras.layers.experimental.preprocessing.RandomFlip("horizontal_and_vertical"),
  tf.keras.layers.experimental.preprocessing.RandomRotation(0.2),
])

第一个层表示进行随机的水平和垂直翻转，而第二个层表示按照 0.2 的弧度值进行随机旋转

# Add the image to a batch.
image = tf.expand_dims(images[i], 0)

将图片添加到批次当中

plt.figure(figsize=(8, 8))
for i in range(9):
    augmented_image = data_augmentation(image)
    ax = plt.subplot(3, 3, i + 1)
    plt.imshow(augmented_image[0])
    plt.axis("off")

四、增强方式

方法一：将其嵌入model中

model = tf.keras.Sequential([
  data_augmentation,
  layers.Conv2D(16, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
])

这样做的好处是：

数据增强这块的工作可以得到GPU的加速（如果你使用了GPU训练的话）

注意：只有在模型训练时（Model.fit）才会进行增强，在模型评估(Model.evaluate)以及预测(Model.predict)时并不会进行增强操作。

方法二：在Dataset数据集中进行数据增强

batch_size = 32
AUTOTUNE = tf.data.AUTOTUNE

def prepare(ds):
    ds = ds.map(lambda x, y: (data_augmentation(x, training=True), y), num_parallel_calls=AUTOTUNE)
    return ds

train_ds = prepare(train_ds)

五训练模型

model = tf.keras.Sequential([
  layers.Conv2D(16, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
  layers.Conv2D(32, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
  layers.Conv2D(64, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
  layers.Flatten(),
  layers.Dense(128, activation='relu'),
  layers.Dense(len(class_names))
])

在准备对模型进行训练之前，还需要再对其进行一些设置。以下内容是在模型的编译步骤中添加的：

损失函数（loss）：用于衡量模型在训练期间的准确率。
优化器（optimizer）：决定模型如何根据其看到的数据和自身的损失函数进行更新。
评价函数（metrics）：用于监控训练和测试步骤。以下示例使用了准确率，即被正确分类的图像的比率。

model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])

epochs=20
history = model.fit(
  train_ds,
  validation_data=val_ds,
  epochs=epochs
)

Epoch 1/20
14/14 [==============================] - 9s 522ms/step - loss: 1.2811 - accuracy: 0.4810 - val_loss: 0.6834 - val_accuracy: 0.5473
Epoch 2/20
14/14 [==============================] - 7s 470ms/step - loss: 0.6916 - accuracy: 0.5167 - val_loss: 0.6863 - val_accuracy: 0.5541
Epoch 3/20
14/14 [==============================] - 7s 449ms/step - loss: 0.6725 - accuracy: 0.5976 - val_loss: 0.6804 - val_accuracy: 0.5203
Epoch 4/20
14/14 [==============================] - 7s 442ms/step - loss: 0.6406 - accuracy: 0.6810 - val_loss: 0.6071 - val_accuracy: 0.7770
Epoch 5/20
14/14 [==============================] - 7s 456ms/step - loss: 0.5705 - accuracy: 0.7405 - val_loss: 0.5379 - val_accuracy: 0.7568
Epoch 6/20
14/14 [==============================] - 7s 472ms/step - loss: 0.4925 - accuracy: 0.7857 - val_loss: 0.5400 - val_accuracy: 0.7432
Epoch 7/20
14/14 [==============================] - 7s 480ms/step - loss: 0.4367 - accuracy: 0.8214 - val_loss: 0.5519 - val_accuracy: 0.7297
Epoch 8/20
14/14 [==============================] - 7s 471ms/step - loss: 0.4035 - accuracy: 0.8214 - val_loss: 0.5957 - val_accuracy: 0.7230
Epoch 9/20
14/14 [==============================] - 7s 457ms/step - loss: 0.3758 - accuracy: 0.8333 - val_loss: 0.4349 - val_accuracy: 0.7905
Epoch 10/20
14/14 [==============================] - 7s 452ms/step - loss: 0.2980 - accuracy: 0.8881 - val_loss: 0.4214 - val_accuracy: 0.8311
Epoch 11/20
14/14 [==============================] - 7s 477ms/step - loss: 0.3206 - accuracy: 0.8643 - val_loss: 0.3316 - val_accuracy: 0.8514
Epoch 12/20
14/14 [==============================] - 7s 471ms/step - loss: 0.2666 - accuracy: 0.8929 - val_loss: 0.3477 - val_accuracy: 0.8649
Epoch 13/20
14/14 [==============================] - 7s 452ms/step - loss: 0.2402 - accuracy: 0.9238 - val_loss: 0.3949 - val_accuracy: 0.8716
Epoch 14/20
14/14 [==============================] - 8s 503ms/step - loss: 0.2027 - accuracy: 0.9333 - val_loss: 0.2815 - val_accuracy: 0.8919
Epoch 15/20
14/14 [==============================] - 7s 454ms/step - loss: 0.2168 - accuracy: 0.9381 - val_loss: 0.2824 - val_accuracy: 0.9122
Epoch 16/20
14/14 [==============================] - 7s 443ms/step - loss: 0.2097 - accuracy: 0.9071 - val_loss: 0.3573 - val_accuracy: 0.8784
Epoch 17/20
14/14 [==============================] - 7s 448ms/step - loss: 0.2030 - accuracy: 0.9071 - val_loss: 0.3212 - val_accuracy: 0.8784
Epoch 18/20
14/14 [==============================] - 7s 444ms/step - loss: 0.1749 - accuracy: 0.9310 - val_loss: 0.2591 - val_accuracy: 0.9257
Epoch 19/20
14/14 [==============================] - 7s 445ms/step - loss: 0.1435 - accuracy: 0.9405 - val_loss: 0.2557 - val_accuracy: 0.8986
Epoch 20/20
14/14 [==============================] - 7s 446ms/step - loss: 0.1288 - accuracy: 0.9500 - val_loss: 0.2351 - val_accuracy: 0.9257

查看测试集准确率

loss, acc = model.evaluate(test_ds)
print("Accuracy", acc)

1/1 [==============================] - 0s 170ms/step - loss: 0.3631 - accuracy: 0.8125
Accuracy 0.8125

六自定义增强函数

import random
# 这是大家可以自由发挥的一个地方
def aug_img(image):
    seed = (random.randint(0,9), 0)
    # 随机改变图像对比度
    stateless_random_brightness = tf.image.stateless_random_contrast(image, lower=0.1, upper=1.0, seed=seed)
    return stateless_random_brightness

image = tf.expand_dims(images[3]*255, 0)
print("Min and max pixel values:", image.numpy().min(), image.numpy().max())

Min and max pixel values: 14.000048 253.28577

plt.figure(figsize=(8, 8))
for i in range(9):
    augmented_image = aug_img(image)
    ax = plt.subplot(3, 3, i + 1)
    plt.imshow(augmented_image[0].numpy().astype("uint8"))

    plt.axis("off")

如何将自定义增强函数应用到我们数据上呢

在数据预处理的步骤中嵌套数据增强的函数即可

def preprocess_image(image, label):
    # 归一化图像
    normalized_image = image / 255.0
    # 应用增强函数
    augmented_image = aug_img(normalized_image)
    return (augmented_image, label)

def aug_img(image):
    seed = (random.randint(0, 9), 0)
    # 随机改变图像对比度
    stateless_random_brightness = tf.image.stateless_random_contrast(image, lower=0.1, upper=1.0, seed=seed)
    return stateless_random_brightness