第T2周：使用TensorFlow实现彩色图片分类

• 🍨 本文为🔗365天深度学习训练营 中的学习记录博客
• 🍖 原作者：K同学啊
  

电脑环境：
语言环境：Python 3.8.0
编译器：Jupyter Notebook
深度学习环境：tensorflow 2.15.0

一、前期工作

1.设置GPU（如果使用的是CPU可以忽略这步）

import tensorflow as tf
gpus = tf.config.list_physical_devices("GPU")

if gpus:
    gpu0 = gpus[0] #如果有多个GPU，仅使用第0个GPU
    tf.config.experimental.set_memory_growth(gpu0, True) #设置GPU显存用量按需使用
    tf.config.set_visible_devices([gpu0],"GPU")

2. 导入数据

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

(train_images, train_labels), (test_images, test_labels) = datasets.cifar10.load_data()

3. 归一化

# 将像素的值标准化至0到1的区间内。
train_images, test_images = train_images / 255.0, test_images / 255.0

train_images.shape,test_images.shape,train_labels.shape,test_labels.shape
'''
输出：((50000, 32, 32, 3), (10000, 32, 32, 3), (50000, 1), (10000, 1))
'''

4.数据可视化

class_names = ['airplane', 'automobile', 'bird', 'cat', 'deer','dog', 'frog', 'horse', 'ship', 'truck']

plt.figure(figsize=(20,10))
for i in range(20):
    plt.subplot(5,10,i+1)
    plt.xticks([])
    plt.yticks([])
    plt.grid(False)
    plt.imshow(train_images[i], cmap=plt.cm.binary)
    plt.xlabel(class_names[train_labels[i][0]])
plt.show()

二、构建CNN网络

1、k同学啊给出的实例代码：

model = models.Sequential([
    layers.Conv2D(32, (3, 3), activation='relu', input_shape=(32, 32, 3)), #卷积层1，卷积核3*3
    layers.MaxPooling2D((2, 2)),                   #池化层1，2*2采样
    layers.Conv2D(64, (3, 3), activation='relu'),  #卷积层2，卷积核3*3
    layers.MaxPooling2D((2, 2)),                   #池化层2，2*2采样
    layers.Conv2D(64, (3, 3), activation='relu'),  #卷积层3，卷积核3*3
    
    layers.Flatten(),                      #Flatten层，连接卷积层与全连接层
    layers.Dense(64, activation='relu'),   #全连接层，特征进一步提取
    layers.Dense(10)                       #输出层，输出预期结果
])

model.summary()  # 打印网络结构

2、自我改进代码：

更改了网络结构并加上了Drop层防止过拟合。

model = models.Sequential([
    
    layers.Conv2D(32, (3, 3), activation='relu', input_shape=(32, 32, 3)),  # 卷积层1
    layers.Conv2D(32, (3, 3), activation='relu'),  # 卷积层2
    layers.MaxPooling2D((2, 2)),  # 池化层1

    layers.Conv2D(64, (3, 3), activation='relu'),  # 卷积层3
    layers.Conv2D(64, (3, 3), activation='relu'),  # 卷积层4
    layers.MaxPooling2D((2, 2)),  # 池化层2

    layers.Dropout(0.4),  # 随机失活

    layers.Flatten(),  # 打平
    layers.Dense(64, activation='relu'),  # 全连接层
    layers.Dense(10)
])

model.summary()

三、编译模型

"""
这里设置优化器、损失函数以及metrics
这三者具体介绍可参考博客：
https://blog.csdn.net/qq_38251616/category_10258234.html
"""
model.compile(
    optimizer='adam',
	loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),# 设置损失函数为交叉熵损失函数
    metrics=['accuracy'])

四、训练模型

"""
这里设置输入训练数据集（图片及标签）、验证数据集（图片及标签）以及迭代次数epochs
关于model.fit()函数的具体介绍可参考博客：
https://blog.csdn.net/qq_38251616/category_10258234.html
"""
history = model.fit(
    # 输入训练集图片
	train_images, 
	# 输入训练集标签
	train_labels, 
	# 设置10个epoch，每一个epoch都将会把所有的数据输入模型完成一次训练。
	epochs=10, 
	# 设置验证集
    validation_data=(test_images, test_labels))

1、示例代码输出：

Epoch 1/10
1563/1563 [==============================] - 10s 6ms/step - loss: 1.5384 - accuracy: 0.4381 - val_loss: 1.3732 - val_accuracy: 0.5031
Epoch 2/10
1563/1563 [==============================] - 9s 6ms/step - loss: 1.1744 - accuracy: 0.5864 - val_loss: 1.1016 - val_accuracy: 0.6104
Epoch 3/10
1563/1563 [==============================] - 8s 5ms/step - loss: 1.0180 - accuracy: 0.6434 - val_loss: 1.0152 - val_accuracy: 0.6463
Epoch 4/10
1563/1563 [==============================] - 8s 5ms/step - loss: 0.9200 - accuracy: 0.6778 - val_loss: 0.9235 - val_accuracy: 0.6776
Epoch 5/10
1563/1563 [==============================] - 8s 5ms/step - loss: 0.8524 - accuracy: 0.7008 - val_loss: 0.9748 - val_accuracy: 0.6666
Epoch 6/10
1563/1563 [==============================] - 8s 5ms/step - loss: 0.7951 - accuracy: 0.7206 - val_loss: 0.8679 - val_accuracy: 0.6979
Epoch 7/10
1563/1563 [==============================] - 9s 6ms/step - loss: 0.7453 - accuracy: 0.7385 - val_loss: 0.8787 - val_accuracy: 0.7000
Epoch 8/10
1563/1563 [==============================] - 8s 5ms/step - loss: 0.7057 - accuracy: 0.7522 - val_loss: 0.8532 - val_accuracy: 0.7064
Epoch 9/10
1563/1563 [==============================] - 8s 5ms/step - loss: 0.6615 - accuracy: 0.7684 - val_loss: 0.9231 - val_accuracy: 0.6927
Epoch 10/10
1563/1563 [==============================] - 8s 5ms/step - loss: 0.6308 - accuracy: 0.7781 - val_loss: 0.8664 - val_accuracy: 0.7188

2、改进后的代码输出：

Epoch 1/20
1563/1563 [==============================] - 17s 7ms/step - loss: 1.5012 - accuracy: 0.4515 - val_loss: 1.2036 - val_accuracy: 0.5765
。。。。。。。。。。
Epoch 13/20
1563/1563 [==============================] - 8s 5ms/step - loss: 0.5693 - accuracy: 0.7970 - val_loss: 0.6858 - val_accuracy: 0.7739
Epoch 14/20
1563/1563 [==============================] - 9s 6ms/step - loss: 0.5473 - accuracy: 0.8076 - val_loss: 0.7089 - val_accuracy: 0.7610
Epoch 15/20
1563/1563 [==============================] - 9s 6ms/step - loss: 0.5463 - accuracy: 0.8074 - val_loss: 0.6955 - val_accuracy: 0.7621
Epoch 16/20
1563/1563 [==============================] - 8s 5ms/step - loss: 0.5192 - accuracy: 0.8160 - val_loss: 0.6945 - val_accuracy: 0.7662
Epoch 17/20
1563/1563 [==============================] - 9s 6ms/step - loss: 0.5088 - accuracy: 0.8195 - val_loss: 0.7065 - val_accuracy: 0.7668
Epoch 18/20
1563/1563 [==============================] - 9s 6ms/step - loss: 0.4958 - accuracy: 0.8226 - val_loss: 0.6850 - val_accuracy: 0.7765
Epoch 19/20
1563/1563 [==============================] - 8s 5ms/step - loss: 0.4728 - accuracy: 0.8327 - val_loss: 0.6813 - val_accuracy: 0.7774
Epoch 20/20
1563/1563 [==============================] - 9s 6ms/step - loss: 0.4707 - accuracy: 0.8300 - val_loss: 0.7123 - val_accuracy: 0.7675

五、模型评估

import matplotlib.pyplot as plt

plt.plot(history.history['accuracy'], label='accuracy')
plt.plot(history.history['val_accuracy'], label = 'val_accuracy')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.ylim([0.5, 1])
plt.legend(loc='lower right')
plt.show()

test_loss, test_acc = model.evaluate(test_images,  test_labels, verbose=2)

1、示例代码输出：

313/313 - 1s - loss: 0.8664 - accuracy: 0.7188 - 647ms/epoch - 2ms/step
print(test_acc)
'''
输出：0.7188000082969666
'''

2、改进后的代码输出：

313/313 - 1s - loss: 0.7123 - accuracy: 0.7675 - 713ms/epoch - 2ms/step
print(test_acc)
'''
输出：0.7674999833106995
'''

六、预测

plt.imshow(test_images[2])

import numpy as np

pre = model.predict(test_images)
print(class_names[np.argmax(pre[2])])

输出：

313/313 [==============================] - 1s 2ms/step
ship

七、总结

在K同学啊给的示例代码的基础上，增加了卷积层和Dropout层，epochs改为20，使训练数据特征进一步提取并且防止过拟合，实验结果显著，test_acc最高达到77%。