【tensorflow】T2.实现cifar10彩色图片分类

>- **🍨 本文为[🔗365天深度学习训练营](https://mp.weixin.qq.com/s/0dvHCaOoFnW8SCp3JpzKxg) 中的学习记录博客**
>- **🍖 原作者：[K同学啊](https://mtyjkh.blog.csdn.net/)** 

我的环境：

语言环境：python 3.11.5

编译器：Spyder

深度学习环境：Tensorflow 2.12.0

一、导入需要的包 & 加载数据

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

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

二、归一化

train_images, test_images = train_images / 255.0, test_images / 255.0

test_images.shape #看一眼形状，还可以加test_label, train的image和label

In [2]: test_images.shape
Out[2]: (10000, 32, 32, 3) 

三、pyplot 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() #spyder里删了好像也不影响

   输出图片： 

 四、构建cnn模型

model = models.Sequential([
    layers.Conv2D(32, (3, 3), activation = "relu", input_shape = (32, 32, 3)),
    layers.MaxPooling2D(2, 2),
    layers.Conv2D(64, (3, 3), activation = "relu"),
    layers.MaxPooling2D(2, 2),
    layers.Conv2D(128, (3, 3), activation = "relu"),
    
    layers.Flatten(),
    layers.Dense(128, activation = "relu"),
    layers.Dense(10)
    
    ])

    然后看一眼模型长啥样

model.summary()

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d (Conv2D)             (None, 30, 30, 32)        896       
                                                                 
 max_pooling2d (MaxPooling2D  (None, 15, 15, 32)       0         
 )                                                               
                                                                 
 conv2d_1 (Conv2D)           (None, 13, 13, 64)        18496     
                                                                 
 max_pooling2d_1 (MaxPooling  (None, 6, 6, 64)         0         
 2D)                                                             
                                                                 
 conv2d_2 (Conv2D)           (None, 4, 4, 128)         73856     
                                                                 
 flatten (Flatten)           (None, 2048)              0         
                                                                 
 dense (Dense)               (None, 128)               262272    
                                                                 
 dense_1 (Dense)             (None, 10)                1290      
                                                                 
=================================================================
Total params: 356,810
Trainable params: 356,810
Non-trainable params: 0
_________________________________________________________________

 五、编译

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

    p.s. 提升准确率的话这里其实也能改，但增加一下神经元的个数已经可以达到72%以上了，所以没改hhhh

六、训练模型

history = model.fit(
    train_images, 
    train_labels, 
    epochs = 8,
    validation_data = (test_images, test_labels)
    
    )

    p.s. 没有checkpoint和earlystopping，然后本来是10轮，在第8轮表现最好，这里epochs就用8了，没什么特定含义。

输出：

Epoch 1/8
1563/1563 [==============================] - 17s 11ms/step - loss: 1.4347 - accuracy: 0.4799 - val_loss: 1.1499 - val_accuracy: 0.5894
Epoch 2/8
1563/1563 [==============================] - 16s 10ms/step - loss: 1.0467 - accuracy: 0.6297 - val_loss: 0.9988 - val_accuracy: 0.6528
Epoch 3/8
1563/1563 [==============================] - 17s 11ms/step - loss: 0.8784 - accuracy: 0.6914 - val_loss: 0.8772 - val_accuracy: 0.6952
Epoch 4/8
1563/1563 [==============================] - 17s 11ms/step - loss: 0.7623 - accuracy: 0.7329 - val_loss: 0.8501 - val_accuracy: 0.7092
Epoch 5/8
1563/1563 [==============================] - 17s 11ms/step - loss: 0.6764 - accuracy: 0.7643 - val_loss: 0.8036 - val_accuracy: 0.7279
Epoch 6/8
1563/1563 [==============================] - 17s 11ms/step - loss: 0.6008 - accuracy: 0.7896 - val_loss: 0.8840 - val_accuracy: 0.7077
Epoch 7/8
1563/1563 [==============================] - 17s 11ms/step - loss: 0.5250 - accuracy: 0.8153 - val_loss: 0.8368 - val_accuracy: 0.7266
Epoch 8/8
1563/1563 [==============================] - 18s 11ms/step - loss: 0.4555 - accuracy: 0.8391 - val_loss: 0.9065 - val_accuracy: 0.7213

七、画个accuracy和loss

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, 1.0])
plt.legend(loc = "lower right")
plt.show()

    输出图片

八、测试模型 

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

313/313 - 1s - loss: 0.9065 - accuracy: 0.7213 - 1s/epoch - 4ms/step 

    p.s. 这里val_accuracy已经达到72%了，任务完成！

九、拿出一张图来pred一下

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

In [2]: print(class_names[np.argmax(pre[2])])
Out[2]: ship

十、总结

     彩色图片不同于灰度图片，由于RGB混合，图片会有三个维度，灰度图片只有一个维度，这是在模型的input shape里就体现出来的。然后要求里说要把准确率提升到72%，所以投机取巧了一下，只增加了第三个卷积层和全连接层的神经元就完成任务了，第二次已经了解简单cnn的程序和结构，希望之后的学习能深化理解。