深度学习实践-1《手写数字识别》

深度学习实践，代码+注释记录每一步

1.准备工作

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

需要提前安装tensorflow2.7 、matplotlib

代码在和鲸社区工作台运行，传送门≯

2.导入数据集

(train_data,train_labels),(test_data,test_labels) = datasets.mnist.load_data("/home/mw/input/mnist2539/mnist.npz")

3.数据可视化

因为导入的是手写数字图片，这里将其中一张图片的灰度直方图进行展示。这里主要是练习通过matplotlib包进行数据可视化。数字表示当前图像灰度平均值，图像为灰度直方图。

avg_image=0.0
for i in range(0,28):
    for j in range(0,28):
        avg_image += train_data[1][i][j]
        #print("i:{:d},j:{:d}".format(i,j))
print(avg_image/(28*28))
#plt.imshow(train_data[1],cmap=plt.cm.binary)
#"{} {}".format("hello", "world")

plt.hist(train_data[1].ravel(),256)
plt.show()

 4.数据归一化

如图所示，所有数据都分布在0-1区间内。

train_data = train_data/255.0
test_data = test_data/255.0
avg_image=0.0

for i in range(0,28):
    for j in range(0,28):
        avg_image += train_data[1][i][j]
        #print(train_data[1][i][j])
print(avg_image/28*28)
#plt.subplot(1,2,1) 
#plt.imshow(train_data[1],cmap=plt.cm.binary)
# plt.subplot(1,2,2) 
plt.hist(train_data[1].ravel(),256)
plt.show()

 5.训练集图片展示

plt.figure(1,figsize=(30,10),facecolor='pink')中，figure用来设置画布，figsieze设置画布大小，facecolor设置画布背景。代码中设置了两个画布。

plt.figure(1,figsize=(30,10),facecolor='pink')
for i in range(30):
    plt.subplot(3,10,i+1)
    plt.xticks([])
    plt.yticks([])
    plt.grid(False)
    plt.imshow(train_data[i],cmap=plt.cm.binary)
    plt.xlabel(train_labels[i])

plt.figure(2,figsize=(30,10),facecolor='green')
for i in range(10):
    plt.subplot(3,10,i+1)
    plt.xticks([])
    plt.yticks([])
    plt.grid(False)
    plt.imshow(train_data[i],cmap=plt.cm.binary)
    plt.xlabel(train_labels[i])
    
plt.show()

 6.数据准备

为了方便模型进行训练，需要将图片的维度从3维改为4维。

print(train_data.shape,test_data.shape)
train_data=train_data.reshape((60000,28,28,1))
test_data=test_data.reshape((10000,28,28,1))
print(train_data.shape,test_data.shape)

(60000, 28, 28) (10000, 28, 28)
(60000, 28, 28, 1) (10000, 28, 28, 1)

结果如上

7.建立模型

这里我们直接使用tensorflow内自带的卷积层、池化层和全连接层进行处理。

model = models.Sequential([
    layers.Conv2D(32,(3,3),activation='relu',input_shape=(28,28,1)),
    layers.MaxPooling2D((2,2)),
    layers.Conv2D(64,(3,3),activation='relu'),
    layers.MaxPooling2D((2,2)),

    layers.Flatten(),
    layers.Dense(64,activation='relu'),
    layers.Dense(10)
])

model.summary()

 

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d (Conv2D)              (None, 26, 26, 32)        320       
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 13, 13, 32)        0         
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 11, 11, 64)        18496     
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 5, 5, 64)          0         
_________________________________________________________________
flatten (Flatten)            (None, 1600)              0         
_________________________________________________________________
dense (Dense)                (None, 64)                102464    
_________________________________________________________________
dense_1 (Dense)              (None, 10)                650       
=================================================================
Total params: 121,930
Trainable params: 121,930
Non-trainable params: 0
_________________________________________________________________

结果如上所示。

8.模型参数设置和训练

optimizer用来设置动量增加模式，损失函数我们使用了tensorflow自带的交叉熵函数。

model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits="Ture"),
              metrics=['accuracy'])
history=model.fit(train_data,train_labels,
                    epochs=20,
                    validation_data=(test_data,test_labels))

 

Epoch 1/20
1875/1875 [==============================] - 70s 37ms/step - loss: 0.1370 - accuracy: 0.9580 - val_loss: 0.0446 - val_accuracy: 0.9853
Epoch 2/20
1875/1875 [==============================] - 67s 36ms/step - loss: 0.0060 - accuracy: 0.9981 - val_loss: 0.0463 - val_accuracy: 0.9891
Epoch 12/20
1875/1875 [==============================] - 67s 36ms/step - loss: 0.0053 - accuracy: 0.9981 - val_loss: 0.0356 - val_accuracy: 0.9913
Epoch 13/20
1875/1875 [==============================] - 67s 36ms/step - loss: 0.0049 - accuracy: 0.9983 - val_loss: 0.0320 - val_accuracy: 0.9914
Epoch 14/20
1875/1875 [==============================] - 67s 36ms/step - loss: 0.0038 - accuracy: 0.9988 - val_loss: 0.0434 - val_accuracy: 0.9913
Epoch 15/20
1875/1875 [==============================] - 67s 36ms/step - loss: 0.0046 - accuracy: 0.9986 - val_loss: 0.0333 - val_accuracy: 0.9919
Epoch 16/20
1875/1875 [==============================] - 67s 36ms/step - loss: 0.0034 - accuracy: 0.9988 - val_loss: 0.0478 - val_accuracy: 0.9903
Epoch 17/20
1875/1875 [==============================] - 67s 36ms/step - loss: 0.0036 - accuracy: 0.9989 - val_loss: 0.0454 - val_accuracy: 0.9911
Epoch 18/20
1875/1875 [==============================] - 66s 35ms/step - loss: 0.0044 - accuracy: 0.9984 - val_loss: 0.0537 - val_accuracy: 0.9901
Epoch 19/20
1875/1875 [==============================] - 66s 35ms/step - loss: 0.0032 - accuracy: 0.9990 - val_loss: 0.0463 - val_accuracy: 0.9913
Epoch 20/20
1875/1875 [==============================] - 66s 35ms/step - loss: 0.0037 - accuracy: 0.9990 - val_loss: 0.0459 - val_accuracy: 0.9917

9.模型保存和加载

model.save(r'/home/mw/project/model.h5')
model = tf.keras.models.load_model(r'/home/mw/project/model.h5')

 10.读取模型进行预测

test_data_pl = test_data.reshape((10000,28,28))
pre=model.predict(test_data)
test_lab = []
for i in range(0,10):
    test_lab.append(np.argmax(pre[i]))
print(test_lab)

plt.figure(figsize=(30,10))
for i in range(10):
    plt.subplot(5,10,i+1)
    plt.xticks([])
    plt.yticks([])
    plt.imshow(test_data_pl[i],cmap=plt.cm.binary)
    plt.xlabel(test_lab[i])
plt.show()

结果如图所示