一、前言

我的環境：

• 語言環境：Python3.6.5
• 編譯器：jupyter notebook
• 深度學習環境：TensorFlow2.4.1

往期精彩內容：

• 卷積神經網絡（CNN）實現mnist手寫數字識別
• 卷積神經網絡（CNN）多種圖片分類的實現
• 卷積神經網絡（CNN）衣服圖像分類的實現
• 卷積神經網絡（CNN）鮮花識別
• 卷積神經網絡（CNN）天氣識別
• 卷積神經網絡（VGG-16）識別海賊王草帽一伙
• 卷積神經網絡（ResNet-50）鳥類識別

來自專欄：機器學習與深度學習算法推薦

二、前期工作

1. 設置GPU（如果使用的是CPU可以忽略這步）

import tensorflow as tfgpus = tf.config.list_physical_devices("GPU")if gpus:tf.config.experimental.set_memory_growth(gpus[0], True)  #設置GPU顯存用量按需使用tf.config.set_visible_devices([gpus[0]],"GPU")

2. 導入數據

import matplotlib.pyplot as plt
# 支持中文
plt.rcParams['font.sans-serif'] = ['SimHei']  # 用來正常顯示中文標簽
plt.rcParams['axes.unicode_minus'] = False  # 用來正常顯示負號import os,PIL# 設置隨機種子盡可能使結果可以重現
import numpy as np
np.random.seed(1)# 設置隨機種子盡可能使結果可以重現
import tensorflow as tf
tf.random.set_seed(1)import pathlib

data_dir = "bird_photos"data_dir = pathlib.Path(data_dir)

3. 查看數據

image_count = len(list(data_dir.glob('*/*')))
print("圖片總數為：",image_count)

圖片總數為： 565

二、數據預處理

文件夾	數量
Bananaquit	166 張
Black Throated Bushtiti	111 張
Black skimmer	122 張
Cockatoo	166張

1. 加載數據

使用image_dataset_from_directory方法將磁盤中的數據加載到tf.data.Dataset中

batch_size = 8
img_height = 227
img_width = 227

train_ds = tf.keras.preprocessing.image_dataset_from_directory(data_dir,validation_split=0.2,subset="training",seed=123,image_size=(img_height, img_width),batch_size=batch_size)

Found 565 files belonging to 4 classes.
Using 452 files for training.

val_ds = tf.keras.preprocessing.image_dataset_from_directory(data_dir,validation_split=0.2,subset="validation",seed=123,image_size=(img_height, img_width),batch_size=batch_size)

Found 565 files belonging to 4 classes.
Using 113 files for validation.

我們可以通過class_names輸出數據集的標簽。標簽將按字母順序對應于目錄名稱。

class_names = train_ds.class_names
print(class_names)

['Bananaquit', 'Black Skimmer', 'Black Throated Bushtiti', 'Cockatoo']

2. 可視化數據

plt.figure(figsize=(10, 5))  # 圖形的寬為10高為5for images, labels in train_ds.take(1):for i in range(8):ax = plt.subplot(2, 4, i + 1)  plt.imshow(images[i].numpy().astype("uint8"))plt.title(class_names[labels[i]])plt.axis("off")

plt.imshow(images[1].numpy().astype("uint8"))

3. 再次檢查數據

for image_batch, labels_batch in train_ds:print(image_batch.shape)print(labels_batch.shape)break

(8, 227, 227, 3)
(8,)

• Image_batch是形狀的張量（8, 224, 224, 3)。這是一批形狀240x240x3的8張圖片（最后一維指的是彩色通道RGB）。
• Label_batch是形狀（8，）的張量，這些標簽對應8張圖片

4. 配置數據集

AUTOTUNE = tf.data.AUTOTUNEtrain_ds = train_ds.cache().shuffle(1000).prefetch(buffer_size=AUTOTUNE)
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)

三、AlexNet (8層）介紹

AleXNet使用了ReLU方法加快訓練速度，并且使用Dropout來防止過擬合

AleXNet (8層）是首次把卷積神經網絡引入計算機視覺領域并取得突破性成績的模型。獲得了ILSVRC 2012年的冠軍，再top-5項目中錯誤率僅僅15.3%，相對于使用傳統方法的亞軍26.2%的成績優良重大突破。和之前的LeNet相比，AlexNet通過堆疊卷積層使得模型更深更寬。

四、構建AlexNet (8層）網絡模型

from tensorflow.keras import layers, models, Input
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Dense, Flatten, Dropout,BatchNormalization,Activationimport numpy as np
seed = 7
np.random.seed(seed)def AlexNet(nb_classes, input_shape):input_tensor = Input(shape=input_shape)# 1st blockx = Conv2D(96, (11,11), strides=4, name='block1_conv1')(input_tensor)x = BatchNormalization()(x)x = Activation('relu')(x)x = MaxPooling2D((3,3), strides=2, name = 'block1_pool')(x)# 2nd blockx = Conv2D(256, (5,5), padding='same', name='block2_conv1')(x)x = BatchNormalization()(x)x = Activation('relu')(x)x = MaxPooling2D((3,3), strides=2, name='block2_pool')(x)# 3rd blockx = Conv2D(384, (3,3), activation='relu', padding='same',name='block3_conv1')(x)# 4th blockx = Conv2D(384, (3,3), activation='relu', padding='same',name='block4_conv1')(x)# 5th blockx = Conv2D(256, (3,3), activation='relu', padding='same',name='block5_conv1')(x)x = MaxPooling2D((3,3), strides=2, name = 'block5_pool')(x)# full connectionx = Flatten()(x)x = Dense(4096, activation='relu',  name='fc1')(x)x = Dropout(0.5)(x)x = Dense(4096, activation='relu', name='fc2')(x)x = Dropout(0.5)(x)output_tensor = Dense(nb_classes, activation='softmax', name='predictions')(x)model = Model(input_tensor, output_tensor)return modelmodel=AlexNet(1000, (img_width, img_height, 3))
model.summary()

Model: "model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         [(None, 227, 227, 3)]     0         
_________________________________________________________________
block1_conv1 (Conv2D)        (None, 55, 55, 96)        34944     
_________________________________________________________________
batch_normalization (BatchNo (None, 55, 55, 96)        384       
_________________________________________________________________
activation (Activation)      (None, 55, 55, 96)        0         
_________________________________________________________________
block1_pool (MaxPooling2D)   (None, 27, 27, 96)        0         
_________________________________________________________________
block2_conv1 (Conv2D)        (None, 27, 27, 256)       614656    
_________________________________________________________________
batch_normalization_1 (Batch (None, 27, 27, 256)       1024      
_________________________________________________________________
activation_1 (Activation)    (None, 27, 27, 256)       0         
_________________________________________________________________
block2_pool (MaxPooling2D)   (None, 13, 13, 256)       0         
_________________________________________________________________
block3_conv1 (Conv2D)        (None, 13, 13, 384)       885120    
_________________________________________________________________
block4_conv1 (Conv2D)        (None, 13, 13, 384)       1327488   
_________________________________________________________________
block5_conv1 (Conv2D)        (None, 13, 13, 256)       884992    
_________________________________________________________________
block5_pool (MaxPooling2D)   (None, 6, 6, 256)         0         
_________________________________________________________________
flatten (Flatten)            (None, 9216)              0         
_________________________________________________________________
fc1 (Dense)                  (None, 4096)              37752832  
_________________________________________________________________
dropout (Dropout)            (None, 4096)              0         
_________________________________________________________________
fc2 (Dense)                  (None, 4096)              16781312  
_________________________________________________________________
dropout_1 (Dropout)          (None, 4096)              0         
_________________________________________________________________
predictions (Dense)          (None, 1000)              4097000   
=================================================================
Total params: 62,379,752
Trainable params: 62,379,048
Non-trainable params: 704
_________________________________________________________________

五、編譯

在準備對模型進行訓練之前，還需要再對其進行一些設置。以下內容是在模型的編譯步驟中添加的：

• 損失函數（loss）：用于衡量模型在訓練期間的準確率。
• 優化器（optimizer）：決定模型如何根據其看到的數據和自身的損失函數進行更新。
• 指標（metrics）：用于監控訓練和測試步驟。以下示例使用了準確率，即被正確分類的圖像的比率。

# 設置優化器，我這里改變了學習率。
# opt = tf.keras.optimizers.Adam(learning_rate=1e-7)model.compile(optimizer="adam",loss='sparse_categorical_crossentropy',metrics=['accuracy'])

六、訓練模型

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

Epoch 1/20
57/57 [==============================] - 5s 30ms/step - loss: 9.2789 - accuracy: 0.2166 - val_loss: 3.2340 - val_accuracy: 0.3363
Epoch 2/20
57/57 [==============================] - 1s 14ms/step - loss: 0.9329 - accuracy: 0.6224 - val_loss: 1.1778 - val_accuracy: 0.5310
Epoch 3/20
57/57 [==============================] - 1s 14ms/step - loss: 0.7438 - accuracy: 0.6747 - val_loss: 1.9651 - val_accuracy: 0.5133
Epoch 4/20
57/57 [==============================] - 1s 14ms/step - loss: 0.8875 - accuracy: 0.7025 - val_loss: 1.5589 - val_accuracy: 0.4602
Epoch 5/20
57/57 [==============================] - 1s 14ms/step - loss: 0.6116 - accuracy: 0.7424 - val_loss: 0.9914 - val_accuracy: 0.4956
Epoch 6/20
57/57 [==============================] - 1s 15ms/step - loss: 0.6258 - accuracy: 0.7520 - val_loss: 1.1103 - val_accuracy: 0.5221
Epoch 7/20
57/57 [==============================] - 1s 13ms/step - loss: 0.5138 - accuracy: 0.8034 - val_loss: 0.7832 - val_accuracy: 0.6726
Epoch 8/20
57/57 [==============================] - 1s 14ms/step - loss: 0.5343 - accuracy: 0.7940 - val_loss: 6.1064 - val_accuracy: 0.4602
Epoch 9/20
57/57 [==============================] - 1s 14ms/step - loss: 0.8667 - accuracy: 0.7606 - val_loss: 0.6869 - val_accuracy: 0.7965
Epoch 10/20
57/57 [==============================] - 1s 16ms/step - loss: 0.5785 - accuracy: 0.8141 - val_loss: 1.3631 - val_accuracy: 0.5310
Epoch 11/20
57/57 [==============================] - 1s 15ms/step - loss: 0.4929 - accuracy: 0.8109 - val_loss: 0.7191 - val_accuracy: 0.7345
Epoch 12/20
57/57 [==============================] - 1s 15ms/step - loss: 0.4141 - accuracy: 0.8507 - val_loss: 0.4962 - val_accuracy: 0.8496
Epoch 13/20
57/57 [==============================] - 1s 15ms/step - loss: 0.2591 - accuracy: 0.9148 - val_loss: 0.8015 - val_accuracy: 0.8053
Epoch 14/20
57/57 [==============================] - 1s 15ms/step - loss: 0.2683 - accuracy: 0.9079 - val_loss: 0.5451 - val_accuracy: 0.8142
Epoch 15/20
57/57 [==============================] - 1s 14ms/step - loss: 0.2925 - accuracy: 0.9096 - val_loss: 0.6668 - val_accuracy: 0.8584
Epoch 16/20
57/57 [==============================] - 1s 14ms/step - loss: 0.4009 - accuracy: 0.8804 - val_loss: 1.1609 - val_accuracy: 0.6372
Epoch 17/20
57/57 [==============================] - 1s 14ms/step - loss: 0.4375 - accuracy: 0.8446 - val_loss: 0.9854 - val_accuracy: 0.7965
Epoch 18/20
57/57 [==============================] - 1s 14ms/step - loss: 0.3085 - accuracy: 0.8926 - val_loss: 0.6477 - val_accuracy: 0.8761
Epoch 19/20
57/57 [==============================] - 1s 15ms/step - loss: 0.1200 - accuracy: 0.9538 - val_loss: 1.8996 - val_accuracy: 0.5398
Epoch 20/20
57/57 [==============================] - 1s 15ms/step - loss: 0.3378 - accuracy: 0.9095 - val_loss: 0.9337 - val_accuracy: 0.8053

七、模型評估

acc = history.history['accuracy']
val_acc = history.history['val_accuracy']loss = history.history['loss']
val_loss = history.history['val_loss']epochs_range = range(epochs)plt.figure(figsize=(12, 4))
plt.subplot(1, 2, 1)plt.plot(epochs_range, acc, label='Training Accuracy')
plt.plot(epochs_range, val_acc, label='Validation Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')plt.subplot(1, 2, 2)
plt.plot(epochs_range, loss, label='Training Loss')
plt.plot(epochs_range, val_loss, label='Validation Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()

八、保存and加載模型

 保存模型
model.save('model/my_model.h5')
# 加載模型
new_model = tf.keras.models.load_model('model/my_model.h5')

九、預測

# 采用加載的模型（new_model）來看預測結果plt.figure(figsize=(10, 5))  # 圖形的寬為10高為5for images, labels in val_ds.take(1):for i in range(8):ax = plt.subplot(2, 4, i + 1)  # 顯示圖片plt.imshow(images[i].numpy().astype("uint8"))# 需要給圖片增加一個維度img_array = tf.expand_dims(images[i], 0) # 使用模型預測圖片中的人物predictions = new_model.predict(img_array)plt.title(class_names[np.argmax(predictions)])plt.axis("off")