深度学习笔记15_TensorFlow实现运动鞋品牌识别

L10732034822024-09-07 7:02

一、我的环境

1.语言环境：Python 3.9

2.编译器：Pycharm

3.深度学习环境：TensorFlow 2.10.0

二、GPU设置

若使用的是cpu则可忽略

复制代码 
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")

三**、导入数据**

复制代码 
data_dir = "./data/"
data_dir = pathlib.Path(data_dir)

image_count = len(list(data_dir.glob('*/*/*.jpg')))

print("图片总数为：",image_count)
#图片总数为：578

四**、数据预处理**

复制代码 
batch_size = 32
img_height = 224
img_width = 224

"""
关于image_dataset_from_directory()的详细介绍可以参考文章：https://mtyjkh.blog.csdn.net/article/details/117018789
"""
train_ds = tf.keras.preprocessing.image_dataset_from_directory(
    "./data/train/",
    seed=123,
    image_size=(img_height, img_width),
    batch_size=batch_size)

"""
关于image_dataset_from_directory()的详细介绍可以参考文章：https://mtyjkh.blog.csdn.net/article/details/117018789
"""
val_ds = tf.keras.preprocessing.image_dataset_from_directory(
    "./data/test/",
    seed=123,
    image_size=(img_height, img_width),
    batch_size=batch_size)

class_names = train_ds.class_names
print(class_names)

运行结果：

复制代码 
['adidas', 'nike']

五、可视化图片

复制代码 
plt.figure(figsize=(20, 10))

for images, labels in train_ds.take(1):
    for i in range(20):
        ax = plt.subplot(5, 10, i + 1)

        plt.imshow(images[i].numpy().astype("uint8"))
        plt.title(class_names[labels[i]])
        
        plt.axis("off")
plt.show()

运行结果：

​​

再次检查数据：

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

运行结果：

复制代码 
(32, 224, 224, 3)
(32,)

六、配置数据集

  • shuffle() ：打乱数据，关于此函数的详细介绍可以参考：https://zhuanlan.zhihu.com/p/42417456
  • prefetch()：预取数据，加速运行
  • cache()：将数据集缓存到内存当中，加速运行
python 复制代码 
AUTOTUNE = tf.data.AUTOTUNE

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

七、构建CNN网络模型

卷积神经网络（CNN）的输入是张量 (Tensor) 形式的 (image_height, image_width, color_channels)，包含了图像高度、宽度及颜色信息。不需要输入batch size。color_channels 为 (R,G,B) 分别对应 RGB 的三个颜色通道（color channel）。在此示例中，我们的 CNN 输入形状是 (180, 180, 3)。我们需要在声明第一层时将形状赋值给参数input_shape。

python 复制代码 
"""
关于卷积核的计算不懂的可以参考文章：https://blog.csdn.net/qq_38251616/article/details/114278995

layers.Dropout(0.4) 作用是防止过拟合，提高模型的泛化能力。
关于Dropout层的更多介绍可以参考文章：https://mtyjkh.blog.csdn.net/article/details/115826689
"""

model = models.Sequential([
    layers.experimental.preprocessing.Rescaling(1./255, input_shape=(img_height, img_width, 3)),
    
    layers.Conv2D(16, (3, 3), activation='relu', input_shape=(img_height, img_width, 3)), # 卷积层1，卷积核3*3  
    layers.AveragePooling2D((2, 2)),               # 池化层1，2*2采样
    layers.Conv2D(32, (3, 3), activation='relu'),  # 卷积层2，卷积核3*3
    layers.AveragePooling2D((2, 2)),               # 池化层2，2*2采样
    layers.Dropout(0.3),  
    layers.Conv2D(64, (3, 3), activation='relu'),  # 卷积层3，卷积核3*3
    layers.Dropout(0.3),  
    
    layers.Flatten(),                       # Flatten层，连接卷积层与全连接层
    layers.Dense(128, activation='relu'),   # 全连接层，特征进一步提取
    layers.Dense(len(class_names))               # 输出层，输出预期结果
])

model.summary()  # 打印网络结构

运行结果：

python 复制代码 
_________________________________________________________________
 Layer (type)                Output Shape              Param #
=================================================================
 rescaling (Rescaling)       (None, 224, 224, 3)       0

 conv2d (Conv2D)             (None, 222, 222, 16)      448

 average_pooling2d (AverageP  (None, 111, 111, 16)     0
 ooling2D)

 conv2d_1 (Conv2D)           (None, 109, 109, 32)      4640

 average_pooling2d_1 (Averag  (None, 54, 54, 32)       0
 ePooling2D)

 dropout (Dropout)           (None, 54, 54, 32)        0

 conv2d_2 (Conv2D)           (None, 52, 52, 64)        18496

 dropout_1 (Dropout)         (None, 52, 52, 64)        0

 flatten (Flatten)           (None, 173056)            0

 dense (Dense)               (None, 128)               22151296

 dense_1 (Dense)             (None, 2)                 258

=================================================================
Total params: 22,175,138
Trainable params: 22,175,138
Non-trainable params: 0
_________________________________________________________________

八、编译

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。
  • 指标（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。
python 复制代码 
# 设置初始学习率
initial_learning_rate = 0.001

lr_schedule = tf.keras.optimizers.schedules.ExponentialDecay(
        initial_learning_rate, 
        decay_steps=10,      # 敲黑板！！！这里是指 steps，不是指epochs
        decay_rate=0.92,     # lr经过一次衰减就会变成 decay_rate*lr
        staircase=True)

# 将指数衰减学习率送入优化器
optimizer = tf.keras.optimizers.Adam(learning_rate=lr_schedule)

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

早停与保存最佳模型参数

python 复制代码 
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping

epochs = 50

# 保存最佳模型参数
checkpointer = ModelCheckpoint('best_model.h5',
                                monitor='val_accuracy',
                                verbose=1,
                                save_best_only=True,
                                save_weights_only=True)

# 设置早停
earlystopper = EarlyStopping(monitor='val_accuracy', 
                             min_delta=0.001,
                             patience=20, 
                             verbose=1)

九、训练模型

python 复制代码 
history = model.fit(train_ds,
                    validation_data=val_ds,
                    epochs=epochs,
                    callbacks=[checkpointer, earlystopper])

运行结果：

python 复制代码 
Epoch 1/50
16/16 [==============================] - ETA: 0s - loss: 3.6308 - accuracy: 0.5000
Epoch 1: val_accuracy improved from -inf to 0.48684, saving model to best_model.h5
16/16 [==============================] - 7s 73ms/step - loss: 3.6308 - accuracy: 0.5000 - val_loss: 0.6932 - val_accuracy: 0.4868
Epoch 2/50
16/16 [==============================] - ETA: 0s - loss: 0.6951 - accuracy: 0.4880
Epoch 2: val_accuracy improved from 0.48684 to 0.50000, saving model to best_model.h5
16/16 [==============================] - 1s 40ms/step - loss: 0.6951 - accuracy: 0.4880 - val_loss: 0.6949 - val_accuracy: 0.5000
Epoch 3/50
15/16 [===========================>..] - ETA: 0s - loss: 0.6928 - accuracy: 0.4979
Epoch 3: val_accuracy did not improve from 0.50000
16/16 [==============================] - 1s 33ms/step - loss: 0.6927 - accuracy: 0.5060 - val_loss: 0.6932 - val_accuracy: 0.5000
Epoch 4/50
15/16 [===========================>..] - ETA: 0s - loss: 0.6922 - accuracy: 0.5553
Epoch 4: val_accuracy improved from 0.50000 to 0.51316, saving model to best_model.h5
16/16 [==============================] - 1s 41ms/step - loss: 0.6920 - accuracy: 0.5578 - val_loss: 0.6925 - val_accuracy: 0.5132
Epoch 5/50
15/16 [===========================>..] - ETA: 0s - loss: 0.6894 - accuracy: 0.5574
Epoch 5: val_accuracy improved from 0.51316 to 0.65789, saving model to best_model.h5
16/16 [==============================] - 1s 39ms/step - loss: 0.6890 - accuracy: 0.5697 - val_loss: 0.6891 - val_accuracy: 0.6579
Epoch 6/50
15/16 [===========================>..] - ETA: 0s - loss: 0.6883 - accuracy: 0.5340
Epoch 6: val_accuracy did not improve from 0.65789
16/16 [==============================] - 1s 33ms/step - loss: 0.6882 - accuracy: 0.5339 - val_loss: 0.6823 - val_accuracy: 0.6184
Epoch 7/50
15/16 [===========================>..] - ETA: 0s - loss: 0.6810 - accuracy: 0.6191
Epoch 7: val_accuracy did not improve from 0.65789
16/16 [==============================] - 1s 33ms/step - loss: 0.6805 - accuracy: 0.6155 - val_loss: 0.6774 - val_accuracy: 0.6316
Epoch 8/50
15/16 [===========================>..] - ETA: 0s - loss: 0.6737 - accuracy: 0.6043
Epoch 8: val_accuracy improved from 0.65789 to 0.71053, saving model to best_model.h5
16/16 [==============================] - 1s 39ms/step - loss: 0.6738 - accuracy: 0.5996 - val_loss: 0.6608 - val_accuracy: 0.7105
Epoch 9/50
15/16 [===========================>..] - ETA: 0s - loss: 0.6461 - accuracy: 0.6979
Epoch 9: val_accuracy did not improve from 0.71053
16/16 [==============================] - 1s 33ms/step - loss: 0.6424 - accuracy: 0.7012 - val_loss: 0.6200 - val_accuracy: 0.6974
Epoch 10/50
15/16 [===========================>..] - ETA: 0s - loss: 0.6148 - accuracy: 0.6979
Epoch 10: val_accuracy did not improve from 0.71053
16/16 [==============================] - 1s 34ms/step - loss: 0.6114 - accuracy: 0.6972 - val_loss: 0.6302 - val_accuracy: 0.6316
Epoch 11/50
15/16 [===========================>..] - ETA: 0s - loss: 0.5956 - accuracy: 0.7234
Epoch 11: val_accuracy improved from 0.71053 to 0.73684, saving model to best_model.h5
16/16 [==============================] - 1s 39ms/step - loss: 0.5968 - accuracy: 0.7191 - val_loss: 0.5779 - val_accuracy: 0.7368
Epoch 12/50
15/16 [===========================>..] - ETA: 0s - loss: 0.5442 - accuracy: 0.7723
Epoch 12: val_accuracy did not improve from 0.73684
16/16 [==============================] - 1s 33ms/step - loss: 0.5505 - accuracy: 0.7570 - val_loss: 0.6001 - val_accuracy: 0.6579
Epoch 13/50
15/16 [===========================>..] - ETA: 0s - loss: 0.5566 - accuracy: 0.7298
Epoch 13: val_accuracy improved from 0.73684 to 0.75000, saving model to best_model.h5
16/16 [==============================] - 1s 40ms/step - loss: 0.5581 - accuracy: 0.7251 - val_loss: 0.5442 - val_accuracy: 0.7500
Epoch 14/50
15/16 [===========================>..] - ETA: 0s - loss: 0.5194 - accuracy: 0.7617
Epoch 14: val_accuracy did not improve from 0.75000
16/16 [==============================] - 1s 33ms/step - loss: 0.5200 - accuracy: 0.7629 - val_loss: 0.5347 - val_accuracy: 0.7368
Epoch 15/50
15/16 [===========================>..] - ETA: 0s - loss: 0.5114 - accuracy: 0.7681
Epoch 15: val_accuracy did not improve from 0.75000
16/16 [==============================] - 1s 33ms/step - loss: 0.5048 - accuracy: 0.7769 - val_loss: 0.5161 - val_accuracy: 0.7500
Epoch 16/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4836 - accuracy: 0.7830
Epoch 16: val_accuracy improved from 0.75000 to 0.76316, saving model to best_model.h5
16/16 [==============================] - 1s 40ms/step - loss: 0.4901 - accuracy: 0.7789 - val_loss: 0.5069 - val_accuracy: 0.7632
Epoch 17/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4636 - accuracy: 0.7809
Epoch 17: val_accuracy did not improve from 0.76316
16/16 [==============================] - 1s 33ms/step - loss: 0.4585 - accuracy: 0.7888 - val_loss: 0.5071 - val_accuracy: 0.7500
Epoch 18/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4717 - accuracy: 0.7723
Epoch 18: val_accuracy did not improve from 0.76316
16/16 [==============================] - 1s 34ms/step - loss: 0.4655 - accuracy: 0.7769 - val_loss: 0.5034 - val_accuracy: 0.7368
Epoch 19/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4610 - accuracy: 0.8064
Epoch 19: val_accuracy did not improve from 0.76316
16/16 [==============================] - 1s 33ms/step - loss: 0.4567 - accuracy: 0.8088 - val_loss: 0.5440 - val_accuracy: 0.7368
Epoch 20/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4547 - accuracy: 0.7872
Epoch 20: val_accuracy improved from 0.76316 to 0.78947, saving model to best_model.h5
16/16 [==============================] - 1s 40ms/step - loss: 0.4507 - accuracy: 0.7948 - val_loss: 0.4812 - val_accuracy: 0.7895
Epoch 21/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4228 - accuracy: 0.8298
Epoch 21: val_accuracy did not improve from 0.78947
16/16 [==============================] - 1s 33ms/step - loss: 0.4238 - accuracy: 0.8287 - val_loss: 0.4926 - val_accuracy: 0.7632
Epoch 22/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4460 - accuracy: 0.8125
Epoch 22: val_accuracy did not improve from 0.78947
16/16 [==============================] - 1s 33ms/step - loss: 0.4386 - accuracy: 0.8187 - val_loss: 0.4857 - val_accuracy: 0.7632
Epoch 23/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4262 - accuracy: 0.8167
Epoch 23: val_accuracy did not improve from 0.78947
16/16 [==============================] - 1s 34ms/step - loss: 0.4204 - accuracy: 0.8227 - val_loss: 0.4718 - val_accuracy: 0.7632
Epoch 24/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4196 - accuracy: 0.8277
Epoch 24: val_accuracy did not improve from 0.78947
16/16 [==============================] - 1s 33ms/step - loss: 0.4208 - accuracy: 0.8247 - val_loss: 0.5068 - val_accuracy: 0.7632
Epoch 25/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4112 - accuracy: 0.8362
Epoch 25: val_accuracy did not improve from 0.78947
16/16 [==============================] - 1s 33ms/step - loss: 0.4118 - accuracy: 0.8347 - val_loss: 0.4658 - val_accuracy: 0.7895
Epoch 26/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4005 - accuracy: 0.8298
Epoch 26: val_accuracy did not improve from 0.78947
16/16 [==============================] - 1s 34ms/step - loss: 0.3981 - accuracy: 0.8347 - val_loss: 0.4822 - val_accuracy: 0.7632
Epoch 27/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4003 - accuracy: 0.8426
Epoch 27: val_accuracy did not improve from 0.78947
16/16 [==============================] - 1s 34ms/step - loss: 0.4038 - accuracy: 0.8406 - val_loss: 0.4756 - val_accuracy: 0.7763
Epoch 28/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3884 - accuracy: 0.8511
Epoch 28: val_accuracy improved from 0.78947 to 0.80263, saving model to best_model.h5
16/16 [==============================] - 1s 40ms/step - loss: 0.3967 - accuracy: 0.8486 - val_loss: 0.4636 - val_accuracy: 0.8026
Epoch 29/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4139 - accuracy: 0.8489
Epoch 29: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 33ms/step - loss: 0.4091 - accuracy: 0.8486 - val_loss: 0.4735 - val_accuracy: 0.7763
Epoch 30/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3857 - accuracy: 0.8617
Epoch 30: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3870 - accuracy: 0.8586 - val_loss: 0.4655 - val_accuracy: 0.7763
Epoch 31/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3853 - accuracy: 0.8447
Epoch 31: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 33ms/step - loss: 0.3908 - accuracy: 0.8347 - val_loss: 0.4688 - val_accuracy: 0.7763
Epoch 32/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3814 - accuracy: 0.8596
Epoch 32: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3869 - accuracy: 0.8546 - val_loss: 0.4728 - val_accuracy: 0.7632
Epoch 33/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3938 - accuracy: 0.8396
Epoch 33: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3887 - accuracy: 0.8446 - val_loss: 0.4798 - val_accuracy: 0.7763
Epoch 34/50
15/16 [===========================>..] - ETA: 0s - loss: 0.4032 - accuracy: 0.8542
Epoch 34: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3955 - accuracy: 0.8586 - val_loss: 0.4708 - val_accuracy: 0.7632
Epoch 35/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3937 - accuracy: 0.8375
Epoch 35: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3865 - accuracy: 0.8426 - val_loss: 0.4695 - val_accuracy: 0.7632
Epoch 36/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3883 - accuracy: 0.8447
Epoch 36: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3862 - accuracy: 0.8486 - val_loss: 0.4700 - val_accuracy: 0.7632
Epoch 37/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3729 - accuracy: 0.8617
Epoch 37: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 33ms/step - loss: 0.3767 - accuracy: 0.8586 - val_loss: 0.4685 - val_accuracy: 0.7632
Epoch 38/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3831 - accuracy: 0.8479
Epoch 38: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3788 - accuracy: 0.8506 - val_loss: 0.4720 - val_accuracy: 0.7632
Epoch 39/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3872 - accuracy: 0.8468
Epoch 39: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 33ms/step - loss: 0.3872 - accuracy: 0.8466 - val_loss: 0.4648 - val_accuracy: 0.7895
Epoch 40/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3800 - accuracy: 0.8489
Epoch 40: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3739 - accuracy: 0.8546 - val_loss: 0.4682 - val_accuracy: 0.7632
Epoch 41/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3800 - accuracy: 0.8511
Epoch 41: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3813 - accuracy: 0.8486 - val_loss: 0.4649 - val_accuracy: 0.7895
Epoch 42/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3727 - accuracy: 0.8617
Epoch 42: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3712 - accuracy: 0.8645 - val_loss: 0.4675 - val_accuracy: 0.7632
Epoch 43/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3803 - accuracy: 0.8468
Epoch 43: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3830 - accuracy: 0.8486 - val_loss: 0.4672 - val_accuracy: 0.7632
Epoch 44/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3648 - accuracy: 0.8745
Epoch 44: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3698 - accuracy: 0.8705 - val_loss: 0.4708 - val_accuracy: 0.7632
Epoch 45/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3742 - accuracy: 0.8489
Epoch 45: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 33ms/step - loss: 0.3695 - accuracy: 0.8546 - val_loss: 0.4683 - val_accuracy: 0.7632
Epoch 46/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3792 - accuracy: 0.8447
Epoch 46: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 33ms/step - loss: 0.3878 - accuracy: 0.8406 - val_loss: 0.4706 - val_accuracy: 0.7632
Epoch 47/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3681 - accuracy: 0.8745
Epoch 47: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3639 - accuracy: 0.8745 - val_loss: 0.4708 - val_accuracy: 0.7632
Epoch 48/50
15/16 [===========================>..] - ETA: 0s - loss: 0.3771 - accuracy: 0.8489
Epoch 48: val_accuracy did not improve from 0.80263
16/16 [==============================] - 1s 34ms/step - loss: 0.3778 - accuracy: 0.8506 - val_loss: 0.4729 - val_accuracy: 0.7632
Epoch 48: early stopping
1/1 [==============================] - 0s 100ms/step

十、模型评估

python 复制代码 
acc = history.history['accuracy']
val_acc = history.history['val_accuracy']

loss = history.history['loss']
val_loss = history.history['val_loss']

epochs_range = range(len(loss))

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()

十一、指定图片预测

python 复制代码 
# 加载效果最好的模型权重
model.load_weights('best_model.h5')
from PIL import Image
import numpy as np


img = Image.open("./data/test/nike/1.jpg")  #这里选择你需要预测的图片
image = tf.image.resize(img, [img_height, img_width])

img_array = tf.expand_dims(image, 0) #/255.0  # 记得做归一化处理（与训练集处理方式保持一致）

predictions = model.predict(img_array) # 这里选用你已经训练好的模型
print("预测结果为：",class_names[np.argmax(predictions)])

运行结果：

python 复制代码 
预测结果为： nike

十二、总结

本周通过学习TensorFlow实现运动鞋品牌识别;首先学习设置动态学习率，在训练神经网络时动态地降低学习率，可以帮助优化算法更有效地收敛到全局最小值，从而提高模型的性能。其次就是学习早停与保存最佳模型参数，模型在指定epoch次都没有提升的情况下，可以提前停止训练。

相关推荐
Norvyn_728 分钟前
LeetCode｜Day18｜20. 有效的括号｜Python刷题笔记
笔记·python·leetcode
AndrewHZ35 分钟前
【图像处理基石】如何入门色彩评估？
图像处理·人工智能·深度学习·色彩科学·hvs·色彩评估·颜色工程
静心问道1 小时前
TrOCR: 基于Transformer的光学字符识别方法，使用预训练模型
人工智能·深度学习·transformer·多模态
亲持红叶1 小时前
GLU 变种：ReGLU 、 GEGLU 、 SwiGLU
人工智能·深度学习·神经网络·激活函数
Y4090012 小时前
C语言转Java语言，相同与相异之处
java·c语言·开发语言·笔记
笑衬人心。2 小时前
TCP 拥塞控制算法 —— 慢启动（Slow Start）笔记
笔记·tcp/ip·php
花海如潮淹2 小时前
前端性能追踪工具：用户体验的毫秒战争
前端·笔记·ux
Andy杨3 小时前
20250718-5-Kubernetes 调度-Pod对象：重启策略+健康检查_笔记
笔记·容器·kubernetes
杭州杭州杭州8 小时前
Python笔记
开发语言·笔记·python
热门推荐
01全球最强模型Grok4，国内已可免费使用！（附教程）02Cursor Claude 模型无法使用的解决方法03KGG转MP3工具|非KGM文件|解密音频04【2025.7.18】更新vscode后所有.vue文件template标签后报红的临时解决办法，Vue - Official 插件3.0.2导致05【无标题】06集群聊天服务器---MySQL数据库的建立07突破限制：使用 Claude Code Proxy 让 Claude Code 自由连接任意模型08绿色建筑新态势：楼宇自控助力能效提升，推动成本优化新路径09使用Ruby接入实时行情API教程10Claude Code 最新版已经支持 Windows 安装使用！