Tensorflow实现深度学习案例5：运动鞋识别

最新推荐文章于 2024-08-20 23:36:24 发布

小叮当爱咖啡

最新推荐文章于 2024-08-20 23:36:24 发布

阅读量667

点赞数 8

文章标签：深度学习 tensorflow 人工智能

本文链接：https://blog.csdn.net/2301_76606951/article/details/141068172

版权

本文为为🔗365天深度学习训练营内部文章

原作者：K同学啊

一前期准备

1.导入数据

from tensorflow       import keras
from tensorflow.keras import layers,models
import os, PIL, pathlib
import matplotlib.pyplot as plt
import tensorflow        as tf

data_dir = "./shoes/"

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

2.查看数据

roses = list(data_dir.glob('train/nike/*.jpg'))
PIL.Image.open(str(roses[0]))

二、数据预处理

1. 加载数据

使用image_dataset_from_directory方法将磁盘中的数据加载到tf.data.Dataset中

tf.keras.preprocessing.image_dataset_from_directory()：是 TensorFlow 的 Keras 模块中的一个函数，用于从目录中创建一个图像数据集（dataset）。这个函数可以以更方便的方式加载图像数据，用于训练和评估神经网络模型。

测试集与验证集的关系：

验证集并没有参与训练过程梯度下降过程的，狭义上来讲是没有参与模型的参数训练更新的。

但是广义上来讲，验证集存在的意义确实参与了一个“人工调参”的过程，我们根据每一个epoch训练之后模型在valid data上的表现来决定是否需要训练进行early stop，或者根据这个过程模型的性能变化来调整模型的超参数，如学习率，batch_size等等。

因此，我们也可以认为，验证集也参与了训练，但是并没有使得模型去overfit验证集

batch_size = 32
img_height = 224
img_width = 224

train_ds = tf.keras.preprocessing.image_dataset_from_directory(
    "./shoes/train/",
    seed=123,
    image_size=(img_height, img_width),
    batch_size=batch_size)

Found 502 files belonging to 2 classes.

val_ds = tf.keras.preprocessing.image_dataset_from_directory(
    "./shoes/test/",
    seed=123,
    image_size=(img_height, img_width),
    batch_size=batch_size)

Found 76 files belonging to 2 classes.

class_names = train_ds.class_names
print(class_names)

['adidas', 'nike']

2. 可视化数据

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")

3.再次检查数据

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

(32, 224, 224, 3)
(32,)

Image_batch是形状的张量（32,224,224,3）。这是一批形状224x224x3的32张图片（最后一维指的是彩色通道RGB）。

Label_batch是形状（32，）的张量，这些标签对应32张图片

4.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

prefetch()功能详细介绍：CPU 正在准备数据时，加速器处于空闲状态。相反，当加速器正在训练模型时，CPU 处于空闲状态。因此，训练所用的时间是 CPU 预处理时间和加速器训练时间的总和。prefetch()将训练步骤的预处理和模型执行过程重叠到一起。当加速器正在执行第 N 个训练步时，CPU 正在准备第 N+1 步的数据。这样做不仅可以最大限度地缩短训练的单步用时（而不是总用时），而且可以缩短提取和转换数据所需的时间。如果不使用prefetch()，CPU 和 GPU/TPU 在大部分时间都处于空闲状态：

AUTOTUNE = tf.data.AUTOTUNE

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

使用prefetch()可显著减少空闲时间：

cache() ：将数据集缓存到内存当中，加速运行

AUTOTUNE = tf.data.AUTOTUNE

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

三、构建CNN网络

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 输入的形状是 (224, 224, 3)即彩色图像。我们需要在声明第一层时将形状赋值给参数input_shape。

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

Model: "sequential"
_________________________________________________________________
 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
_________________________________________________________________

四、训练模型

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。

1.设置动态学习率

📮 ExponentialDecay函数：

tf.keras.optimizers.schedules.ExponentialDecay是 TensorFlow 中的一个学习率衰减策略，用于在训练神经网络时动态地降低学习率。学习率衰减是一种常用的技巧，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。

🔎 主要参数：

initial_learning_rate（初始学习率）：初始学习率大小。
decay_steps（衰减步数）：学习率衰减的步数。在经过 decay_steps 步后，学习率将按照指数函数衰减。例如，如果 decay_steps 设置为 10，则每10步衰减一次。
decay_rate（衰减率）：学习率的衰减率。它决定了学习率如何衰减。通常，取值在 0 到 1 之间。
staircase（阶梯式衰减）：一个布尔值，控制学习率的衰减方式。如果设置为 True，则学习率在每个 decay_steps 步之后直接减小，形成阶梯状下降。如果设置为 False，则学习率将连续衰减。

# 设置初始学习率
initial_learning_rate = 0.1

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'])

注：这里设置的动态学习率为：指数衰减型（ExponentialDecay）。在每一个epoch开始前，学习率（learning_rate）都将会重置为初始学习率（initial_learning_rate），然后再重新开始衰减。计算公式如下：

learning_rate = initial_learning_rate * decay_rate ^ (step / decay_steps)

学习率大与学习率小的优缺点分析:

学习率大

优点：

1、加快学习速率。

2、有助于跳出局部最优值。

缺点：

1、导致模型训练不收敛。

2、单单使用大学习率容易导致模型不精确。

学习率小

优点：

1、有助于模型收敛、模型细化。

2、提高模型精度。

缺点：

1、很难跳出局部最优值。

2、收敛缓慢。

2.早停与保存最佳模型参数

EarlyStopping()参数说明：

monitor: 被监测的数据。

min_delta: 在被监测的数据中被认为是提升的最小变化，例如，小于 min_delta 的绝对变化会被认为没有提升。

patience: 没有进步的训练轮数，在这之后训练就会被停止。

verbose: 详细信息模式。

mode: {auto, min, max} 其中之一。在 min 模式中，当被监测的数据停止下降，训练就会停止；在 max 模式中，当被监测的数据停止上升，训练就会停止；在 auto 模式中，方向会自动从被监测的数据的名字中判断出来。

baseline: 要监控的数量的基准值。如果模型没有显示基准的改善，训练将停止。

estore_best_weights: 是否从具有监测数量的最佳值的时期恢复模型权重。如果为 False，则使用在训练的最后一步获得的模型权重。

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)

3.模型训练

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

Epoch 1/50
16/16 [==============================] - ETA: 0s - loss: 2.5806 - accuracy: 0.4821
Epoch 1: val_accuracy improved from -inf to 0.50000, saving model to best_model.h5
16/16 [==============================] - 11s 656ms/step - loss: 2.5806 - accuracy: 0.4821 - val_loss: 0.6933 - val_accuracy: 0.5000
Epoch 2/50
16/16 [==============================] - ETA: 0s - loss: 0.6937 - accuracy: 0.5040
Epoch 2: val_accuracy did not improve from 0.50000
16/16 [==============================] - 9s 551ms/step - loss: 0.6937 - accuracy: 0.5040 - val_loss: 0.6932 - val_accuracy: 0.5000
Epoch 3/50
16/16 [==============================] - ETA: 0s - loss: 0.6916 - accuracy: 0.5339
Epoch 3: val_accuracy improved from 0.50000 to 0.51316, saving model to best_model.h5
16/16 [==============================] - 10s 656ms/step - loss: 0.6916 - accuracy: 0.5339 - val_loss: 0.6929 - val_accuracy: 0.5132
Epoch 4/50
16/16 [==============================] - ETA: 0s - loss: 0.6868 - accuracy: 0.5299
Epoch 4: val_accuracy improved from 0.51316 to 0.53947, saving model to best_model.h5
16/16 [==============================] - 10s 661ms/step - loss: 0.6868 - accuracy: 0.5299 - val_loss: 0.6822 - val_accuracy: 0.5395
Epoch 5/50
16/16 [==============================] - ETA: 0s - loss: 0.6867 - accuracy: 0.5139
Epoch 5: val_accuracy did not improve from 0.53947
16/16 [==============================] - 9s 571ms/step - loss: 0.6867 - accuracy: 0.5139 - val_loss: 0.6668 - val_accuracy: 0.5000
Epoch 6/50
16/16 [==============================] - ETA: 0s - loss: 0.6774 - accuracy: 0.5558
Epoch 6: val_accuracy did not improve from 0.53947
16/16 [==============================] - 9s 567ms/step - loss: 0.6774 - accuracy: 0.5558 - val_loss: 0.6570 - val_accuracy: 0.5263
Epoch 7/50
16/16 [==============================] - ETA: 0s - loss: 0.6658 - accuracy: 0.6016
Epoch 7: val_accuracy improved from 0.53947 to 0.65789, saving model to best_model.h5
16/16 [==============================] - 10s 659ms/step - loss: 0.6658 - accuracy: 0.6016 - val_loss: 0.6493 - val_accuracy: 0.6579
Epoch 8/50
16/16 [==============================] - ETA: 0s - loss: 0.6479 - accuracy: 0.6056
Epoch 8: val_accuracy did not improve from 0.65789
16/16 [==============================] - 9s 578ms/step - loss: 0.6479 - accuracy: 0.6056 - val_loss: 0.6391 - val_accuracy: 0.6053
Epoch 9/50
16/16 [==============================] - ETA: 0s - loss: 0.6262 - accuracy: 0.6414
Epoch 9: val_accuracy improved from 0.65789 to 0.67105, saving model to best_model.h5
16/16 [==============================] - 11s 702ms/step - loss: 0.6262 - accuracy: 0.6414 - val_loss: 0.6079 - val_accuracy: 0.6711
Epoch 10/50
16/16 [==============================] - ETA: 0s - loss: 0.5956 - accuracy: 0.6713
Epoch 10: val_accuracy improved from 0.67105 to 0.71053, saving model to best_model.h5
16/16 [==============================] - 11s 691ms/step - loss: 0.5956 - accuracy: 0.6713 - val_loss: 0.5862 - val_accuracy: 0.7105
Epoch 11/50
16/16 [==============================] - ETA: 0s - loss: 0.5662 - accuracy: 0.7131
Epoch 11: val_accuracy improved from 0.71053 to 0.72368, saving model to best_model.h5
16/16 [==============================] - 11s 668ms/step - loss: 0.5662 - accuracy: 0.7131 - val_loss: 0.5471 - val_accuracy: 0.7237
Epoch 12/50
16/16 [==============================] - ETA: 0s - loss: 0.5252 - accuracy: 0.7490
Epoch 12: val_accuracy did not improve from 0.72368
16/16 [==============================] - 9s 586ms/step - loss: 0.5252 - accuracy: 0.7490 - val_loss: 0.5893 - val_accuracy: 0.6842
Epoch 13/50
16/16 [==============================] - ETA: 0s - loss: 0.5170 - accuracy: 0.7430
Epoch 13: val_accuracy did not improve from 0.72368
16/16 [==============================] - 10s 596ms/step - loss: 0.5170 - accuracy: 0.7430 - val_loss: 0.5504 - val_accuracy: 0.7105
Epoch 14/50
16/16 [==============================] - ETA: 0s - loss: 0.4729 - accuracy: 0.8028
Epoch 14: val_accuracy did not improve from 0.72368
16/16 [==============================] - 9s 555ms/step - loss: 0.4729 - accuracy: 0.8028 - val_loss: 0.5343 - val_accuracy: 0.6711
Epoch 15/50
16/16 [==============================] - ETA: 0s - loss: 0.4450 - accuracy: 0.8088
Epoch 15: val_accuracy improved from 0.72368 to 0.75000, saving model to best_model.h5
16/16 [==============================] - 11s 666ms/step - loss: 0.4450 - accuracy: 0.8088 - val_loss: 0.5025 - val_accuracy: 0.7500
Epoch 16/50
16/16 [==============================] - ETA: 0s - loss: 0.4374 - accuracy: 0.7928
Epoch 16: val_accuracy did not improve from 0.75000
16/16 [==============================] - 9s 582ms/step - loss: 0.4374 - accuracy: 0.7928 - val_loss: 0.5023 - val_accuracy: 0.7368
Epoch 17/50
16/16 [==============================] - ETA: 0s - loss: 0.4026 - accuracy: 0.8327
Epoch 17: val_accuracy did not improve from 0.75000
16/16 [==============================] - 9s 588ms/step - loss: 0.4026 - accuracy: 0.8327 - val_loss: 0.5010 - val_accuracy: 0.7368
Epoch 18/50
16/16 [==============================] - ETA: 0s - loss: 0.3861 - accuracy: 0.8406
Epoch 18: val_accuracy did not improve from 0.75000
16/16 [==============================] - 9s 573ms/step - loss: 0.3861 - accuracy: 0.8406 - val_loss: 0.4924 - val_accuracy: 0.7500
Epoch 19/50
16/16 [==============================] - ETA: 0s - loss: 0.3663 - accuracy: 0.8506
Epoch 19: val_accuracy did not improve from 0.75000
16/16 [==============================] - 9s 563ms/step - loss: 0.3663 - accuracy: 0.8506 - val_loss: 0.4749 - val_accuracy: 0.7500
Epoch 20/50
16/16 [==============================] - ETA: 0s - loss: 0.3476 - accuracy: 0.8725
Epoch 20: val_accuracy improved from 0.75000 to 0.76316, saving model to best_model.h5
16/16 [==============================] - 11s 715ms/step - loss: 0.3476 - accuracy: 0.8725 - val_loss: 0.4688 - val_accuracy: 0.7632
Epoch 21/50
16/16 [==============================] - ETA: 0s - loss: 0.3424 - accuracy: 0.8705
Epoch 21: val_accuracy improved from 0.76316 to 0.77632, saving model to best_model.h5
16/16 [==============================] - 11s 683ms/step - loss: 0.3424 - accuracy: 0.8705 - val_loss: 0.4727 - val_accuracy: 0.7763
Epoch 22/50
16/16 [==============================] - ETA: 0s - loss: 0.3318 - accuracy: 0.8825
Epoch 22: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 574ms/step - loss: 0.3318 - accuracy: 0.8825 - val_loss: 0.4668 - val_accuracy: 0.7763
Epoch 23/50
16/16 [==============================] - ETA: 0s - loss: 0.3398 - accuracy: 0.8904
Epoch 23: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 571ms/step - loss: 0.3398 - accuracy: 0.8904 - val_loss: 0.4669 - val_accuracy: 0.7763
Epoch 24/50
16/16 [==============================] - ETA: 0s - loss: 0.3356 - accuracy: 0.8606
Epoch 24: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 559ms/step - loss: 0.3356 - accuracy: 0.8606 - val_loss: 0.4666 - val_accuracy: 0.7763
Epoch 25/50
16/16 [==============================] - ETA: 0s - loss: 0.3260 - accuracy: 0.8685
Epoch 25: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 558ms/step - loss: 0.3260 - accuracy: 0.8685 - val_loss: 0.4697 - val_accuracy: 0.7763
Epoch 26/50
16/16 [==============================] - ETA: 0s - loss: 0.3083 - accuracy: 0.8865
Epoch 26: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 565ms/step - loss: 0.3083 - accuracy: 0.8865 - val_loss: 0.4671 - val_accuracy: 0.7763
Epoch 27/50
16/16 [==============================] - ETA: 0s - loss: 0.3073 - accuracy: 0.8904
Epoch 27: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 567ms/step - loss: 0.3073 - accuracy: 0.8904 - val_loss: 0.4662 - val_accuracy: 0.7763
Epoch 28/50
16/16 [==============================] - ETA: 0s - loss: 0.2896 - accuracy: 0.8984
Epoch 28: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 564ms/step - loss: 0.2896 - accuracy: 0.8984 - val_loss: 0.4694 - val_accuracy: 0.7632
Epoch 29/50
16/16 [==============================] - ETA: 0s - loss: 0.2895 - accuracy: 0.8884
Epoch 29: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 568ms/step - loss: 0.2895 - accuracy: 0.8884 - val_loss: 0.4652 - val_accuracy: 0.7632
Epoch 30/50
16/16 [==============================] - ETA: 0s - loss: 0.2913 - accuracy: 0.8884
Epoch 30: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 556ms/step - loss: 0.2913 - accuracy: 0.8884 - val_loss: 0.4658 - val_accuracy: 0.7632
Epoch 31/50
16/16 [==============================] - ETA: 0s - loss: 0.2802 - accuracy: 0.9024
Epoch 31: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 568ms/step - loss: 0.2802 - accuracy: 0.9024 - val_loss: 0.4606 - val_accuracy: 0.7632
Epoch 32/50
16/16 [==============================] - ETA: 0s - loss: 0.2933 - accuracy: 0.8745
Epoch 32: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 559ms/step - loss: 0.2933 - accuracy: 0.8745 - val_loss: 0.4631 - val_accuracy: 0.7632
Epoch 33/50
16/16 [==============================] - ETA: 0s - loss: 0.3122 - accuracy: 0.8884
Epoch 33: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 573ms/step - loss: 0.3122 - accuracy: 0.8884 - val_loss: 0.4572 - val_accuracy: 0.7632
Epoch 34/50
16/16 [==============================] - ETA: 0s - loss: 0.2859 - accuracy: 0.8964
Epoch 34: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 594ms/step - loss: 0.2859 - accuracy: 0.8964 - val_loss: 0.4593 - val_accuracy: 0.7632
Epoch 35/50
16/16 [==============================] - ETA: 0s - loss: 0.2888 - accuracy: 0.8924
Epoch 35: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 573ms/step - loss: 0.2888 - accuracy: 0.8924 - val_loss: 0.4558 - val_accuracy: 0.7632
Epoch 36/50
16/16 [==============================] - ETA: 0s - loss: 0.2860 - accuracy: 0.9104
Epoch 36: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 568ms/step - loss: 0.2860 - accuracy: 0.9104 - val_loss: 0.4548 - val_accuracy: 0.7632
Epoch 37/50
16/16 [==============================] - ETA: 0s - loss: 0.2973 - accuracy: 0.8865
Epoch 37: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 563ms/step - loss: 0.2973 - accuracy: 0.8865 - val_loss: 0.4617 - val_accuracy: 0.7632
Epoch 38/50
16/16 [==============================] - ETA: 0s - loss: 0.2973 - accuracy: 0.8904
Epoch 38: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 570ms/step - loss: 0.2973 - accuracy: 0.8904 - val_loss: 0.4594 - val_accuracy: 0.7632
Epoch 39/50
16/16 [==============================] - ETA: 0s - loss: 0.2827 - accuracy: 0.8944
Epoch 39: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 571ms/step - loss: 0.2827 - accuracy: 0.8944 - val_loss: 0.4558 - val_accuracy: 0.7632
Epoch 40/50
16/16 [==============================] - ETA: 0s - loss: 0.2877 - accuracy: 0.8924
Epoch 40: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 560ms/step - loss: 0.2877 - accuracy: 0.8924 - val_loss: 0.4571 - val_accuracy: 0.7632
Epoch 41/50
16/16 [==============================] - ETA: 0s - loss: 0.2867 - accuracy: 0.8924
Epoch 41: val_accuracy did not improve from 0.77632
16/16 [==============================] - 9s 577ms/step - loss: 0.2867 - accuracy: 0.8924 - val_loss: 0.4586 - val_accuracy: 0.7632
Epoch 41: early stopping

五、模型评估

1. Loss与Accuracy图

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()

2. 指定图片进行预测

# 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("./45-data/Monkeypox/M06_01_04.jpg")  #这里选择你需要预测的图片
img = Image.open("./46-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)])

1/1 [==============================] - 0s 30ms/step
预测结果为： nike

心得：

使用早停法可以防止过拟合、减少计算资源、避免过度调优

实现早停法的注意事项

监控指标：选择合适的监控指标（如验证集上的损失或准确率），以便准确判断何时停止训练。
容忍度设置：设置容忍度（即等待多少轮后验证性能没有改善才停止训练）可以避免过早停止导致的欠拟合。
保存最佳模型：通常在训练过程中会保存验证集上表现最好的模型，以确保最终使用的是最佳模型。