文章目录

一、环境准备二、代码编写1. 准备VGG类2. 导入相关库3.导入数据4.数据预处理5. 实例化VGG-16模型6.设置优化器7.训练模型8.模型评估

---

一、环境准备

语言环境：Python3.8.0编译器：PyCharm 深度学习环境：TensorFlow-gpu 2.5

---

二、代码编写

1. 准备VGG类

# -*-coding:utf-8-*-from tensorflow.keras import layers, Model, SequentialCONV_KERNEL_INITIALIZER = {'class_name': 'VarianceScaling','config': {'scale': 2.0,'mode': 'fan_out','distribution': 'truncated_normal'}}DENSE_KERNEL_INITIALIZER = {'class_name': 'VarianceScaling','config': {'scale': 1. / 3.,'mode': 'fan_out','distribution': 'uniform'}}# 输出预测层def VGG(feature, im_height=224, im_width=224, num_classes=1000):# tensorflow中的tensor通道排序是NHWCinput_image = layers.Input(shape=(im_height, im_width, 3), dtype="float32")x = feature(input_image) # 将提取特征层和下面的层进行连接x = layers.Flatten()(x) # 展平层，将输入数据展平成1维x = layers.Dropout(rate=0.5)(x) # 随机让50%的神经元失活，缓解过拟合x = layers.Dense(2048, activation='relu',kernel_initializer=DENSE_KERNEL_INITIALIZER)(x) # 全连接层，激活函数为relu，有2048个神经元x = layers.Dropout(rate=0.5)(x)x = layers.Dense(2048, activation='relu',kernel_initializer=DENSE_KERNEL_INITIALIZER)(x)x = layers.Dense(num_classes,kernel_initializer=DENSE_KERNEL_INITIALIZER)(x)output = layers.Softmax()(x) # softmax处理多分类问题，得到类别概率分布model = Model(inputs=input_image, outputs=output) # inputs：函数第一行中定义的input节点；outputs：softmax得到的概率分布return model# 提取特征函数def make_feature(cfg):feature_layers = []for v in cfg:if v == "M":feature_layers.append(layers.MaxPool2D(pool_size=2, strides=2))else:conv2d = layers.Conv2D(v, kernel_size=3, padding="SAME", activation="relu",kernel_initializer=CONV_KERNEL_INITIALIZER)feature_layers.append(conv2d)return Sequential(feature_layers, name="feature")# 各种vgg模型结构cfgs = {# M代表最大池化层'vgg11': [64, 'M', 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],'vgg13': [64, 64, 'M', 128, 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],'vgg16': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'M', 512, 512, 512, 'M', 512, 512, 512, 'M'],'vgg19': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 256, 'M', 512, 512, 512, 512, 'M', 512, 512, 512, 512, 'M'],}# 生成VGG模型，默认使用vgg16结构； im_height：图像高度；im_width：图像宽度；num_classes：图像类别个数def vgg(model_name="vgg16", im_height=224, im_width=224, num_classes=1000):assert model_name in cfgs.keys(), "not support model {}".format(model_name)cfg = cfgs[model_name]model = VGG(make_feature(cfg), im_height=im_height, im_width=im_width, num_classes=num_classes)return model

2. 导入相关库

import osos.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"os.environ["CUDA_VISIBLE_DEVICES"] = "0"import tensorflow as tfimport matplotlib.pyplot as pltimport numpy as npfrom tensorflow import kerasfrom tensorflow.keras import layers,modelsimport pathlibfrom VGG_Model import vgg

3.导入数据

#设置随机种子尽可能使结果得以重现np.random.seed(520)tf.random.set_seed(520)#导入数据data_dir = r"你的数据集路径"data_dir = pathlib.Path(data_dir)# 设置GPU运行gpus = tf.config.experimental.list_physical_devices("GPU")if gpus:try:for gpu in gpus:tf.config.experimental.set_memory_growth(gpu, True)except RuntimeError as e:print(e)exit(-1)# 查看数据imageCount = len(list(data_dir.glob('*/*.png')))print("图片的总数为:",imageCount)

4.数据预处理

# 1.加载数据batch_size = 32img_height = 224img_width = 224train_ds = tf.keras.preprocessing.image_dataset_from_directory(data_dir,validation_split=0.2,subset="training",seed=520,image_size=(img_height,img_width),batch_size=batch_size)val_ds = tf.keras.preprocessing.image_dataset_from_directory(data_dir,validation_split=0.2,subset="validation",seed=520,image_size=(img_height,img_width),batch_size=batch_size)class_names = train_ds.class_namesprint(class_names)# 2.可视化数据plt.figure(figsize=(10,5),dpi=100)for 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.再次检查数据for image_batch, labels_batch in train_ds:print(image_batch.shape)print(labels_batch.shape)break# 4.配置数据集# shuffle()：打乱数据# 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AUTOTUNE = tf.data.AUTOTUNEtrain_ds = train_ds.cache().shuffle(1000).prefetch(buffer_size=AUTOTUNE)val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)# 5.归一化normalization_layer = layers.experimental.preprocessing.Rescaling(1./255)normalization_train_ds = train_ds.map(lambda x, y: (normalization_layer(x), y))val_ds = val_ds.map(lambda x, y: (normalization_layer(x), y))image_batch, labels_batch = next(iter(val_ds))first_image = image_batch[0]# 6.查看归一化后的数据print(np.min(first_image), np.max(first_image))

Found 621 files belonging to 7 classes.

Using 497 files for training.

Found 621 files belonging to 7 classes.

Using 124 files for validation.

[‘lufei’, ‘luobin’, ‘namei’, ‘qiaoba’, ‘shanzhi’, ‘suolong’, ‘wusuopu’]

(32, 224, 224, 3)

(32,)

Image_batch是形状的张量（32,180,180,3）。这是一批形状180x180x3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

5. 实例化VGG-16模型

model = vgg("vgg16", 224, 224, 5)model.summary()

模型参数：

6.设置优化器

opt = tf.keras.optimizers.Adam(learning_rate=1e-4)pile(optimizer=opt,loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),metrics=['accuracy'])

7.训练模型

由于我是用CPU训练的，速度会很慢，所以这里只迭代了10次，识别准确率也不是很高

epochs = 10history = model.fit(train_ds,validation_data=val_ds,epochs=epochs)

Epoch 1/10

16/16 [] - 195s 12s/step - loss: 2.0244 - accuracy: 0.1408 - val_loss: 1.9719 - val_accuracy: 0.1935

Epoch 2/10

16/16 [] - 195s 12s/step - loss: 2.0326 - accuracy: 0.1328 - val_loss: 1.9719 - val_accuracy: 0.1935

Epoch 3/10

16/16 [] - 194s 12s/step - loss: 2.0326 - accuracy: 0.1328 - val_loss: 1.9719 - val_accuracy: 0.1935

Epoch 4/10

16/16 [] - 196s 12s/step - loss: 2.0326 - accuracy: 0.1328 - val_loss: 1.9719 - val_accuracy: 0.1935

Epoch 5/10

16/16 [] - 194s 12s/step - loss: 2.0326 - accuracy: 0.1328 - val_loss: 1.9719 - val_accuracy: 0.1935

Epoch 6/10

16/16 [] - 196s 12s/step - loss: 2.0326 - accuracy: 0.1328 - val_loss: 1.9719 - val_accuracy: 0.1935

Epoch 7/10

16/16 [] - 196s 12s/step - loss: 2.0326 - accuracy: 0.1328 - val_loss: 1.9719 - val_accuracy: 0.1935

Epoch 8/10

16/16 [] - 196s 12s/step - loss: 2.0326 - accuracy: 0.1328 - val_loss: 1.9719 - val_accuracy: 0.1935

Epoch 9/10

16/16 [] - 194s 12s/step - loss: 2.0326 - accuracy: 0.1328 - val_loss: 1.9719 - val_accuracy: 0.1935

Epoch 10/10

16/16 [] - 194s 12s/step - loss: 2.0326 - accuracy: 0.1328 - val_loss: 1.9719 - val_accuracy: 0.1935

8.模型评估

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()

可以看到，训练集和测试集的准确率都均没有超过20%（模型应该是没有问题的，在其他训练集上取得过95+%的识别准确率），应该是迭代次数还远远不够。本文海贼王数据和部分思路来自（k同学啊）

如果觉得《【机器视觉】VGG-16实现海贼王人物识别》对你有帮助，请点赞、收藏，并留下你的观点哦！