keras 利用vgg16进行十分类模型训练_10分类模型训练

导入模块

import os
import numpy as np
import tensorflow as tf
import random
import seaborn as sns
import matplotlib.pyplot as plt
 
from keras.models import Sequential, Model
from keras.layers import Dense, Dropout, Activation, Flatten, Input
from keras.layers.convolutional import Conv2D, MaxPooling2D
from keras.optimizers import RMSprop, Adam, SGD
from keras.preprocessing import image
from keras.preprocessing.image import ImageDataGenerator
from keras.utils import np_utils
from keras.applications.vgg16 import VGG16, preprocess_input
from sklearn.model_selection import train_test_split

读取图片函数

def read_and_process_image(data_dir,width=32, height=32, channels=3, preprocess=False):
    
    train_classes= [data_dir +  i for i in os.listdir(data_dir) ]
    train_images = []
    for train_class in train_classes:
        train_images= train_images + [train_class + "/" + i for i in os.listdir(train_class)]
    
    random.shuffle(train_images)
    
    def read_image(file_path, preprocess):
        img = image.load_img(file_path, target_size=(height, width))
        x = image.img_to_array(img)
        x = np.expand_dims(x, axis=0)
        
        if preprocess:
            x = preprocess_input(x)
        return x
    
    def prep_data(images, proprocess):
        count = len(images)
        data = np.ndarray((count, height, width, channels), dtype = np.float32)
        
        for i, image_file in enumerate(images):
            image = read_image(image_file, preprocess)
            data[i] = image
        
        return data
    
    def read_labels(file_path):
        labels = []
        for i in file_path:
            if 'airplane' in i:
                label = 0
            elif 'automobile' in i:
                label = 1
            elif 'bird' in i:
                label = 2
            elif 'cat' in i:
                label = 3
            elif 'deer' in i:
                label = 4
            elif 'dog' in i:
                label = 5
            elif 'frog' in i:
                label = 6
            elif 'horse' in i:
                label = 7
            elif 'ship' in i:
                label = 8
            elif 'truck' in i:
                label = 9
            labels.append(label)
        
        return labels
    
    X = prep_data(train_images, preprocess)
    labels = read_labels(train_images)
    
    assert X.shape[0] == len(labels)
    
    print("Train shape: {}".format(X.shape))
    
    return X, labels

读取训练集测试集

# 读取训练集图片
WIDTH = 48
HEIGHT = 48
CHANNELS = 3
X, y = read_and_process_image('D:/Python Project/cifar-10/train/',width=WIDTH, height=HEIGHT, channels=CHANNELS)
test_X, test_y = read_and_process_image('D:/Python Project/cifar-10/test/',width=WIDTH, height=HEIGHT, channels=CHANNELS)

查看训练集和测试集类型

# 统计y
sns.countplot(y)
sns.countplot(test_y)

将训练集和测试集标签one-hot编码

train_y = np_utils.to_categorical(y)
test_y = np_utils.to_categorical(test_y)

查看几张图片

# 显示图片
def show_picture(X, idx):
    plt.figure(figsize=(10,5), frameon=True)
    img = X[idx,:,:,::-1]
    img = img/255
    plt.imshow(img)
    plt.show()
 
for idx in range(0,3):
    show_picture(X, idx)

定义外接vgg网络

def vgg16_model(input_shape= (HEIGHT,WIDTH,CHANNELS)):
    vgg16 = VGG16(include_top=False, weights='imagenet',input_shape=input_shape)
    
    for layer in vgg16.layers:
        layer.trainable = False
    last = vgg16.output
    # 后面加入自己的模型
    x = Flatten()(last)
    x = Dense(256, activation='relu')(x)
    x = Dropout(0.5)(x)
    x = Dense(256, activation='relu')(x)
    x = Dropout(0.5)(x)
    x = Dense(num_classes, activation='softmax')(x)
    
    model = Model(inputs=vgg16.input, outputs=x)
    
    return model

创建模型

model_vgg16 = vgg16_model()
model_vgg16.summary()
model_vgg16.compile(loss='categorical_crossentropy',optimizer = Adam(0.0001), metrics = ['accuracy'])

训练模型

history = model_vgg16.fit(X,train_y, validation_data=(test_X, test_y),epochs=20,batch_size=100,verbose=True)
score = model_vgg16.evaluate(test_X, test_y, verbose=0)
print("Large CNN Error: %.2f%%" %(100-score[1]*100))