基于Keras的卷积神经网络用于猫狗分类(进行了数据增强)+卷积层可视化

接着我上一篇博客,https://blog.csdn.net/fanzonghao/article/details/81149153

在上一篇基础上对数据集进行数据增强。函数如下:

"""
查看图像增强是否发生作用
"""
def see_pic_aug():
    train_datagen = ImageDataGenerator(
        rescale=1./255,
        rotation_range=40,
        width_shift_range=0.2,
        height_shift_range=0.2,
        shear_range=0.2,
        zoom_range=0.2,
        horizontal_flip=True,
        fill_mode='nearest')
    # 从训练集例返回图片的地址
    train_dir, validation_dir, cat_img_files, dog_img_files = data_read.read_data()
    # 返回随机一张图片的地址
    img_path = random.choice(cat_img_files + dog_img_files)
    img = load_img(img_path, target_size=(150, 150))
    x = img_to_array(img)
    # 变成(1,150,150,3)
    x = x.reshape((1,) + x.shape)
    i = 0
    for batch in train_datagen.flow(x, batch_size=1):
        plt.figure(i)
        plt.imshow(array_to_img(batch[0]))
        i += 1
        if i % 5 == 0:
            break
    plt.show()

打印5张查看:

确实发生了一些改变。

下面就用数据增强的样本训练模型,代码如下:

import numpy as np
import matplotlib.pyplot as plt
import random
import data_read
import tensorflow as tf
from keras.models import Model
from keras import  layers,optimizers
from keras import backend as K
from keras.preprocessing.image import ImageDataGenerator,img_to_array,load_img,array_to_img
"""
获得所需求的图片--进行了图像增强
"""
def data_deal_overfit():
    # 获取数据的路径
    train_dir, validation_dir, next_cat_pix, next_dog_pix = data_read.read_data()
    #图像增强
    train_datagen=ImageDataGenerator(
        rescale=1./255,
        rotation_range=40,
        width_shift_range=0.2,
        height_shift_range=0.2,
        shear_range=0.2,
        zoom_range=0.2,
        horizontal_flip=True,
        fill_mode='nearest')
    test_datagen=ImageDataGenerator(rescale=1./255)
    #从文件夹获取所需要求的图片
    train_generator=train_datagen.flow_from_directory(
          train_dir,
          target_size=(150,150),
          batch_size=20,
          class_mode='binary')
    test_generator = test_datagen.flow_from_directory(
        validation_dir,
        target_size=(150, 150),
        batch_size=20,
        class_mode='binary')
    return train_generator,test_generator
"""
定义模型并加入了dropout
"""
def define_model():
#定义TF backend session
    # tf_config=tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True))
    # K.set_session(tf.Session(config=tf_config))
    #卷积过程 三层卷积
    img_input=layers.Input(shape=(150,150,3))
    x=layers.Conv2D(filters=16,kernel_size=(3,3),activation='relu')(img_input)
    print('第一次卷积尺寸={}'.format(x.shape))
    x=layers.MaxPooling2D(strides=(2,2))(x)
    print('第一次池化尺寸={}'.format(x.shape))
    x=layers.Conv2D(filters=32,kernel_size=(3,3),activation='relu')(x)
    print('第二次卷积尺寸={}'.format(x.shape))
    x=layers.MaxPooling2D(strides=(2,2))(x)
    print('第二次池化尺寸={}'.format(x.shape))
    x=layers.Conv2D(filters=64,kernel_size=(3,3),activation='relu')(x)
    print('第三次卷积尺寸={}'.format(x.shape))
    x=layers.MaxPooling2D(strides=(2,2))(x)
    print('第三次池化尺寸={}'.format(x.shape))
    #全连接层
    x=layers.Flatten()(x)
    x=layers.Dense(512,activation='relu')(x)
    output=layers.Dense(1,activation='sigmoid')(x)
    model=Model(inputs=img_input,outputs=output,name='CAT_DOG_Model')
    return img_input,model
"""
训练模型
"""
def train_model():
    #构建网络模型
    img_input,model=define_model()
    #编译模型
    model.compile(optimizer=optimizers.RMSprop(lr=0.001),loss='binary_crossentropy',metrics=['accuracy'])
    train_generator,test_generator=data_deal_overfit()
    #verbose:日志显示,0为不在标准输出流输出日志信息,1为输出进度条记录,2为每个epoch输出一行记录
    #训练模型 返回history包含各种精度和损失
    history=model.fit_generator(
        train_generator,
        steps_per_epoch=100,#2000 images=batch_szie*steps
        epochs=1,
        validation_data=test_generator,
        validation_steps=50,#1000=20*50
        verbose=2)
    # 模型参数个数
    model.summary()
    #精度
    acc=history.history['acc']
    val_acc=history.history['val_acc']
    #损失
    loss=history.history['loss']
    val_loss=history.history['val_loss']
    #epochs的数量
    epochs=range(len(acc))

    plt.plot(epochs,acc)
    plt.plot(epochs, val_acc)
    plt.title('training and validation accuracy')

    plt.figure()
    plt.plot(epochs, loss)
    plt.plot(epochs, val_loss)
    plt.title('training and validation loss')
    plt.show()

    #测试图片
    # 从训练集例返回图片的地址
    train_dir, validation_dir, cat_img_files, dog_img_files = data_read.read_data()
    # 返回随机一张图片的地址
    img_path = random.choice(cat_img_files + dog_img_files)
    img = load_img(img_path, target_size=(150, 150))
    plt.imshow(img)
    plt.show()
    x = img_to_array(img)
    # 变成(1,150,150,3)
    x = x.reshape((1,) + x.shape)
    y_pred=model.predict(x)
    print('预测值y={}'.format(y_pred))

#图形化形式查看卷积层生成的图
def visualize_model():
    img_input,model=define_model()
    # print(model.layers)
    #存储每一层的tensor的shape 类型等
    successive_outputs=[layer.output for layer in model.layers]
    # print(successive_outputs)
    visualization_model=Model(img_input,successive_outputs)
    #从训练集例返回图片的地址
    train_dir, validation_dir, cat_img_files,dog_img_files = data_read.read_data()
    #返回随机一张图片的地址
    img_path=random.choice(cat_img_files+dog_img_files)
    img=load_img(img_path,target_size=(150,150))
    x=img_to_array(img)
    #print(x.shape)
    #变成(1,150,150,3)
    x=x.reshape((1,)+x.shape)
    x/=255
    #(samples,150,150,3) 存储10层的信息
    successive_feature_maps=visualization_model.predict(x)
    # print(len(successive_feature_maps))
    # for i in range(len(successive_feature_maps)):
    #     print(successive_feature_maps[i].shape)

    layer_names=[layer.name for layer in model.layers]
    #zip 打包成一个个元组以列表形式返回[(),()]
    #并且遍历元组里的内容
    for layer_name,feature_map in zip(layer_names,successive_feature_maps):
        if len(feature_map.shape)==4:#只查看卷积层
            n_features=feature_map.shape[-1]#(1,150,150,3)取3 取出深度
            size=feature_map.shape[1]##(1,150,150,3)取150  尺寸大小
            display_grid=np.zeros((size,size*n_features))
            for i in range(n_features):
                x=feature_map[0,:,:,i]
                x-=x.mean()
                x/=x.std()
                x*=64
                x+=128
                #限定x的值大小 小于0 则为0  大于255则为255
                x=np.clip(x,0,255).astype('uint8')
                display_grid[:,i*size:(i+1)*size]=x
            #显示
            scale=64./n_features
            plt.figure(figsize=(scale*n_features,scale))
            plt.title(layer_name)
            plt.grid(False)
            plt.imshow(display_grid,aspect='auto',cmap='Oranges')
    plt.show()
"""
查看图像增强是否发生作用
"""
def see_pic_aug():
    train_datagen = ImageDataGenerator(
        rescale=1./255,
        rotation_range=40,
        width_shift_range=0.2,
        height_shift_range=0.2,
        shear_range=0.2,
        zoom_range=0.2,
        horizontal_flip=True,
        fill_mode='nearest')
    # 从训练集例返回图片的地址
    train_dir, validation_dir, cat_img_files, dog_img_files = data_read.read_data()
    # 返回随机一张图片的地址
    img_path = random.choice(cat_img_files + dog_img_files)
    img = load_img(img_path, target_size=(150, 150))
    x = img_to_array(img)
    # 变成(1,150,150,3)
    x = x.reshape((1,) + x.shape)
    i = 0
    #32个训练样本
    for batch in train_datagen.flow(x, batch_size=32):
        plt.figure(i)
        plt.imshow(array_to_img(batch[0]))
        i += 1
        if i % 5 == 0:
            break
    plt.show()
if __name__ == '__main__':
    # see_pic_aug()
    train_model()
    #visualize_model()
    # 像素缩小到0~1

迭代100次结果:可看出相比上一篇文章,精度是稳定的,损失值也几乎是稳定的,数据增强还是起了防止过拟合的作用。

同样可视化卷积层:

猜你喜欢

转载自blog.csdn.net/fanzonghao/article/details/81172899