keras分类猫狗数据（中）使用CNN分类模型

keras分类猫狗数据（上）数据预处理
keras分类猫狗数据（中）使用CNN分类模型
keras分类猫狗数据（下）迁移学习
keras分类猫狗数据（番外篇）深度学习CNN连接SVM分类

1 . 如下代码，数据增强，并根据文件夹创建数据流(我的保存为了catvsdogs/morph.py)。

from keras.preprocessing.image import ImageDataGenerator

train_dir="E:/MLdata/kaggle_Dogsvs.Cats/min_trainfordata/train"

test_dir="E:/MLdata/kaggle_Dogsvs.Cats/min_trainfordata/test"

train_pic_gen=ImageDataGenerator(rescale=1./255,rotation_range=20,width_shift_range=0.2,height_shift_range=0.2,
                                 shear_range=0.2,zoom_range=0.5,horizontal_flip=True,fill_mode='nearest')

test_pic_gen=ImageDataGenerator(rescale=1./255)


train_flow=train_pic_gen.flow_from_directory(train_dir,(128,128),batch_size=32,class_mode='binary')

test_flow=test_pic_gen.flow_from_directory(test_dir,(128,128),batch_size=32,class_mode='binary')

# print(train_flow.class_indices)

2 .构建模型并训练

from keras.models import Sequential
from keras.layers import Convolution2D,MaxPool2D,Flatten,Dense,Dropout
from keras.callbacks import TensorBoard

model=Sequential([
    Convolution2D(32,3,3,input_shape=(128,128,3),activation='relu'),
    MaxPool2D(pool_size=(2,2)),
    Convolution2D(64,3,3,input_shape=(128,128,3),activation='relu'),
    MaxPool2D(pool_size=(2,2)),
    Flatten(),
    Dense(64,activation='relu'),
    Dropout(0.5),
    Dense(1,activation='sigmoid')
])

model.summary()

model.compile(optimizer='rmsprop',loss='binary_crossentropy',metrics=['accuracy'])

import catvsdogs.morph as morph#引用上文1的数据增加代码

model.fit_generator(
    morph.train_flow,steps_per_epoch=100,epochs=50,verbose=1,validation_data=morph.test_flow,validation_steps=100,
    callbacks=[TensorBoard(log_dir='./logs/1')]
)

model.save('outputs/catdogs_model.h5')

3 .也可以将步骤2的模型数据重新加载继续训练。

from keras.callbacks import TensorBoard
from keras.models import load_model

model=load_model('outputs/catdogs_model.h5')

model.summary()

import catvsdogs.morph as morph

model.fit_generator(
    morph.train_flow,steps_per_epoch=100,epochs=50,verbose=1,validation_data=morph.test_flow,validation_steps=100,
    callbacks=[TensorBoard(log_dir='./logs/2')]
)

model.save('outputs/catdogs_model.h5')

4 . 以下加载模型数据并参与预测：

from keras import models
import numpy as np
import cv2

def get_inputs(src=[]):
    pre_x = []
    for s in src:
        input = cv2.imread(s)
        input = cv2.resize(input, (128, 128))
        input = cv2.cvtColor(input, cv2.COLOR_BGR2RGB)
        pre_x.append(input)  # input一张图片
    pre_x = np.array(pre_x) / 255.0
    return pre_x

def put_prey(pre_y,label):
    output=[]
    for y in pre_y:
        if y[0]<0.5:#二分类，此处只用一个神经元输出
            output.append([label[0],1-y[0]])
        else:
            output.append([label[1], y[0]])
    return output

model=models.load_model('outputs/catdogs_model.h5')

pre_x=get_inputs(['../data/rotman.jpg','../data/cat.jpg'])


pre_y=model.predict(pre_x)

import catvsdogs.morph as mp
output=put_prey(pre_y,list(mp.train_flow.class_indices.keys()))
print(output)

[[‘dogs’, 0.8959254], [‘cats’, 0.6214156150817871]]

测试所用图片：

5 .第二次训练验证数据正确率和损失如下：

可以看到准确率在0.78左右。