CNN-kersa，基本步骤

本文记录用keras的cnn进行图像分类。原始数据为图片

基本流程

1. 加载图像数据
2. 图像数据预处理
3. 训练模型
4. 保存模型与模型的可视化
5. 训练过程可视化

代码结构

整个过程可以在一个.py文件中完成，思路按照上面的流程

# 导包
# 
from keras.utils import to_categorical
from sklearn.model_selection import train_test_split
from PIL import Image
# 必须，可视化用得到
import numpy as np
import os


# 1. 读取图片并转成numpy类型函数
def read_image(imageName):
    im = Image.open(imageName).convert('L')
    data = np.array(im)
    return data

# 读取在data里面有几个文件夹
text = os.listdir('./data')

images = []
labels = []
# 把文件夹里面的图片和其对应的故障类型对应
for textPath in text:
    for fn in os.listdir(os.path.join('data', textPath)):
        if fn.endswith('.png'):
            fd = os.path.join('./data', textPath, fn)
            images.append(read_image(fd))
            labels.append(textPath)

X = np.array(images)
y = np.array(list(map(int, labels)))

# 按三七分把这些数据分为测试集和训练集
train_X, test_X, train_y, test_y = train_test_split(X, y, test_size=0.30, random_state=30)

# 将原始的特征矩阵做数据处理形成模型需要的数据
train_X = train_X.reshape(-1, 64, 64, 1)
train_X = train_X.astype('float32')
train_X /= 255
# 使用keras中的方法对数字的标签分类做One-Hot编码
train_y = to_categorical(train_y, 10)

# 将原始的特征矩阵做数据处理形成模型需要的数据
test_X = test_X.reshape(-1, 64, 64, 1)
test_X = test_X.astype('float32')
test_X /= 255
# 使用keras中的方法对数字的标签分类做One-Hot编码
test_y = to_categorical(test_y, 10)

from keras.models import Sequential
from keras.layers import Conv2D, MaxPool2D, Flatten, Dropout, Dense
from keras.losses import categorical_crossentropy
from keras.optimizers import Adadelta

model = Sequential()
model.add(Conv2D(32, (5,5), activation='relu', input_shape=[64, 64, 1]))
model.add(Conv2D(64, (5,5), activation='relu'))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Flatten())
model.add(Dropout(0.5))
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(10, activation='softmax'))

model.compile(loss=categorical_crossentropy,
             optimizer=Adadelta(),
             metrics=['accuracy'])

batch_size = 8      # 一次训练的样本数
epochs = 16         # 所有样本遍历次数
model.fit(train_X, train_y,
         batch_size=batch_size,
         epochs=epochs)

loss, accuracy = model.evaluate(test_X, test_y, verbose=1)
print('loss:%.4f accuracy:%.4f' %(loss, accuracy))

保存模型

model.summary()

if not os.path.isdir(save_dir) :
    os.makedirs(save_dir)
model_path = os.path.join(save_dir, model_name)
model.save(model_path)
print('Saved trained model at %s' % model_path)

可视化

import matplotlib.pyplot as plt  
# 绘制训练 & 验证的准确率值  
plt.plot(history.history['acc'])  
plt.plot(history.history['val_acc'])  
plt.title('Model accuracy')  
plt.ylabel('Accuracy')  
plt.xlabel('Epoch')  
plt.legend(['Train', 'Valid'], loc='upper left')  
plt.savefig('tradition_cnn_valid_acc.png')  
plt.show()  
   
 # 绘制训练 & 验证的损失值  
plt.plot(history.history['loss'])  
plt.plot(history.history['val_loss'])  
plt.title('Model loss')  
plt.ylabel('Loss')  
plt.xlabel('Epoch')  
plt.legend(['Train', 'Valid'], loc='upper left')  
plt.savefig('tradition_cnn_valid_loss.png')  
plt.show()  

历史网络结构

04091

model = Sequential()
model.add(Conv2D(32, (5,5), activation='relu', input_shape=[64, 64, 1]))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Dropout(0.5))

model.add(Conv2D(32, (5,5), activation='relu', input_shape=[64, 64, 1]))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Dropout(0.5))

model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(10, activation='softmax'))

04092

百度的LeCNN，准确度不错了

model.add(Conv2D(32,(5,5),strides=(1,1),input_shape=(64,64,1),padding='valid',activation='relu',kernel_initializer='uniform'))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Conv2D(64,(5,5),strides=(1,1),padding='valid',activation='relu',kernel_initializer='uniform'))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Flatten())
model.add(Dense(128,activation='relu'))
model.add(Dense(10,activation='softmax'))

Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 60, 60, 32)        832       
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 30, 30, 32)        0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 26, 26, 64)        51264     
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 13, 13, 64)        0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 10816)             0         
_________________________________________________________________
dense_1 (Dense)              (None, 128)               1384576   
_________________________________________________________________
dense_2 (Dense)              (None, 10)                1290      
=================================================================
Total params: 1,437,962
Trainable params: 1,437,962

后面

尝试改小图片？
尝试新的结构，比如上面的加入dropout

翻译论文的方法

epoch：18，38

L1 Conv（5×5×32）
L2 Maxpool（2×2）
L3 Conv（3×3×64）
L4 Maxpool（2×2）
L5 Conv（3×3×128）
L6 Maxpool（2×2）
L7 Conv（3×3×256）
L8 Maxpool（2×2）

64*64

model.add(Conv2D(32,(5,5),strides=(1,1),input_shape=(64,64,1),padding='valid',activation='relu',kernel_initializer='uniform'))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Conv2D(64,(3,3),strides=(1,1),padding='valid',activation='relu',kernel_initializer='uniform'))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Conv2D(128,(3,3),strides=(1,1),padding='valid',activation='relu',kernel_initializer='uniform'))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Conv2D(256,(3,3),strides=(1,1),padding='valid',activation='relu',kernel_initializer='uniform'))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Flatten())
model.add(Dense(2560,activation='relu'))
model.add(Dense(10,activation='softmax'))

实验效果不佳

32x32新方法

减少了第一个卷积层的维度，32-》16

model = Sequential()
model.add(Conv2D(16,(5,5),strides=(1,1),input_shape=(32,32,1),padding='valid',activation='relu',kernel_initializer='uniform'))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Conv2D(32,(3,3),strides=(1,1),padding='valid',activation='relu',kernel_initializer='uniform'))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Conv2D(64,(3,3),strides=(1,1),padding='valid',activation='relu',kernel_initializer='uniform'))
model.add(MaxPool2D(pool_size=(2,2)))

model.add(Flatten())
model.add(Dense(256,activation='relu'))
model.add(Dense(10,activation='softmax'))