卷积神经网络convolution_CIFAR10_keras_参数调优和图像增强代码

#!/usr/bin/env python # coding: utf-8 # In[1]: # 加载函数库 from keras.layers.convolutional import Conv2D from keras.layers.convolutional import MaxPooling2D from keras.layers.core import Activation, Dropout from keras.layers.core import Flatten from keras.layers.core import Dense from keras.datasets import mnist, cifar10 #from keras.datasets import cifar10 from keras.utils import np_utils,generic_utils from keras.preprocessing.image import ImageDataGenerator import keras from keras.optimizers import SGD, RMSprop, Adam import numpy as np import matplotlib.pyplot as plt from keras.models import Sequential from keras.utils.vis_utils import plot_model # In[2]: # 初始化模型参数 # 随机种子 np.random.seed(666) # 训练轮数 NB_EPOCH = 5 # batch size BATCH_SIZE = 256 # 四层神经元数量总数 N_ADDITION = 1024 #verbose：日志显示 #verbose = 0 为不在标准输出流输出日志信息 #verbose = 1 为输出进度条记录 #verbose = 2 为每个epoch输出一行记录 VERBOSE = 1 # 分类标签种类数量 NB_CLASSES = 10 #OPTIMIZER = SGD() # 隐藏层神经元数量 # 第一层 N_HIDDEN = 128 #validation_split用于在没有提供验证集的时候 #按一定比例从训练集中取出一部分作为验证集 VALIDATION_SPLIT = 0.2 # In[3]: data_generator = ImageDataGenerator(rotation_range=90, width_shift_range=0.15 ,height_shift_range=0.15, zoom_range=0.3 ,horizontal_flip=True ,fill_mode='nearest' ) # In[4]: # 准备数据 (X_train,y_train),(X_test,y_test) = cifar10.load_data() # convolution不需要做数据结构的修改 # 还是32*32， 不用转化为一维数组 # 数据类型 X_train = X_train.astype('float32') X_test = X_test.astype('float32') # 数据预处理 X_train /= 255 X_test /= 255 # 分类标签处理:转化为二值化序列 y_train = np_utils.to_categorical(y_train, NB_CLASSES) y_test = np_utils.to_categorical(y_test,NB_CLASSES) # 图片shape input_shape = (32,32,3) print(X_train.shape) # In[5]: # 构建卷积模型 model = Sequential() model.add(Conv2D(20, kernel_size=5, padding="same", input_shape=input_shape)) model.add(Activation("relu")) model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2), padding='valid')) # 增加Dropout层 # 在第一池化层/第二池化层/最后一个池化层的效果都不一样 # 第一个添加的效果最好 # 思考为什么？ model.add(Dropout(0.15)) # 添加规则化器的约束 model.add(Conv2D(50, kernel_size=3, padding="same", kernel_regularizer=keras.regularizers.l2(0.01))) model.add(Activation('relu')) model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2), padding='valid')) # 增加新的卷积层、池化层 model.add(Conv2D(20, kernel_size=5, padding="same", input_shape=input_shape)) model.add(Activation("relu")) model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2), padding='valid')) # 扁平化处理 model.add(Flatten()) model.add(Dense(500)) model.add(Activation("relu")) model.add(Dense(10)) model.add(Activation("softmax")) # In[6]: model.compile(loss='categorical_crossentropy', optimizer='sgd', metrics=['accuracy']) # 初始化回调函数 CSV_log = keras.callbacks.CSVLogger(filename='csv/cnn_CIFAR10_image.log', separator=',', append=False) # 训练模型 model.fit(X_train, y_train, batch_size=BATCH_SIZE, epochs=NB_EPOCH ,verbose=VERBOSE, validation_split=VALIDATION_SPLIT, callbacks=[CSV_log]) # In[ ]: # In[ ]: ''' # 编译模型 #model.compile(loss='categorical_crossentropy', optimizer='sgd', metrics=['accuracy']) model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy']) print("Training on Data Agumentation...") # 手动定义训练模型的训练，首先定义 for e in range(NB_EPOCH): print('EPOCH:{0}/{1}'.format(e+1, NB_EPOCH)) print('Training...') # 用于呈现每一轮的训练速度 progbar = generic_utils.Progbar(X_train.shape[0]) batch_size = 0 # 通过数据发生器以原始数据X_train，y_train为原型生成批评批量大小BATCH_SIZE数据 for x_batch,ybatch in data_generator.flow(X_train,y_train,batch_size=BATCH_SIZE): loss, train_acc = model.train_on_batch(x_batch,y_batch) batch_size += x_batch.shape[0] if batch_size > X_train.shape[0]: break # 每跑完一个批量，probar便累计 progbar.add(x_batch.shape[0], values[('train loss', loss), ('train acc', train_acc)]) ''' # In[ ]: ''' # 用原始数据中的测试集验证模型 loss, acc = model.evaluate(X_test, y_test, batch_size=32) print('Val Loss:', loss) print('Val Accuracy:', acc) # 初始化回调函数 CSV_log = keras.callbacks.CSVLogger(filename='csv/cnn_CIFAR10_param.log', separator=',', append=False) # 训练模型 model.fit(X_train, y_train, batch_size=BATCH_SIZE, epochs=NB_EPOCH ,verbose=VERBOSE, validation_split=VALIDATION_SPLIT, callbacks=[CSV_log]) ''' # In[7]: # 模型评分 score = model.evaluate(X_test, y_test, verbose=VERBOSE) print("test loss : ", score[0]) print("test accuracy : ", score[1]) # summary()用于显示模型详细信息 model.summary() # plot_model函数用于绘制神经网络模型结构 # 输入所需要绘制模型的model # 输出图片为model.png,并展示各网络层的大小 plot_model(model, to_file='model_convolution_CIFAR10_image.png',show_shapes=True) # In[8]: # 对日志进行抽取并显示 from pandas import DataFrame import pandas as pd import seaborn as sns import matplotlib.pyplot as plt # In[10]: # 读取各个不同的日志文件 path = 'csv\cnn_CIFAR10_image.log' data = pd.read_csv(path, sep=',') # 配置排版 plt.subplot() #绘制曲线 sns.lineplot(data=data[['accuracy','val_accuracy']]) # 设置标题 plt.title('model CNN_CIFAR10_image:' '\nbest acc is {0:.4}' '\nbest val_acc is {1:.4}'.format(data[['accuracy']].max().values[0], data[['val_accuracy']].max().values[0]) ) #plt.figure(figsize=(16, 12))344opo;[0--] #plt.tight_layout(pad=0.1, w_pad=1.0, h_pad=1.0) plt.tight_layout() plt.show() # In[ ]: # In[ ]: # In[ ]: # In[ ]: # In[ ]: # In[ ]: # In[ ]: