首先要把对自己的数据进行处理：
文件名为load_img.py
# -*- coding: utf-8 -*-
import numpy as np
import os
from PIL import Image
def load_data(file_dir,nums):
j = 0
i = 0
A0 = 0
A1 = 0
A2 = 0
A3 = 0
# 从data里面选择前number张作为总数据
number = nums
# 分为4类，从每个类的数量里面取前validation_split * number个样本
number1 = int(nums/4)
data = np.empty((number, 224, 224, 3), dtype="float32")
label = np.empty((number,), dtype="uint8")
for file in os.listdir(file_dir):
for file1 in os.listdir(file_dir + '/' + file):
for file2 in os.listdir(file_dir + '/' + file + '/' + file1):
for file3 in os.listdir(file_dir + '/' + file + '/' + file1 + '/' + file2):
dir = file_dir + '/' + file + '/' + file1 + '/' + file2 + '/' + file3
if 'C.jpg' in file3:
# print(Image.open(dir))
try:
image = Image.open(dir)
except IOError:
print(dir)
print ("Error: 没有找到文件或读取文件失败")
else:
if i == number:
j = j + 1
else:
if A0 == number1:
j = j + 1
else:
if image.mode == 'RGB':
# print(dir)
img = Image.open(dir)
# img= img.convert('L') # 3通道图转化为单通道图
img = img.resize((224, 224), Image.ANTIALIAS)
arr = np.asarray(img, dtype="float32")
data[i, :, :, :] = arr
label[i] = int(0)
A0 = A0 + 1
i = i + 1
if i % 500 == 0:
print(i)
else:
j = j + 1
elif 'CB.jpg' in file3:
# print(Image.open(dir))
try:
image = Image.open(dir)
except IOError:
print(dir)
print("Error: 没有找到文件或读取文件失败")
else:
if i == number:
j = j + 1
else:
if A3 == number1:
j = j + 1
else:
if image.mode == 'RGB':
# print(dir)
img = Image.open(dir)
# img = img.convert('L') # 3通道图转化为单通道图
img = img.resize((224, 224), Image.ANTIALIAS)
arr = np.asarray(img, dtype="float32")
data[i, :, :, :] = arr
label[i] = int(3)
A3 = A3 + 1
i = i + 1
if i % 500 == 0:
print(i)
else:
j = j + 1
elif 'CA.jpg' in file3:
# print(Image.open(dir))
try:
image = Image.open(dir)
except IOError:
print(dir)
print ("Error: 没有找到文件或读取文件失败")
else:
if i == number:
j = j + 1
else:
if A1 == number1:
j = j + 1
else:
if image.mode == 'RGB':
# print(dir)
img = Image.open(dir)
# img = img.convert('L') # 3通道图转化为单通道图
img = img.resize((224, 224), Image.ANTIALIAS)
arr = np.asarray(img, dtype="float32")
data[i, :, :, :] = arr
label[i] = int(1)
A1 = A1 + 1
i = i + 1
if i % 500 == 0:
print(i)
else:
j = j + 1
elif ('C.jpg' not in file3) and ('CA.jpg' not in file3) and ('CB.jpg' not in file3) and (
'.jpg' in file3):
# print(Image.open(dir))
try:
image = Image.open(dir)
except IOError:
print(dir)
print ("Error: 没有找到文件或读取文件失败")
else:
if i == number:
j = j + 1
else:
if A2 == number1:
j = j + 1
else:
if image.mode == 'RGB':
# print(dir)
img = Image.open(dir)
# img = img.convert('L') # 3通道图转化为单通道图
img = img.resize((224, 224), Image.ANTIALIAS)
arr = np.asarray(img, dtype="float32")
data[i, :, :, :] = arr
label[i] = int(2)
A2 = A2 + 1
i = i + 1
if (i % 500 == 0):
print(i)
else:
j = j + 1
else:
j = j + 1
print(i)
return data, label
第二部就要把自己的数据导入goodlenet网络里；
# -*- coding: utf-8 -*-
from __future__ import absolute_import
from __future__ import print_function
# 1. 导入各种模块
# 基本形式为：
# import 模块名
# from 某个文件 import 某个模块
# 指定GPU+固定显存
import os
import tensorflow as tf
os.environ["CUDA_VISIBLE_DEVICES"]="3" # 调用gpu哪块显卡
from keras.backend.tensorflow_backend import set_session
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.9 # 为显卡利用率设置阈值
set_session(tf.Session(config=config))
from PIL import Image # 图像读写
import numpy as np # 数组或者矩阵计算
from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential # 新建模型时用到
from keras.layers.advanced_activations import PReLU # 比较新的激活函数
from keras.optimizers import SGD, Adadelta, Adagrad # 各种优化器
from keras.utils import np_utils, generic_utils #标签二值化
from keras.utils.vis_utils import plot_model # 画出网络模型
from six.moves import range
from keras.layers import Input, Dense, Dropout, BatchNormalization, Conv2D, MaxPooling2D, AveragePooling2D,\
concatenate,Activation, Flatten
from keras.layers.convolutional import Conv2D, MaxPooling2D, AveragePooling2D
from load_img import load_data # 加载数据集，返回data，label
from keras.models import Model # 构建网络模型需要用的模块
from keras import backend as K # 去除历史计算图
from sklearn.cross_validation import train_test_split # 用于分类train ：test
from keras.wrappers import scikit_learn
from keras.callbacks import ModelCheckpoint
import copy
import types
np.random.seed(1337) # for reproducibility
# load and pre-process data
def preprocess_input(x):
return x.astype('float32').reshape((-1,) + input_shape) / 255
def preprocess_output(y):
return np_utils.to_categorical(y)
def Conv2d_BN(x, nb_filter, kernel_size, padding='same', strides=(1, 1), name=None):
if name is not None:
bn_name = name + '_bn'
conv_name = name + '_conv'
else:
bn_name = None
conv_name = None
x = Conv2D(nb_filter, kernel_size, padding=padding, strides=strides, activation='relu', name=conv_name)(x)
x = BatchNormalization(axis=3, name=bn_name)(x)
return x
def Inception(x, nb_filter):
branch1x1 = Conv2d_BN(x, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
branch3x3 = Conv2d_BN(x, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
branch3x3 = Conv2d_BN(branch3x3, nb_filter, (3, 3), padding='same', strides=(1, 1), name=None)
branch5x5 = Conv2d_BN(x, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
branch5x5 = Conv2d_BN(branch5x5, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
branchpool = MaxPooling2D(pool_size=(3, 3), strides=(1, 1), padding='same')(x)
branchpool = Conv2d_BN(branchpool, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
x = concatenate([branch1x1, branch3x3, branch5x5, branchpool], axis=3)
return x
def create_model(dropout):
input_shape = (224, 224, 3)
inputTensor = Input(input_shape)
# padding = 'same'，填充为(步长-1）/2,还可以用ZeroPadding2D((3,3))
# model = Sequential()
x = Conv2d_BN(inputTensor, 64, (7, 7), strides=(2, 2), padding='same')
# print('1 x_Conv2d_BN shape:', x.shape)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same')(x)
# print(' x_MaxPooling2D shape:', x.shape)
x = Conv2d_BN(x, 192, (3, 3), strides=(1, 1), padding='same')
# print('2 x_Conv2d_BN shape:', x.shape)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same')(x)
x = Inception(x, 64) # 256
# print('3 x_Inception shape:', x.shape)
x = Inception(x, 120) # 480
# print('4 x_Inception shape:', x.shape)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same')(x)
x = Inception(x, 128) # 512
# print('5 x_Inception shape:', x.shape)
x = Inception(x, 128)
# print('6 x_Inception shape:', x.shape)
x = Inception(x, 128)
# print('7 x_Inception shape:', x.shape)
x = Inception(x, 132) # 528
# print('8 x_Inception shape:', x.shape)
x = Inception(x, 208) # 832
# print('9 x_Inception shape:', x.shape)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same')(x)
x = Inception(x, 208)
# print('10 x_Inception shape:', x.shape)
x = Inception(x, 256) # 1024
# print('11 x_Inception shape:', x.shape)
x = AveragePooling2D(pool_size=(7, 7), strides=(7, 7), padding='same')(x)
x = Flatten(name='flatten')(x)
x = Dropout(dropout)(x)
# print('11 x_Dropout(0.4) shape:', x.shape)
# print('12 x_Flatten shape:', x.shape)
x = Dense(1000, activation='relu')(x)
# print('12 x_Dense_1 shape:', x.shape)
x1 = Dense(4, activation='softmax')(x)
# print('13 x_Dense_2 shape:', x1.shape)
model = Model(input=inputTensor, output=x1)
return model
if __name__ == '__main__':
# 对数据进行预处理
num_classes = 4 # number of classes
batch_size = 100
epochs = 40
# 加载数据
input_shape = (224, 224, 3)
if K.image_data_format() == 'channels_first':
input_shape = (3, 224, 224) # image shape
else:
input_shape = (224, 224, 3) # image shape
Data_Path = '/home/4T/2018_runpu_data/原始数据/河北数据4_20180409'
data, label = load_data(Data_Path, 40000)
label = np_utils.to_categorical(label, 4)
x_train, x_test, y_train, y_test = train_test_split(data, label, test_size=0.2, random_state=20)
print('Loading data...')
print('x_train shape:', x_train.shape, 'y_train shape:', y_train[1].shape)
print('x_test shape:', x_test.shape, 'y_test shape', y_test[1].shape)
save_dir = '/home/4T/wdd/keras_study/For_Classification/model'
model_name = 'pan_obj_googlenet_40000.h5'
# model_name='googlenet-ep{epoch:03d}-loss{loss:.3f}-val_loss{val_loss:.3f}.h5'
model_path = os.path.join(save_dir, model_name)
model = create_model(0.4)
model.compile(loss='categorical_crossentropy', optimizer='sgd', metrics=['accuracy'])
# 该回调函数将在最好的epoch后保存模型到model_path
checkpoint = ModelCheckpoint(model_name,
monitor='val_acc', verbose=1, save_best_only=True, mode='max')
model.fit(data, label, batch_size=batch_size, epochs=epochs, verbose=1,callbacks=[checkpoint],
validation_data=(x_test, y_test),shuffle=True, initial_epoch=0)
model.save('pan_obj_googlenet_40000.h5')
print('Saved trained model at %s ' % model_path)
K.clear_session()也可以测试不同的bach和epoch
# -*- coding: utf-8 -*-
from __future__ import absolute_import
from __future__ import print_function
# 1. 导入各种模块
# 基本形式为：
# import 模块名
# from 某个文件 import 某个模块
# 指定GPU+固定显存
import os
import tensorflow as tf
os.environ["CUDA_VISIBLE_DEVICES"]="1" # 调用gpu哪块显卡
from keras.backend.tensorflow_backend import set_session
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.9 # 为显卡利用率设置阈值
set_session(tf.Session(config=config))
from PIL import Image # 图像读写
import numpy as np # 数组或者矩阵计算
from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential # 新建模型时用到
from keras.layers.advanced_activations import PReLU # 比较新的激活函数
from keras.optimizers import SGD, Adadelta, Adagrad # 各种优化器
from keras.utils import np_utils, generic_utils
# from keras.utils.np_utils import to_categorical # 标签二值化
from keras.utils.vis_utils import plot_model # 画出网络模型
from six.moves import range
from keras.layers import Input, Dense, Dropout, BatchNormalization, Conv2D, MaxPooling2D, AveragePooling2D,\
concatenate,Activation, Flatten
from keras.layers.convolutional import Conv2D, MaxPooling2D, AveragePooling2D
from load_img import load_data # 加载数据集，返回data，label
from keras.models import Model # 构建网络模型需要用的模块
from keras import backend as K # 去除历史计算图
from sklearn.cross_validation import train_test_split # 用于分类train ：test
from sklearn.model_selection import GridSearchCV
from keras.wrappers import scikit_learn
from keras.wrappers.scikit_learn import KerasClassifier
from keras.callbacks import ModelCheckpoint,Callback
import copy
import types
np.random.seed(1337) # for reproducibility
class LossHistory(Callback):
def on_train_begin(self,logs={}):
self.losses = []
def on_batch_end(self, batch, logs={}):
self.losses.append(logs.get('loss'))
# load and pre-process data
def preprocess_input(x):
return x.astype('float32').reshape((-1,) + input_shape) / 255
def preprocess_output(y):
return np_utils.to_categorical(y)
def Conv2d_BN(x, nb_filter, kernel_size, padding='same', strides=(1, 1), name=None):
if name is not None:
bn_name = name + '_bn'
conv_name = name + '_conv'
else:
bn_name = None
conv_name = None
x = Conv2D(nb_filter, kernel_size, padding=padding, strides=strides, activation='relu', name=conv_name)(x)
x = BatchNormalization(axis=3, name=bn_name)(x)
return x
def Inception(x, nb_filter):
branch1x1 = Conv2d_BN(x, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
branch3x3 = Conv2d_BN(x, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
branch3x3 = Conv2d_BN(branch3x3, nb_filter, (3, 3), padding='same', strides=(1, 1), name=None)
branch5x5 = Conv2d_BN(x, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
branch5x5 = Conv2d_BN(branch5x5, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
branchpool = MaxPooling2D(pool_size=(3, 3), strides=(1, 1), padding='same')(x)
branchpool = Conv2d_BN(branchpool, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
x = concatenate([branch1x1, branch3x3, branch5x5, branchpool], axis=3)
return x
def create_model(dropout):
input_shape = (224, 224, 3)
inputTensor = Input(input_shape)
# padding = 'same'，填充为(步长-1）/2,还可以用ZeroPadding2D((3,3))
# model = Sequential()
x = Conv2d_BN(inputTensor, 64, (7, 7), strides=(2, 2), padding='same')
# print('1 x_Conv2d_BN shape:', x.shape)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same')(x)
# print(' x_MaxPooling2D shape:', x.shape)
x = Conv2d_BN(x, 192, (3, 3), strides=(1, 1), padding='same')
# print('2 x_Conv2d_BN shape:', x.shape)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same')(x)
x = Inception(x, 64) # 256
# print('3 x_Inception shape:', x.shape)
x = Inception(x, 120) # 480
# print('4 x_Inception shape:', x.shape)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same')(x)
x = Inception(x, 128) # 512
# print('5 x_Inception shape:', x.shape)
x = Inception(x, 128)
# print('6 x_Inception shape:', x.shape)
x = Inception(x, 128)
# print('7 x_Inception shape:', x.shape)
x = Inception(x, 132) # 528
# print('8 x_Inception shape:', x.shape)
x = Inception(x, 208) # 832
# print('9 x_Inception shape:', x.shape)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same')(x)
x = Inception(x, 208)
# print('10 x_Inception shape:', x.shape)
x = Inception(x, 256) # 1024
# print('11 x_Inception shape:', x.shape)
x = AveragePooling2D(pool_size=(7, 7), strides=(7, 7), padding='same')(x)
x = Flatten(name='flatten')(x)
x = Dropout(dropout)(x)
# print('11 x_Dropout(0.4) shape:', x.shape)
# print('12 x_Flatten shape:', x.shape)
x = Dense(1000, activation='relu')(x)
# print('12 x_Dense_1 shape:', x.shape)
x1 = Dense(4, activation='softmax')(x)
# print('13 x_Dense_2 shape:', x1.shape)
model = Model(input=inputTensor, output=x1)
model.compile(loss='categorical_crossentropy', optimizer='sgd', metrics=['accuracy'])
# model.summary() # 打印出模型概况，它实际调用的是keras.utils.print_summary
return model
if __name__ == '__main__':
# 对数据进行预处理
# num_classes = 4 # number of classes
# batch_size = 256
# epochs = 50
# 加载数据
input_shape = (224, 224, 3)
if K.image_data_format() == 'channels_first':
input_shape = (3, 224, 224) # image shape
else:
input_shape = (224, 224, 3) # image shape
Data_Path = '/home/4T/2018_runpu_data/原始数据/河北数据4_20180409'
data, label = load_data(Data_Path, 30000)
label = np_utils.to_categorical(label, 4)
x_train, x_test, y_train, y_test = train_test_split(data, label, test_size=0.2, random_state=20)
# x_train, x_test = map(preprocess_input, [x_train, x_test])
# y_train, y_test = map(preprocess_output,[y_train, y_test])
print('Loading data...')
print('x_train shape:', x_train.shape, 'y_train shape:', y_train[1].shape)
print('x_test shape:', x_test.shape, 'y_test shape', y_test[1].shape)
# 设置参数候选值
batch_size = [32, 64,100]
epochs = [20,40,60]
my_classifier = scikit_learn.KerasClassifier(build_fn=create_model,dropout=0.4)
# 创建GridSearchCV，并训练
param_grid = dict(batch_size=batch_size, epochs=epochs)
grid = GridSearchCV(estimator=my_classifier, param_grid=param_grid, n_jobs=1)
grid_result = grid.fit(x_train, y_train,validation_data=(x_test, y_test))
# 打印结果
print('Best: {} using {}'.format(grid_result.best_score_, grid_result.best_params_))
means = grid_result.cv_results_['mean_test_score']
stds = grid_result.cv_results_['std_test_score']
params = grid_result.cv_results_['params']
for mean, std, param in zip(means, stds, params):
print("%f (%f) with: %r" % (mean, std, param))
checkpointer = ModelCheckpoint(filepath="model/googlenet_best.hdf5", verbose=1, save_best_only=True)
history = LossHistory()
with open('log_googlenet_train.txt','w') as f:
f.write(str(history.history))
# model.fit(data, label, batch_size=32, epochs=1, verbose=1,validation_split=0.2,shuffle=True, initial_epoch=0)
# 神经网络模型可视化
# plot_model(model, to_file='model.png',show_shapes=True)
# 保存模型数据的文件
# 用于保存验证集误差最小的参数，当验证集误差减少时，立马保存下来
# save_dir = '/home/4T/wdd/keras_study/'
# model_name = 'pan_obj_googlenet_24000.h5'
# model_path = os.path.join(save_dir, model_name)
# model_1.save(model_path)
# print('Saved trained model at %s ' % model_path)
K.clear_session()
最后一步就是，训练好模型以后，怎嘛进行调用（c++也是可以调用的）
# -*- coding: utf-8 -*-
from __future__ import absolute_import
from __future__ import print_function
# 1. 导入各种模块
# 基本形式为：
# import 模块名
# from 某个文件 import 某个模块
# 指定GPU+固定显存
import os
import tensorflow as tf
os.environ["CUDA_VISIBLE_DEVICES"]="0" # 调用gpu哪块显卡
from keras.backend.tensorflow_backend import set_session
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.9 # 为显卡利用率设置阈值
#set_session(tf.Session(config=config))
from PIL import Image # 图像读写
import numpy as np # 数组或者矩阵计算
from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential # 新建模型时用到
from keras.layers.advanced_activations import PReLU # 比较新的激活函数
from keras.optimizers import SGD, Adadelta, Adagrad # 各种优化器
from keras.utils import np_utils, generic_utils
# from keras.utils.np_utils import to_categorical # 标签二值化
from keras.utils.vis_utils import plot_model # 画出网络模型
from six.moves import range
from keras.layers import Input, Dense, Dropout, BatchNormalization, Conv2D, MaxPooling2D, AveragePooling2D,\
concatenate,Activation, Flatten
from keras.layers.convolutional import Conv2D, MaxPooling2D, AveragePooling2D
#from load_img import load_data # 加载数据集，返回data，label
from keras.models import Model # 构建网络模型需要用的模块
from keras.models import load_model
from keras import backend as K # 去除历史计算图
from sklearn.cross_validation import train_test_split # 用于分类train ：test
from sklearn.model_selection import GridSearchCV
from keras.wrappers import scikit_learn
from keras.wrappers.scikit_learn import KerasClassifier
from keras.callbacks import ModelCheckpoint,Callback
import copy
import types
import cv2
np.random.seed(1337) # for reproducibility
# load and pre-process data
def preprocess_input(x):
return x.astype('float32').reshape((-1,) + input_shape) / 255
def preprocess_output(y):
return np_utils.to_categorical(y)
def Conv2d_BN(x, nb_filter, kernel_size, padding='same', strides=(1, 1), name=None):
if name is not None:
bn_name = name + '_bn'
conv_name = name + '_conv'
else:
bn_name = None
conv_name = None
x = Conv2D(nb_filter, kernel_size, padding=padding, strides=strides, activation='relu', name=conv_name)(x)
x = BatchNormalization(axis=3, name=bn_name)(x)
return x
def Inception(x, nb_filter):
branch1x1 = Conv2d_BN(x, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
branch3x3 = Conv2d_BN(x, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
branch3x3 = Conv2d_BN(branch3x3, nb_filter, (3, 3), padding='same', strides=(1, 1), name=None)
branch5x5 = Conv2d_BN(x, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
branch5x5 = Conv2d_BN(branch5x5, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
branchpool = MaxPooling2D(pool_size=(3, 3), strides=(1, 1), padding='same')(x)
branchpool = Conv2d_BN(branchpool, nb_filter, (1, 1), padding='same', strides=(1, 1), name=None)
x = concatenate([branch1x1, branch3x3, branch5x5, branchpool], axis=3)
return x
def create_model(dropout):
input_shape = (224, 224, 3)
inputTensor = Input(input_shape)
# padding = 'same'，填充为(步长-1）/2,还可以用ZeroPadding2D((3,3))
# model = Sequential()
x = Conv2d_BN(inputTensor, 64, (7, 7), strides=(2, 2), padding='same')
# print('1 x_Conv2d_BN shape:', x.shape)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same')(x)
# print(' x_MaxPooling2D shape:', x.shape)
x = Conv2d_BN(x, 192, (3, 3), strides=(1, 1), padding='same')
# print('2 x_Conv2d_BN shape:', x.shape)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same')(x)
x = Inception(x, 64) # 256
# print('3 x_Inception shape:', x.shape)
x = Inception(x, 120) # 480
# print('4 x_Inception shape:', x.shape)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same')(x)
x = Inception(x, 128) # 512
# print('5 x_Inception shape:', x.shape)
x = Inception(x, 128)
# print('6 x_Inception shape:', x.shape)
x = Inception(x, 128)
# print('7 x_Inception shape:', x.shape)
x = Inception(x, 132) # 528
# print('8 x_Inception shape:', x.shape)
x = Inception(x, 208) # 832
# print('9 x_Inception shape:', x.shape)
x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same')(x)
x = Inception(x, 208)
# print('10 x_Inception shape:', x.shape)
x = Inception(x, 256) # 1024
# print('11 x_Inception shape:', x.shape)
x = AveragePooling2D(pool_size=(7, 7), strides=(7, 7), padding='same')(x)
x = Flatten(name='flatten')(x)
x = Dropout(dropout)(x)
# print('11 x_Dropout(0.4) shape:', x.shape)
# print('12 x_Flatten shape:', x.shape)
x = Dense(1000, activation='relu')(x)
# print('12 x_Dense_1 shape:', x.shape)
x1 = Dense(4, activation='softmax')(x)
# print('13 x_Dense_2 shape:', x1.shape)
model = Model(inputs=inputTensor, output=x1)
model.compile(loss='categorical_crossentropy', optimizer='sgd', metrics=['accuracy'])
# model.summary() # 打印出模型概况，它实际调用的是keras.utils.print_summary
return model
def load_testdata(test_image):
testImg = Image.open(test_image)
testImg = testImg.resize((224, 224), Image.ANTIALIAS)
imgArr = np.asarray(testImg, dtype="float32")
testData = np.empty((1, 224, 224, 3), dtype="float32")
testData[0, :, :,:] = imgArr
return testData
if __name__ == '__main__':
# 对数据进行预处理
num_classes = 4 # number of classes
batch_size = 100
epochs = 10
# 加载数据
input_shape = (224, 224, 3)
#if K.image_data_format() == 'channels_first':
if K.image_dim_ordering() == 'th':
input_shape = (3, 224, 224) # image shape
else:
input_shape = (224, 224, 3) # image shape
# Data_Path = '/home/4T/2018_runpu_data/原始数据/河北数据4_20180409'
# #data, label = load_data(Data_Path, 200)
# label = np_utils.to_categorical(label, 4)
# x_train, x_test, y_train, y_test = train_test_split(data, label, test_size=0.2, random_state=20)
# # x_train, x_test = map(preprocess_input, [x_train, x_test])
# # y_train, y_test = map(preprocess_output,[y_train, y_test])
# print('Loading data...')
# print('x_train shape:', x_train.shape, 'y_train shape:', y_train[1].shape)
# print('x_test shape:', x_test.shape, 'y_test shape', y_test[1].shape)
model = create_model(dropout=0.4)
# model.fit(data, label, batch_size=32, epochs=1, verbose=1,validation_split=0.2,shuffle=True, initial_epoch=0)
#model.fit(x_train, y_train,batch_size = batch_size,epochs=epochs,verbose=1,validation_data=(x_test, y_test),shuffle=True)
# 保存模型数据的文件
# save_dir = '/home/4T/wdd/keras_study/'
# model_name = 'pan_obj_googlenet_24000.h5'
# model_path = os.path.join(save_dir, model_name)
# model.save(model_path)
#print('Saved trained model at %s ' % model_path)
# 神经网络模型可视化
# plot_model(model, to_file='model.png',show_shapes=True)
test_image = '/home/4T/2018_runpu_data/测试样本抽取/12/1/裁定书/（2017）冀0109财保10号/2CB.jpg'
#src=cv2.imread(test_image)
#cv2.imshow("test image: ",src)
model_path='/home/4T/2018_runpu_data/pan_obj_googlenet_40000.h5'
model=load_model(model_path)
# model.load_weights(model_path)
testData=load_testdata(test_image)
yy=model.predict(testData, batch_size=1, verbose=0)
result=np.argmax(yy)
print('待测的图片是 : ',result)
K.clear_session()
这里这能进行一张一张的测试，进行批测试，需要修改一下代码
8009
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
