图像数据识别的模型

315 0 0

作者:夏至秋颜

模型参数设置与模型构建及训练

from keras.models import Sequential
from keras.layers import Dense, Activation
from keras.callbacks import ModelCheckpoint

model = Sequential()
model.add(Dense(units=64, input_dim=100))
model.add(Activation("relu"))
model.add(Dense(units=64, input_dim=100))
model.add(Activation("softmax"))
#完成模型的搭建后，我们需要使用.compile()方法来编译模型：
model.compile(loss='categorical_croosentropy',metrics=['accuracy'])
model.fit(x_train, y_train, epochs=5, batch_size=32)
loss_and_metrics = model.evaluate(x_test, y_test, batch_size=128)
classes = model.predict(x_test, batch_size=128)
model.save('my_model.h5')

#更改loss函数和优化器
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy])

checkpointer = ModelCheckpoint(filepath="checkpoint-{epoch:02d}e-val_acc_{val_acc:2f}.hdf5"
,save_best_only=true, verbose=1, period=50)

model.fit(data,labels, epoch=10,batch_size=32, callbacks=[checkpointer])

#调用Checkpoint保存的model
model = load_model('checkpoint-05e-val_acc_0.58.hdf5')


#模型选取
from keras.application.vgg16 import VGG16
from keras.application.vgg19 import VGG19
from keras.application.inception_v3 import InceptionV3
from keras.application.resnet50 import ResNet50

model_vgg16_conv = VGG16(weights=None, include_top=False, pooling='avg')
output_vgg16_conv = model_vgg16_conv(input)
x = output_vgg16_conv
input = Input(shape=(width,height,channel),name='image_input')
x = Dense(clazz, activation='softmax', name='predictions')(x)

#Create your own model
model = Model(inputs=input, outputs=x)
model.complie(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adam(lr=lr,decay=0),metrics=['acc])

#load all Images
def LoadImageGen(files_data, labels_data,batch=32, label="label"):
      start = 0
      while start < len(file_data):
            stop = start + batch
            if stop > len(files_data):
                  stop = len(file_data)
            imgs = []
            labels = []
            for i in range(start, stop):
                  imgs.append(LoadImage(file_data[i]))
                  labels.append(label_data[i])
            yield(np.array(imgs),np.array(labels))
            if start + batch < len(files_data):
                  start +=batch
            else:
                  zip_data = list(zip(files_data,labels_data))
                  random.shuffle(zip_data)
                  files_data, labels_data = zip(*zip_data)
                  start=0
# load Images to training model
model.fit_generator(
      LoadImageGen(train_x,train_y, batch=batch,label = "train"),
      steps_per_epoch=int(len(train_x)/batch),
      epochs = epoch,
      verbose = 1,
      validation_data = LoadImageGen(test_x,test_y, batch=batch,label = "test"),
      validation_steps = int(len(test_x)/batch),
      callbacks=[
            EarlyStopping(monitor='val_acc',patience=patienceEpoch)),
            modelCheckpoint
            ]
      )

VGG16:VGG(visual geometry group,超分辨率测试序列)

参考：https://zhuanlan.zhihu.com/p/41423739
共包含13卷积层(Convolutional Layer，表示为conv3-XXXX)+3个连接层(Fully connected Layer，表示为FC-XXXX)+5个池化层(Pool layer，表示maxpool)，VGG16的16代表权重系数，maxpool没有权重系数，故16=13+3.