把文件vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5放入C:\Users\DELL.keras\models目录下，本文采用VGG16模型训练好的权重和偏置值搭建卷积神经网络，其中没有更改卷积层和池化层模型结构，使用两层神经网络简单识别data目录下的五种图片进行分类。
from tensorflow.python.keras.applications.vgg16 import VGG16, preprocess_input
from tensorflow import keras
from tensorflow.python.keras.preprocessing.image import ImageDataGenerator, load_img, img_to_array
import tensorflow as tf
import numpy as np
# 读取本地图片
class TransferModel(object):
def __init__(self):
# 定义测试和图片的变化方式
self.train_generator = ImageDataGenerator(rescale=1.0 / 255.0)
self.test_generator = ImageDataGenerator(rescale=1.0 / 255.0)
# 指定训练数据和测试数据的目录
self.train_dir = "./data/train"
self.test_dir = "./data/text"
# 定义图片网络大小参数
self.image_size = (224, 224)
self.batch_size = 32
# 定义迁移学习的类型模型,不包含全连接层模型加载
self.base_model = VGG16(include_top=False)
self.lable_dict = {
'0': 'bus',
'1': 'cat',
'2': 'dog',
'3': 'flower',
'4': 'tree'
}
def get_local_data(self):
# 读取本地数据
# 获取数据集
train_gen = self.train_generator.flow_from_directory(self.train_dir, shuffle=True,target_size=self.image_size, class_mode='binary', batch_size=self.batch_size)
# 训练数据集
test_gen = self.test_generator.flow_from_directory(self.test_dir, shuffle=True, target_size=self.image_size, class_mode='binary', b atch_size=self.batch_size)
return train_gen, test_gen
def refine_base_model(self):
# 微调VGG,减少迁移学习参数数量
# 获取原先的notop输出，在输出后面增加结构定义新的迁移学习模型
x = self.base_model.outputs[0]
x = keras.layers.GlobalAveragePooling2D()(x)
x = keras.layers.Dense(1024, activation=tf.nn.relu)(x)
y_prediction = keras.layers.Dense(5, activation=tf.nn.softmax)(x)
five_model = keras.models.Model(inputs=self.base_model.inputs, outputs=y_prediction)
return five_model
def freeze_model(self):
# 冻结Vgg模型，根据数据量来判断,获取所有车=层，返回层的列表
for layers in self.base_model.layers:
layers.trainable = False
def compile(self, model):
model.compile(optimizer=keras.optimizers.Adam(), loss=keras.losses.sparse_categorical_crossentropy,
metrics=["accuracy"])
return None
def fit_generator(self, model, train_gen, test_gen):
# 每一次迭代的准确数
modelckpt = keras.callbacks.ModelCheckpoint(
'D:/PYCHARM ITEM/vgg16pretrain/ckpt/transfer_{epoch:02d}-{val_acc:.2f}.h5',
monitor='val_acc',
save_weights_only=True,
save_best_only=True,
mode='auto',
period=1)
model.fit_generator(train_gen, epochs=3, validation_data=test_gen, callbacks=[modelckpt])
return None
def predict(self, model):
# 加载模型
model.load_weights("./ckpt/transfer_02-0.73.h5")
# 读取图片
image = load_img("D:/PYCHARM ITEM/vgg16pretrain/data/text/flower/src=http___img.redocn.com_sheying_20160618_yiduohongsexiaohuaduo_6507369.jpg&refer=http___img.redocn.jpg",
target_size=(224, 224))
image = img_to_array(image)
img = image.reshape([1, image.shape[0], image.shape[1], image.shape[2]])
# model.predict()
# 预测结果进行处理
image = preprocess_input(img)
predictions = model.predict(image)
res = np.argmax(predictions, axis=1)
print(self.lable_dict[str(res[0])])
# def predict():
# model = VGG16()
# print(model.summary())
if __name__ == '__main__':
tm = TransferModel()
# 训练
# train_gen, test_gen = tm.get_local_data()
# # print(train_gen)
# # print(test_gen)
# # for data in train_gen:
# # print(data[0].shape, data[1].shape)
# # print(tm.base_model.summary())
# # print(tm.refine_base_model())
# model = tm.refine_base_model()
# tm.freeze_model()
# tm.compile(model)
# tm.fit_generator(model, train_gen, test_gen)
# print(model.summary())
# 测试
model = tm.refine_base_model()
tm.predict(model)
Original: https://blog.csdn.net/weixin_43723578/article/details/121518659
Author: 如若是说
Title: 利用VGG16卷积神经网络模型数据做五种图片检测
原创文章受到原创版权保护。转载请注明出处：https://www.johngo689.com/62416/
转载文章受原作者版权保护。转载请注明原作者出处！
