如何解决使用任何CNN模型包括VGG16进行二进制分类时,验证精度均达到50%
我是CNN的新手,我遇到了二进制分类问题,验证准确度为50%。
为便于记录,我正在尝试将脑部扫描的MRI归类为阿尔茨海默氏病和健康对照。
默认情况下,图片为黑色和白色.png格式的图片为250x250px,我总共大约有1000张图片。
我尝试创建自己的模型并实现VGG16,所有这些都使验证精度达到了50%左右。
我开始认为这可能是数据处理不正确的问题,但是由于不确定,因此我不确定。也许与黑白图像被解释为RGB有关吗?但老实说,我不确定。
将感谢任何有兴趣的人,谢谢!
import tensorflow as tf
from keras.layers.convolutional import Convolution2D
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Activation,Dense,Flatten,BatchNormalization,Conv2D,MaxPool2D,Dropout
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.preprocessing.image import ImageDataGenerator
import os
import warnings
import matplotlib.pyplot as plt
from tensorflow.python.keras.applications.vgg16 import VGG16
from tensorflow.python.keras.layers import ZeroPadding2D,MaxPooling2D
warnings.simplefilter(action='ignore',category=FutureWarning)
os.chdir('C:/Users/dancu/PycharmProjects/firstCNN/data/ad-vs-cn')
physical_devices = tf.config.experimental.list_physical_devices('GPU')
print("Num GPUs Available: ",len(physical_devices))
tf.config.experimental.set_memory_growth(physical_devices[0],True)
# Define paths for image data
train_path = "C:/Users/dancu/PycharmProjects/firstCNN\data/ad-vs-cn/train"
test_path = "C:/Users/dancu/PycharmProjects/firstCNN\data/ad-vs-cn/test"
valid_path = "C:/Users/dancu/PycharmProjects/firstCNN\data/ad-vs-cn/valid"
# Use ImageDataGenerator to create 3 lots of batches
train_batches = ImageDataGenerator(
rescale=1/255).flow_from_directory(directory=train_path,target_size=(64,64),classes=['cn','ad'],batch_size=20,color_mode="rgb")
valid_batches = ImageDataGenerator(
rescale=1/255).flow_from_directory(directory=valid_path,color_mode="rgb")
test_batches = ImageDataGenerator(
rescale=1/255).flow_from_directory(directory=test_path,target_size=(256,240),batch_size=10,color_mode="rgb")
imgs,labels = next(train_batches)
# Test to see normalisation has occurred properly
print(imgs[1][16])
print(labels)
# Define method to plot MRIs
def plotImages(images_arr):
fig,axes = plt.subplots(1,10,figsize=(20,20))
axes = axes.flatten()
for img,ax in zip( images_arr,axes):
ax.imshow(img)
ax.axis('off')
plt.tight_layout()
plt.show()
# Plot a sample of MRIs
plotImages(imgs)
# Define the model
# VGG16
model = Sequential()
model.add(Conv2D(input_shape=(64,64,3),filters=64,kernel_size=(3,padding="same",activation="relu"))
model.add(Conv2D(filters=64,activation="relu"))
model.add(MaxPool2D(pool_size=(2,2),strides=(2,2)))
model.add(Conv2D(filters=128,activation="relu"))
model.add(Conv2D(filters=128,2)))
model.add(Conv2D(filters=256,activation="relu"))
model.add(Conv2D(filters=256,2)))
model.add(Conv2D(filters=512,activation="relu"))
model.add(Conv2D(filters=512,2)))
model.add(Flatten())
model.add(Dense(units=4096,activation="relu"))
model.add(Dense(units=4096,activation="relu"))
model.add(Dense(units=2,activation="softmax"))
# This model hits around 70% train acc,50% val acc
# model = Sequential([
# Conv2D(filters=32,activation='relu',padding = 'same',input_shape=(64,3)),# MaxPool2D(pool_size=(2,strides=2),# Dropout(0.2),# # BatchNormalization(),# Conv2D(filters=64,padding='same'),# Dropout(0.3),# # BatchNormalization(),# Conv2D(filters=128,# Dropout(0.4),# Flatten(),# Dense(units=2,activation='softmax')
# ])
## This model hits around 68% training accuracy at it's peak
# base_model = Sequential([
# Conv2D(filters=32,input_shape=(256,256,# Dropout(0.1),activation='softmax')
# ])
# Summarise each layer of the model
print(model.summary())
# Compile and train the model
model.compile(optimizer=Adam(),loss='categorical_crossentropy',metrics=['accuracy'])
model.fit(x=train_batches,steps_per_epoch=len(train_batches),validation_data=valid_batches,validation_steps=len(valid_batches),epochs=35,verbose=1
)
编辑:到目前为止,感谢大家的答复,他们的见解都很深刻。 我认为我们得出的结论是,这种情况的样本量很小,而且这些图像很难使用2D CNN进行分类。明天,我将尝试使用原始的.nii文件将基本的3D CNN放在一起,看看是否可以提高准确性。
解决方法
重复所有图像3次以符合RGB 3通道。如有损失,请使用binary_crossentropy
。在测试图像生成器中,您有(256,240),使其与您的训练大小相同。
也请尝试使用网络的原始大小(224x224)。
,您只有2个类别:是和否。因此,我建议仅通过“ Sigmoid”激活来生成一个通道的输出,即网络的最后一层应该是:
Dense(units=1,activation='sigmoid')
然后使用“ binary_crossentropy”训练您的网络。首先使用简单的小型CNN。
,在训练和验证生成器中,图像大小仅为64 X 64,而原始大小为250 X250。减少图像丢失的细节,网络可以从中学习。我建议您将这些值设置为224 X 224。 您有1000张图像,可能不足以获得较高的准确性。您可能想通过使用图像数据生成器中可用的扩充来扩充数据。文档为here.。在测试生成器中,应使图像大小与用于训练的大小相同。当您使用VGG16时,请确保将weights = imagenet设置为224 X224。您正在执行二进制分类,因此我建议顶层为Dense(units = 1,activation ='sigmoid')。然后,您可以测试输出是否高于或低于.5。编译模型时,请使用loss =“ binary_crossentropy”。还要设置top = False和pooling = max,然后添加密集层。我不使用Vgg16,因为它具有超过4000万个可训练的参数。如果您训练完整模型,则训练时间可能会很长。我更喜欢使用Mobilenet,它只有400万个参数,而且本质上也一样准确。
版权声明:本文内容由互联网用户自发贡献,该文观点与技术仅代表作者本人。本站仅提供信息存储空间服务,不拥有所有权,不承担相关法律责任。如发现本站有涉嫌侵权/违法违规的内容, 请发送邮件至 dio@foxmail.com 举报,一经查实,本站将立刻删除。