即使在具有高训练精度后，该模型也不能预测正确的事情，原因确实是一个问题，它不能很好地预测。假设将疟疾图像分为感染和未感染，但似乎没有做到这一点，该模型是使用cnn迁移学习开发的
from tensorflow.keras.layers import Input, Lambda, Dense, Flatten,Conv2D
from tensorflow.keras.models import Model
from tensorflow.keras.applications.vgg19 import VGG19
from tensorflow.keras.applications.resnet50 import preprocess_input
from tensorflow.keras.preprocessing import image
from tensorflow.keras.preprocessing.image import ImageDataGenerator,load_img
from tensorflow.keras.models import Sequential
import numpy as np
from glob import glob
import matplotlib.pyplot as plt
import tensorflow as tf
print(tf.__version__)
# re-size all the images to this
IMAGE_SIZE = [224, 224 , 3]
# Import the Vgg 16 library as shown below and add preprocessing layer to the front of VGG
# Here we will be using imagenet weights
mobilnet = VGG19(input_shape=IMAGE_SIZE , weights='imagenet', include_top=False)
from google.colab import drive
drive.mount('/content/drive')
# don't train existing weights
for layer in mobilnet.layers:
layer.trainable = False
# useful for getting number of output classes
# folders = glob('Dataset/Train/*')
enter code here
folders = glob('/content/drive/MyDrive/Dataset/Train/*')
folders
# our layers - you can add more if you want
x = Flatten()(mobilnet.output)
prediction = Dense(len(folders), activation='softmax')(x)
# create a model object
model = Model(inputs=mobilnet.input, outputs=prediction)
# view the structure of the model
model.summary()
from tensorflow.keras.layers import MaxPooling2D
# tell the model what cost and optimization method to use
model.compile(
loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy']
)
# Use the Image Data Generator to import the images from the dataset
from tensorflow.keras.preprocessing.image import ImageDataGenerator
train_datagen = ImageDataGenerator(rescale = 1./255,
shear_range = 0.2,
zoom_range = 0.2,
horizontal_flip = True)
test_datagen = ImageDataGenerator(rescale = 1./255)
# # Make sure you provide the same target size as initialied for the image size
training_set = train_datagen.flow_from_directory(directory='/content/drive/MyDrive/Dataset/Train',
target_size = (224, 224),
batch_size = 32,
class_mode = 'categorical')
training_set
test_set = test_datagen.flow_from_directory('/content/drive/MyDrive/Dataset/Test',
target_size = (224, 224),
batch_size = 32,
class_mode = 'categorical')
test_set
from keras.callbacks import ModelCheckpoint, EarlyStopping
# simple early stopping
es = EarlyStopping(monitor='val_loss', mode='min', verbose=1)
# fit the model
# Run the cell. It will take some time to execute
r = model.fit_generator(
training_set,
validation_data=test_set,
epochs= 20,
steps_per_epoch=len(training_set),
validation_steps=len(test_set),
# callbacks=[es]
)
# plot the loss
plt.plot(r.history['loss'], label='train loss')
plt.plot(r.history['val_loss'], label='val loss')
plt.legend()
plt.show()
plt.savefig('LossVal_loss')
# plot the accuracy
plt.plot(r.history['accuracy'], label='train acc')
plt.plot(r.history['val_accuracy'], label='val acc')
plt.legend()
plt.show()
plt.savefig('AccVal_acc')
# save it as a h5 file
from tensorflow.keras.models import load_model
model.save('model_vgg19.h5')
y_pred = model.predict(test_set)
y_pred
import numpy as np
y_pred = np.argmax(y_pred, axis=1)
y_pred
print(y_pred.shape)
from tensorflow.keras.models import load_model
from tensorflow.keras.preprocessing import image
model=load_model('model_vgg19.h5')
img=image.load_img('/content/drive/MyDrive/Dataset/Train/Parasite/C111.png',target_size=(224,224))
x=image.img_to_array(img)
x
x.shape
x=x/255
x=np.expand_dims(x,axis=0)
image_data = preprocess_input(x)
print(image_data.shape)
model.predict(image_data)
a=np.argmax(model.predict(image_data), axis=1)
print(a)
if(a==1):
print("Not infected")
else:
print("Infected")
请注意modelfit_generator正在折旧。model.fit现在可以处理生成器。每个历元的步骤实际上应该是len(training_set)/batch_size，验证步骤也是如此。在test_set中，flow_from_directory设置为shuffle=False。
您知道数据集的余额值吗？这是训练集中样本感染样本/未感染样本之间的比率。如果这是<。5你的模型将倾向于预测未被感染。例如，如果你有100个感染者样本和1000个未感染者样本，那么只要始终预测未感染者，你的模型的准确率将达到90%。我建议你修改你的提前停止回调如下
es=tf.keras.callbacks.EarlyStopping( monitor="val_loss", min_delta=0,
patience=4, verbose=1, mode="auto",
baseline=None, restore_best_weights=True)
增加你的纪元数量，比如说50，这样提前停止就会被激活。如果restore_best_weights=True，它将加载具有最低验证损失的epoch的权重的模型。我还建议您使用Keras回调ReduceLROnPlateau使用可调整的学习率。如果验证损失未能提高历元的耐心数，这将通过参数因子自动降低学习率。我建议的代码如下：
tf.keras.callbacks.ReduceLROnPlateau(monitor="val_loss", factor=0.5, patience=1,
verbose=1, mode="auto", min_delta=0.0001,
cooldown=0, min_lr=0)
在model.fit中设置callbacks=[es，rlronp]
您不是在训练Vgg模型，这很好，但您可能需要微调该模型。完成培训后，您可以进行
mobilnet.trainable=True
fine_tune_epochs=10
total_epochs=epochs + fine_tune_epochs
r=model.fit(training_set, validation_data=test_set, epochs= total_epochs,
steps_per_epoch=len(training_set)/batch_size,
validation_steps=len(test_set)/batch_siz,
callbacks=[es, rlronp], initial_epoch=epoch)
要查看您的模型是否真的表现良好，您应该创建一个混淆矩阵。在测试集上获得一组预测，如下
classes=['infected', 'normal']
import seaborn as sns
sns.set_style('darkgrid')
from sklearn.metrics import confusion_matrix, classification_report
y_pred=[]
preds=model.predict(train_set, batch_size=batch_size)
for p in preds:
y_pred.append(np.argmax(p))
y_true=test_set.labels # this is why you set shuffle=False for the train_set
y_true= np.array(labels)
y_pred=np.array(y_pred)
cm = confusion_matrix(y_true, y_pred )
clr = classification_report(y_true, y_pred, target_names=classes)
fig_width=8
fig_height=8
plt.figure(figsize=(fig_width, fig_height))
sns.heatmap(cm, annot=True, vmin=0, fmt='g', cmap='Blues', cbar=False)
plt.xticks(np.arange(length)+.5, classes, rotation= 90)
plt.yticks(np.arange(length)+.5, classes, rotation=0)
plt.xlabel("Predicted")
plt.ylabel("Actual")
plt.title("Confusion Matrix")
plt.show()
print("Classification Report:n----------------------n", clr)