keras 二分类预测结果几乎全是一个值

Question

程序是用来对蜜蜂(bee)和黄蜂(wasp)分类的，用的模型是在vgg16上拼接的，代码如下

from keras.applications.vgg16 import VGG16
from keras.layers import Dense, Flatten, Activation, Dropout
from keras.models import Sequential
from keras.preprocessing.image import ImageDataGenerator
import keras
import shutil
import os

def creatDataGenerator(train_dir, test_dir):
    train_data_generator = ImageDataGenerator(rescale=.1/255)
    test_data_generator = ImageDataGenerator(rescale=.1/255)

    train_generator = train_data_generator.flow_from_directory(train_dir,
                                                            target_size=(150,150),
                                                            batch_size=32,
                                                            class_mode='binary')
    test_generator = test_data_generator.flow_from_directory(test_dir,
                                                            target_size=(150,150),
                                                            batch_size=32,
                                                            class_mode='binary')
    return train_generator, test_generator

vgg_model = VGG16(weights='imagenet', include_top=False, input_shape=(150,150,3))

cla_model = Sequential()
cla_model.add(Flatten())
cla_model.add(Dense(512, activation='relu'))
cla_model.add(Dropout(0.5))
cla_model.add(Dense(1, activation='sigmoid'))

model = Sequential()
model.add(vgg_model)
model.add(cla_model)

model.compile(loss='binary_crossentropy', optimizer='RMSprop', metrics=['accuracy'])

train_generator, test_generator = creatDataGenerator(train_dir=r'C:\Users\ayana\.keras\datasets\bee-vs-wasp\train',
                                                    test_dir=r'C:\Users\ayana\.keras\datasets\bee-vs-wasp\test')
H = model.fit(train_generator,
              steps_per_epoch=50,
              epochs=30,
              validation_data=test_generator,
              validation_steps=50)

然后训练以后进行预测，选择的是黄蜂的10张图（蜜蜂预测出来也是同样的结果）

顺便训练的准确率也比较低，不到0.6，也一直不知道怎么能高一些

from keras.preprocessing.image import load_img, img_to_array
import numpy as np

def predict(i):
    img_path = os.listdir(r'C:\Users\ayana\.keras\datasets\bee-vs-wasp\test\wasp')[i]
    img = load_img(path='C:\\Users\\ayana\\.keras\\datasets\\bee-vs-wasp\\test\\wasp\\'+img_path, 
                    target_size=(150,150))
    img = np.expand_dims(img, axis=0)/255
    prediction = model.predict(img)
    return prediction

for i in range(10):
    print(predict(i))

#>>>[[0.4714901]]
#    [[0.4714901]]
#    [[0.4714901]]
#    [[0.4714901]]
#    [[0.4714901]]
#    [[0.4714901]]
#    [[0.4714901]]
#    [[0.4714901]]
#    [[0.4714901]]
#    [[0.4714901]]

再用np.argmax()的话就都是0了

被困了一天了，#求救

Answer 1

应该是这个吧，你训练之后得到的模型是H，prediction = H.predict(img)