Question

我正在使用keras和tensorflow学习CNN。我已经完成了图像的所有预处理，例如将图像转换为数组，调整大小并将其转换为灰度图像。之后，我将数据输入到模型中，但精度始终为50％。我做错什么了吗？

我还尝试过调整随机种子，批大小，减小数据集大小，并且还与优化程序一起使用，但没有任何帮助

count_dog=0
count_cat=0
#Reading cats and converting them from rgb to black and white
print('-------------------------')
for i in range(5000):
    try:
        caturl=r"C:/Users/Gowtam/Downloads/Compressed/PetImages/Cat/"""
        caturl=caturl+str(i)+'.jpg'
        img = cv2.imread(caturl,cv2.IMREAD_GRAYSCALE)
        img = cv2.resize(img,(50,50))
        imglist.append(img_to_array(img))
    except Exception as e:
        print(str(e))
        count_cat+=1


#Reading dogs and converting them frrom rgb to black and white    
for i in range(5000):
    try:
        caturl=r"C:/Users/Gowtam/Downloads/Compressed/PetImages/Dog/"""
        caturl=caturl+str(i)+'.jpg'
        img = cv2.imread(caturl,cv2.IMREAD_GRAYSCALE)
        img = cv2.resize(img,(50,50))
        imglist.append(img_to_array(img))
    except Exception as e:
        print(str(e))
        count_dog+=1

print(count_cat,'',count_dog)    

#now we have RGB to GREYSCALE converted images| lets create labels
print(len(imglist))

labels=[]

for i in range(5000-count_cat):
    labels.append("cat")
for i in range(5000-count_dog):
    labels.append("dog")
#one hot encoding..!
labelencoder = LabelEncoder()
labels50=(labelencoder.fit_transform(labels))

#now lets scramble the data(imgs) and labelencoder in same way

shuffimgs=shuffle(imglist,random_state=20)
shufflabels=shuffle(labels50,random_state=20)



#input shape of image (width,height,depth) depth means how many colors or channels that is 1 as b&W

#now lets normalize the shuffeled and resized imgs

fData=tf.keras.utils.normalize(shuffimgs)

#now lets build the model
#fData = fData[0].reshape(28,28,1)


model = Sequential()
#add model layers
model.add(Conv2D(32, (3,3), activation='relu', input_shape=(50,50,1)))
model.add(MaxPooling2D(2,2))
model.add(Conv2D(64, (3,3), activation='relu'))
model.add(MaxPooling2D(2,2))
model.add(Conv2D(128, (3,3), activation='relu'))
model.add(MaxPooling2D(2,2))
model.add(Flatten())
model.add(Dense(2, activation=tf.nn.softmax))

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

model.fit(fData,labels50,epochs=5)

Answer 1

尝试增加纪元数，也可以使用VGG16等预建模型。最后这是keras创作者的教程

https://blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html

Answer 2

我给您的第一个建议是，当它是二进制分类时，它使用二进制交叉熵来弥补损失，因为作为Cat Vs Dog，您必须将其分类为一个或另一个，然后仅分类两个。激活我建议使用S形，因为您需要两个值或0或1。尝试这种方式。

model = Sequential()
model.add(Conv2D(32,(3,3),activation='relu',input_shape=(150,150,3)))
model.add(MaxPooling2D((2,2)))
model.add(Conv2D(64,(3,3),activation='relu'))
model.add(MaxPooling2D((2,2)))
model.add(Conv2D(128,(3,3),activation='relu'))
model.add(MaxPooling2D((2,2)))
model.add(Conv2D(128,(3,3),activation='relu'))
model.add(MaxPooling2D((2,2)))
model.add(Flatten())
model.add(Dropout(0.5))
model.add(Dense(512,activation='relu'))
model.add(Dense(1,activation='sigmoid'))

model.summary()

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

train_datagen = ImageDataGenerator(
    rescale=1./255,
    rotation_range=40,
    width_shift_range=0.2,
    height_shift_range=0.2,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True,
    )
test_datagen = ImageDataGenerator(rescale=1./255)

train_generator = train_datagen.flow_from_directory(
    train_directory,
    target_size=(150,150),
    batch_size=20,
    class_mode='binary'
)

validation_generator = test_datagen.flow_from_directory(
    validation_directory,
    target_size=(150,150),
    batch_size=20,
    class_mode='binary'
)

Tensorflow-Keras CNN图像分类器的准确性始终为50％

2 个答案: