Keras:Unet网络实现多类语义分割方式

Love The Way You Lie 2023-02-20 02:08 294阅读 0赞

更多python视频教程请到菜鸟教程https://www.piaodoo.com/

**1 介绍**

U-Net最初是用来对医学图像的语义分割，后来也有人将其应用于其他领域。但大多还是用来进行二分类，即将原始图像分成两个灰度级或者色度，依次找到图像中感兴趣的目标部分。

本文主要利用U-Net网络结构实现了多类的语义分割，并展示了部分测试效果，希望对你有用！

**2 源代码**

（1）训练模型

from __future__ import print_function
    import os
    import datetime
    import numpy as np
    from keras.models import Model
    from keras.layers import Input, concatenate, Conv2D, MaxPooling2D, Conv2DTranspose, AveragePooling2D, Dropout, \
     BatchNormalization
    from keras.optimizers import Adam
    from keras.layers.convolutional import UpSampling2D, Conv2D
    from keras.callbacks import ModelCheckpoint
    from keras import backend as K
    from keras.layers.advanced_activations import LeakyReLU, ReLU
    import cv2

PIXEL = 512 \#set your image size  
BATCH\_SIZE = 5  
lr = 0.001  
EPOCH = 100  
X\_CHANNEL = 3 \# training images channel  
Y\_CHANNEL = 1 \# label iamges channel  
X\_NUM = 422 \# your traning data number

pathX = ‘I:\\Pascal VOC Dataset\\train1\\images\\’ \#change your file path  
pathY = ‘I:\\Pascal VOC Dataset\\train1\\SegmentationObject\\’ \#change your file path

\#data processing  
def generator(pathX, pathY,BATCH\_SIZE):  
while 1:  
X\_train\_files = os.listdir(pathX)  
Y\_train\_files = os.listdir(pathY)  
a = (np.arange(1, X\_NUM))  
X = \[\]  
Y = \[\]  
for i in range(BATCH\_SIZE):  
index = np.random.choice(a)

# print(index) #

img = cv2.imread(pathX + X\_train\_files\[index\], 1)  
img = np.array(img).reshape(PIXEL, PIXEL, X\_CHANNEL)  
X.append(img)  
img1 = cv2.imread(pathY + Y\_train\_files\[index\], 1)  
img1 = np.array(img1).reshape(PIXEL, PIXEL, Y\_CHANNEL)  
Y.append(img1)

X = np.array(X)  
Y = np.array(Y)  
yield X, Y

\#creat unet network  
inputs = Input((PIXEL, PIXEL, 3))  
conv1 = Conv2D(8, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(inputs)  
pool1 = AveragePooling2D(pool\_size=(2, 2))(conv1) \# 16

conv2 = BatchNormalization(momentum=0.99)(pool1)  
conv2 = Conv2D(64, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv2)  
conv2 = BatchNormalization(momentum=0.99)(conv2)  
conv2 = Conv2D(64, 1, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv2)  
conv2 = Dropout(0.02)(conv2)  
pool2 = AveragePooling2D(pool\_size=(2, 2))(conv2) \# 8

conv3 = BatchNormalization(momentum=0.99)(pool2)  
conv3 = Conv2D(128, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv3)  
conv3 = BatchNormalization(momentum=0.99)(conv3)  
conv3 = Conv2D(128, 1, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv3)  
conv3 = Dropout(0.02)(conv3)  
pool3 = AveragePooling2D(pool\_size=(2, 2))(conv3) \# 4

conv4 = BatchNormalization(momentum=0.99)(pool3)  
conv4 = Conv2D(256, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv4)  
conv4 = BatchNormalization(momentum=0.99)(conv4)  
conv4 = Conv2D(256, 1, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv4)  
conv4 = Dropout(0.02)(conv4)  
pool4 = AveragePooling2D(pool\_size=(2, 2))(conv4)

conv5 = BatchNormalization(momentum=0.99)(pool4)  
conv5 = Conv2D(512, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv5)  
conv5 = BatchNormalization(momentum=0.99)(conv5)  
conv5 = Conv2D(512, 1, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv5)  
conv5 = Dropout(0.02)(conv5)  
pool4 = AveragePooling2D(pool\_size=(2, 2))(conv4)

# conv5 = Conv2D(35, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv4) #

# drop4 = Dropout(0.02)(conv5) #

pool4 = AveragePooling2D(pool\_size=(2, 2))(pool3) \# 2  
pool5 = AveragePooling2D(pool\_size=(2, 2))(pool4) \# 1

conv6 = BatchNormalization(momentum=0.99)(pool5)  
conv6 = Conv2D(256, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv6)

conv7 = Conv2D(256, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv6)  
up7 = (UpSampling2D(size=(2, 2))(conv7)) \# 2  
conv7 = Conv2D(256, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(up7)  
merge7 = concatenate(\[pool4, conv7\], axis=3)

conv8 = Conv2D(128, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(merge7)  
up8 = (UpSampling2D(size=(2, 2))(conv8)) \# 4  
conv8 = Conv2D(128, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(up8)  
merge8 = concatenate(\[pool3, conv8\], axis=3)

conv9 = Conv2D(64, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(merge8)  
up9 = (UpSampling2D(size=(2, 2))(conv9)) \# 8  
conv9 = Conv2D(64, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(up9)  
merge9 = concatenate(\[pool2, conv9\], axis=3)

conv10 = Conv2D(32, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(merge9)  
up10 = (UpSampling2D(size=(2, 2))(conv10)) \# 16  
conv10 = Conv2D(32, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(up10)

conv11 = Conv2D(16, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv10)  
up11 = (UpSampling2D(size=(2, 2))(conv11)) \# 32  
conv11 = Conv2D(8, 3, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(up11)

# conv12 = Conv2D(3, 1, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv11) #

conv12 = Conv2D(3, 1, activation=‘relu’, padding=‘same’, kernel\_initializer=‘he\_normal’)(conv11)

model = Model(input=inputs, output=conv12)  
print(model.summary())  
model.compile(optimizer=Adam(lr=1e-3), loss=‘mse’, metrics=\[‘accuracy’\])

history = model.fit\_generator(generator(pathX, pathY,BATCH\_SIZE),  
steps\_per\_epoch=600, nb\_epoch=EPOCH)  
end\_time = datetime.datetime.now().strftime(’%Y-%m-%d %H:%M:%S’)

\#save your training model  
model.save(r’V1\_828.h5’)

\#save your loss data  
mse = np.array((history.history\[‘loss’\]))  
np.save(r’V1\_828.npy’, mse)

（2）测试模型

from keras.models import load_model
    import numpy as np
    import matplotlib.pyplot as plt
    import os
    import cv2

model = load\_model(‘V1\_828.h5’)  
test\_images\_path = ‘I:\\Pascal VOC Dataset\\test\\test\_images\\’  
test\_gt\_path = ‘I:\\Pascal VOC Dataset\\test\\SegmentationObject\\’  
pre\_path = ‘I:\\Pascal VOC Dataset\\test\\pre\\’

X = \[\]  
for info in os.listdir(test\_images\_path):  
A = cv2.imread(test\_images\_path + info)  
X.append(A)

# i += 1 #

X = np.array(X)  
print(X.shape)  
Y = model.predict(X)

groudtruth = \[\]  
for info in os.listdir(test\_gt\_path):  
A = cv2.imread(test\_gt\_path + info)  
groudtruth.append(A)  
groudtruth = np.array(groudtruth)

i = 0  
for info in os.listdir(test\_images\_path):  
cv2.imwrite(pre\_path + info,Y\[i\])  
i += 1

a = range(10)  
n = np.random.choice(a)  
cv2.imwrite(‘prediction.png’,Y\[n\])  
cv2.imwrite(‘groudtruth.png’,groudtruth\[n\])  
fig, axs = plt.subplots(1, 3)

# cnt = 1 #

# for j in range(1): #

axs\[0\].imshow(np.abs(X\[n\]))  
axs\[0\].axis(‘off’)  
axs\[1\].imshow(np.abs(Y\[n\]))  
axs\[1\].axis(‘off’)  
axs\[2\].imshow(np.abs(groudtruth\[n\]))  
axs\[2\].axis(‘off’)

# cnt += 1 #

fig.savefig(“imagestest.png”)  
plt.close()

**3 效果展示**

说明：从左到右依次是预测图像，真实图像，标注图像。可以看出，对于部分数据的分割效果还有待改进，主要原因还是数据集相对复杂，模型难于找到其中的规律。

![20200611103733.jpg][]

以上这篇Keras:Unet网络实现多类语义分割方式就是小编分享给大家的全部内容了，希望能给大家一个参考，也希望大家多多

茂名论坛https://www.hnthzk.com/

化州橘红http://www.sfkyty.com/

茂名论坛http://www.nrso.net/

源码搜藏网http://www.cntkd.net/

茂名市高级技工学校(茂名一技)http://www.szsyby.net/

茂名一技https://www.mmbbu.com/