https://arxiv.org/abs/1505.04597
https://www.seoxiehui.cn/article-106308-1.html
簡(jiǎn)介:介紹在圖像分割中，機(jī)器必須將圖像分割成不同的segments，每個(gè)segment代表不同的實(shí)體。圖像分割示例正如你在上面看到的，圖像如何變成兩個(gè)部分，一個(gè)代表貓，另一個(gè)代表背景。圖像分割在從自動(dòng)駕駛汽車到衛(wèi)星的許多 ...
介紹
在圖像分割中，機(jī)器必須將圖像分割成不同的segments，每個(gè)segment代表不同的實(shí)體。
圖像分割在從自動(dòng)駕駛汽車到衛(wèi)星的許多領(lǐng)域都很有用。也許其中最重要的是醫(yī)學(xué)影像。醫(yī)學(xué)圖像的微妙之處是相當(dāng)復(fù)雜的。一臺(tái)能夠理解這些細(xì)微差別并識(shí)別出必要區(qū)域的機(jī)器，可以對(duì)醫(yī)療保健產(chǎn)生深遠(yuǎn)的影響。
卷積神經(jīng)網(wǎng)絡(luò)在簡(jiǎn)單的圖像分割問題上取得了不錯(cuò)的效果，但在復(fù)雜的圖像分割問題上卻沒有取得任何進(jìn)展。這就是UNet的作用。UNet最初是專門為醫(yī)學(xué)圖像分割而設(shè)計(jì)的。該方法取得了良好的效果，并在以后的許多領(lǐng)域得到了應(yīng)用。在本文中，我們將討論UNet工作的原因和方式
UNet架構(gòu)
該架構(gòu)看起來像一個(gè)'U'。該體系結(jié)構(gòu)由三部分組成：contraction，bottleneck和expansion 部分。contraction部分由許多contraction塊組成。每個(gè)塊接受一個(gè)輸入，應(yīng)用兩個(gè)3X3的卷積層，然后是一個(gè)2X2的最大池化。在每個(gè)塊之后，核或特征映射的數(shù)量會(huì)加倍，這樣體系結(jié)構(gòu)就可以有效地學(xué)習(xí)復(fù)雜的結(jié)構(gòu)。最底層介于contraction層和expansion 層之間。它使用兩個(gè)3X3 CNN層，然后是2X2 up convolution層。
這種架構(gòu)的核心在于expansion 部分。與contraction層類似，它也包含幾個(gè)expansion 塊。每個(gè)塊將輸入傳遞到兩個(gè)3X3 CNN層，然后是2X2上采樣層。此外，卷積層使用的每個(gè)塊的feature map數(shù)量得到一半，以保持對(duì)稱性。每次輸入也被相應(yīng)的收縮層的 feature maps所附加。這個(gè)動(dòng)作將確保在contracting 圖像時(shí)學(xué)習(xí)到的特征將被用于重建圖像。expansion 塊的數(shù)量與contraction塊的數(shù)量相同。之后，生成的映射通過另一個(gè)3X3 CNN層，feature map的數(shù)量等于所需的segment的數(shù)量。
UNet中的損失計(jì)算
UNet對(duì)每個(gè)像素使用了一種新穎的損失加權(quán)方案，使得分割對(duì)象的邊緣具有更高的權(quán)重。這種損失加權(quán)方案幫助U-Net模型以不連續(xù)的方式分割生物醫(yī)學(xué)圖像中的細(xì)胞，以便在binary segmentation map中容易識(shí)別單個(gè)細(xì)胞。
首先，在所得圖像上應(yīng)用pixel-wise softmax，然后是交叉熵?fù)p失函數(shù)。所以我們將每個(gè)像素分類為一個(gè)類。我們的想法是，即使在分割中，每個(gè)像素都必須存在于某個(gè)類別中，我們只需要確保它們可以。因此，我們只是將分段問題轉(zhuǎn)換為多類分類問題，與傳統(tǒng)的損失函數(shù)相比，它表現(xiàn)得非常好。
def unet(pretrained_weights = None,input_size = (256,256,1)):
inputs = Input(input_size)
conv1 = Conv2D(64, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(inputs)
conv1 = Conv2D(64, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv1)
pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
conv2 = Conv2D(128, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(pool1)
conv2 = Conv2D(128, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv2)
pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)
conv3 = Conv2D(256, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(pool2)
conv3 = Conv2D(256, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv3)
pool3 = MaxPooling2D(pool_size=(2, 2))(conv3)
conv4 = Conv2D(512, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(pool3)
conv4 = Conv2D(512, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv4)
drop4 = Dropout(0.5)(conv4)
pool4 = MaxPooling2D(pool_size=(2, 2))(drop4)
conv5 = Conv2D(1024, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(pool4)
conv5 = Conv2D(1024, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv5)
drop5 = Dropout(0.5)(conv5)
up6 = Conv2D(512, 2, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(UpSampling2D(size = (2,2))(drop5))
merge6 = concatenate([drop4,up6], axis = 3)
conv6 = Conv2D(512, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(merge6)
conv6 = Conv2D(512, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv6)
up7 = Conv2D(256, 2, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(UpSampling2D(size = (2,2))(conv6))
merge7 = concatenate([conv3,up7], axis = 3)
conv7 = Conv2D(256, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(merge7)
conv7 = Conv2D(256, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv7)
up8 = Conv2D(128, 2, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(UpSampling2D(size = (2,2))(conv7))
merge8 = concatenate([conv2,up8], axis = 3)
conv8 = Conv2D(128, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(merge8)
conv8 = Conv2D(128, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv8)
up9 = Conv2D(64, 2, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(UpSampling2D(size = (2,2))(conv8))
merge9 = concatenate([conv1,up9], axis = 3)
conv9 = Conv2D(64, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(merge9)
conv9 = Conv2D(64, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv9)
conv9 = Conv2D(2, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv9)
conv10 = Conv2D(1, 1, activation = 'sigmoid')(conv9)
model = Model(input = inputs, output = conv10)
model.compile(optimizer = Adam(lr = 1e-4), loss = 'binary_crossentropy', metrics = ['accuracy'])
#model.summary()
if(pretrained_weights):
model.load_weights(pretrained_weights)
return model
