Keras实现U-net

最新推荐文章于 2024-06-03 08:56:19 发布

Grack_skw

最新推荐文章于 2024-06-03 08:56:19 发布

阅读量770

点赞数

分类专栏： Keras 深度学习

本文链接：https://blog.csdn.net/qq_35565669/article/details/106728781

版权

这篇博客介绍了如何使用Keras实现U-Net网络，包括网络结构的详细说明，并提供了模型的概览及可能遇到的plot_model问题的解决办法。

摘要由CSDN通过智能技术生成

U-net论文地址：https://arxiv.org/pdf/1505.04597v1.pdf

U-net的网络结构如图

import tensorflow as tf
from keras.layers import *
import numpy as np
import keras
from keras.models import *
from keras.optimizers import *


def unet(pretrained_weight=None, input_size=(512, 512, 3)):
    input = Input(shape=input_size)
    conv1 = Conv2D(64, kernel_size=3, padding='same', activation='relu', kernel_initializer='he_normal', name='encoder_conv1_1')(input)
    conv1 = Conv2D(64, kernel_size=3, padding='same', activation='relu', kernel_initializer='he_normal', name='encoder_conv1_2')(conv1)
    maxpool1 = MaxPooling2D((2, 2), name='encoder_maxpool1')(conv1)

    conv2 = Conv2D(128, kernel_size=3, padding='same', activation='relu', kernel_initializer='he_normal', name='encoder_conv2_1')(maxpool1)
    conv2 = Conv2D(128, kernel_size=3, padding='same', activation='relu', kernel_initializer='he_normal', name='encoder_conv2_2')(conv2)
    maxpool2 = MaxPooling2D((2, 2), name='encoder_maxpool2')(conv2)

    conv3 = Conv2D(256, kernel_size=3, padding='same', activation='relu', kernel_initializer='he_normal', name='encoder_conv3_1')(maxpool2)
    conv3 = Conv2D(256, kernel_size=3, padding='same', activation='relu', kernel_initializer='he_normal', name='encoder_conv3_2')(conv3)
    maxpool3 = MaxPooling2D((2, 2), name='encoder_maxpool3')(conv3)

    conv4 = Conv2D(512, kernel_size=3, padding='same', activation='relu', kernel_initializer='he_normal', name='encoder_conv4_1')(maxpool3)
    conv4 = Conv2D(512, kernel_size=3, padding='same', activation='relu', kernel_initializer='he_normal', name='encoder_conv4_2')(conv4)
    maxpool4 = MaxPooling2D((2, 2), name='encoder_maxpool4')(conv4)

    conv5 = Conv2D(1024, kernel_size=3, padding='same', activation='relu', kernel_initializer='he_normal', name='encoder_conv5_1')(maxpool4)
    conv5 = Conv2D(1024, kernel_size=3, padding='same', activation='relu', kernel_initializer='he_normal', name='encoder_conv5_2')(conv5)
    # maxpool4=MaxPooling2D((2,2))(conv4)

    up1 = UpSampling2D(size=(2, 2), name='decoder_upsample1')(conv5)
    upconv1 = Conv2D(512, kernel_size=2, padding='same', activation='relu', kernel_initializer='he_normal', name='decoder_upconv1')(up1)
    upconv1 = concatenate([upconv1, conv4], axis=-1)
    upconv1 = Conv2D(512, kernel_size=3, padding='same', activation='relu', kernel_initializer='he_normal', name='decoder_conv1_1')(upconv1)
    upconv1 = Conv2D(512, kernel_size=3, padding='same', activation='relu', kernel_initializer='he_normal', name='decoder_conv1_2')(upconv1)

    up2 = UpSampling2D(size=(2, 2), name='decoder_upsample2')(upconv1)
    upconv2 = Conv2D(256, kernel_size=2, padding='same', activation=