全卷积神经网路【U-net项目实战】U-Net源码上实现自己数据集的分割任务

环境：

win7+python 3.5+ tensorflow 1.9.0 + keras 2.2.4

参考博客： https://blog.csdn.net/Snowy_susu/article/details/88998936?depth_1-utm_source=distribute.pc_relevant.none-task&utm_source=distribute.pc_relevant.none-task
U-net源码讲解（Keras）源码解析很棒很细！

              全卷积神经网络图像分割(U-net)-keras实现代码根本来源于该博主的Github

               深度学习数据增强（data_augmentation）:Keras ImageDataGenerator : keras的图像增强讲解，方便理解源码函数；

               Keras中文文档——图片预处理 ： 无意中翻到的，学习keras可以留存。

目的 ： 利用U-Net （keras）实现自己数据集的分割任务，二类。

数据集准备：

样本： size： 256*256 位深为8，.png格式；

标签： size： 256*256 ，.png格式，使用photoshop软件标注控制位深为8，图片名于样本名字相同。

存储位置： 同一个根目录下的两个文件夹中，文件名为：image, label

测试图片：位深为8，

主要代码：

model.py:存放U-Net网络模型，keras写起来真的很简单！！

# model.py
import numpy as np 
import os
import skimage.io as io
import skimage.transform as trans
import numpy as np
from keras.models import *
from keras.layers import *
from keras.optimizers import *
from keras.callbacks import ModelCheckpoint, LearningRateScheduler
from keras import backend as keras
 
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
 
 

data.py:

# data.py 对图片的相关处理
from __future__ import print_function
from keras.preprocessing.image import ImageDataGenerator
import numpy as np 
import os
import glob
import skimage.io as io
import skimage.transform as trans
 
Sky = [128,128,128]
Building = [128,0,0]
Pole = [192,192,128]
Road = [128,64,128]
Pavement = [60,40,222]
Tree = [128,128,0]
SignSymbol = [192,128,128]
Fence = [64,64,128]
Car = [64,0,128]
Pedestrian = [64,64,0]
Bicyclist = [0,128,192]
Unlabelled = [0,0,0]
 
COLOR_DICT = np.array([Sky, Building, Pole, Road, Pavement,
                          Tree, SignSymbol, Fence, Car, Pedestrian, Bicyclist, Unlabelled])
 
 
#adjustData()函数主要是对训练集的数据和标签的像素值进行归一化
def adjustData(img,mask,flag_multi_class,num_class):
    if(flag_multi_class):#此程序中不是多类情况，所以不考虑这个
        img = img / 255.0
        mask = mask[:,:,:,0] if(len(mask.shape) == 4) else mask[:,:,0]
		# if else的简洁写法，一行表达式，为真时放在前面，不明白mask.shape=4的情况是什么，
		# 由于有batch_size，所以mask就有3维[batch_size,wigth,heigh],估计mask[:,:,0]是写错了，应该写成[0,:,:],这样可以得到一片图片，
        new_mask = np.zeros(mask.shape + (num_class,))
		# np.zeros里面是shape元组，此目的是将数据厚度扩展到num_class层，以在层的方向实现one-hot结构
 
        for i in range(num_class):
            #for one pixel in the image, find the class in mask and convert it into one-hot vector
            #index = np.where(mask == i)
            #index_mask = (index[0],index[1],index[2],np.zeros(len(index[0]),dtype = np.int64) + i) if (len(mask.shape) == 4) else (index[0],index[1],np.zeros(len(index[0]),dtype = np.int64) + i)
            #new_mask[index_mask] = 1
            new_mask[mask == i,i] = 1#将平面的mask的每类，都单独变成一层，
        new_mask = np.reshape(new_mask,(new_mask.shape[0],new_mask.shape[1]*new_mask.shape[2],new_mask.shape[3])) if flag_multi_class else np.reshape(new_mask,(new_mask.shape[0]*new_mask.shape[1],new_mask.shape[2]))
        mask = new_mask
    elif(np.max(img) > 1):
        img = img / 255.0
        mask = mask /255.0
        mask[mask > 0.5] = 1
        mask[mask <= 0.5] = 0
    return (img,mask)
 
 
# trainGenerator()函数主要是产生一个数据增强的图片生成器，方便后面使用这个生成器不断生成图片
def trainGenerator(batch_size,train_path,image_folder,mask_folder,aug_dict,image_color_mode = "grayscale",
                    mask_color_mode = "grayscale",image_save_prefix  = "image",mask_save_prefix  = "mask",
                    flag_multi_class = False,num_class = 2,save_to_dir = None,target_size = (256,256),seed = 1):
    '''
    can generate image and mask at the same time
    use the same seed for image_datagen and mask_datagen to ensure the transformation for image and mask is the same
    if you want to visualize the results of generator, set save_to_dir = "your path"
    '''
    image_datagen = ImageDataGenerator(**aug_dict)
    mask_datagen = ImageDataGenerator(**aug_dict)
    image_generator = image_datagen.flow_from_directory(
        train_path,#训练数据文件夹路径
        classes = [image_folder],#类别文件夹,对哪一个类进行增强
        class_mode = None,#不返回标签
        color_mode = image_color_mode,#灰度，单通道模式
        target_size = target_size,#转换后的目标图片大小
        batch_size = batch_size,#每次产生的（进行转换的）图片张数
        save_to_dir = save_to_dir,#保存的图片路径
        save_prefix  = image_save_prefix,#生成图片的前缀，仅当提供save_to_dir时有效
        seed = seed)
    mask_generator = mask_datagen.flow_from_directory(
        train_path,
        classes = [mask_folder],
        class_mode = None,
        color_mode = mask_color_mode,
        target_size = target_size,
        batch_size = batch_size,
        save_to_dir = save_to_dir,
        save_prefix  = mask_save_prefix,
        seed = seed)
    train_generator = zip(image_generator, mask_generator)#组合成一个生成器
    for (img,mask) in train_generator:
		#由于batch是2，所以一次返回两张，即img是一个2张灰度图片的数组，[2,256,256]
        img,mask = adjustData(img,mask,flag_multi_class,num_class)#返回的img依旧是[2,256,256]
        yield (img,mask)
		#每次分别产出两张图片和标签，不懂yield的请看https://blog.csdn.net/mieleizhi0522/article/details/82142856
 
 
# testGenerator()函数主要是对测试图片进行规范，使其尺寸和维度上和训练图片保持一致
def testGenerator(test_path,num_image = 30,target_size = (256,256),flag_multi_class = False,as_gray = True):
    for i in range(num_image):
        img = io.imread(os.path.join(test_path,"%d.png"%i),as_gray = as_gray)
        img = img / 255.0
        img = trans.resize(img,target_size)
        img = np.reshape(img,img.shape+(1,)) if (not flag_multi_class) else img
        img = np.reshape(img,(1,)+img.shape)
		#将测试图片扩展一个维度，与训练时的输入[2,256,256]保持一致
        yield img
 
 
# geneTrainNpy()函数主要是分别在训练集文件夹下和标签文件夹下搜索图片，
# 然后扩展一个维度后以array的形式返回，是为了在没用数据增强时的读取文件夹内自带的数据
def geneTrainNpy(image_path,mask_path,flag_multi_class = False,num_class = 2,image_prefix = "image",mask_prefix = "mask",image_as_gray = True,mask_as_gray = True):
    image_name_arr = glob.glob(os.path.join(image_path,"%s*.png"%image_prefix))
	#相当于文件搜索，搜索某路径下与字符匹配的文件
    image_arr = []
    mask_arr = []
    for index,item in enumerate(image_name_arr):#enumerate是枚举，输出[(0,item0),(1,item1),(2,item2)]
        img = io.imread(item,as_gray = image_as_gray)
        img = np.reshape(img,img.shape + (1,)) if image_as_gray else img
        mask = io.imread(item.replace(image_path,mask_path).replace(image_prefix,mask_prefix),as_gray = mask_as_gray)
		#重新在mask_path文件夹下搜索带有mask字符的图片（标签图片）
        mask = np.reshape(mask,mask.shape + (1,)) if mask_as_gray else mask
        img,mask = adjustData(img,mask,flag_multi_class,num_class)
        image_arr.append(img)
        mask_arr.append(mask)
    image_arr = np.array(image_arr)
    mask_arr = np.array(mask_arr)#转换成array
    return image_arr,mask_arr
 
 
# labelVisualize()函数是给出测试后的输出之后，为输出涂上不同的颜色，多类情况下才起作用，两类的话无用
def labelVisualize(num_class,color_dict,img):
    img = img[:,:,0] if len(img.shape) == 3 else img
    img_out = np.zeros(img.shape + (3,))
	#变成RGB空间，因为其他颜色只能再RGB空间才会显示
    for i in range(num_class):
        img_out[img == i,:] = color_dict[i]
		#为不同类别涂上不同的颜色，color_dict[i]是与类别数有关的颜色，img_out[img == i,:]是img_out在img中等于i类的位置上的点
    return img_out / 255.0
 
 
 
# # 直接将在0-1的浮点数直接保存成图片,生成的是灰度图
# def saveResult(save_path,npyfile,flag_multi_class = False,num_class = 2):
    # for i,item in enumerate(npyfile):
        # img = labelVisualize(num_class,COLOR_DICT,item) if flag_multi_class else item[:,:,0]
        # io.imsave(os.path.join(save_path,"%d_predict.png"%i),img)
 
# 生成二值图片		
def saveResult(save_path,npyfile,flag_multi_class = False,num_class = 2):
    for i,item in enumerate(npyfile):
        if flag_multi_class:
            img = labelVisualize(num_class,COLOR_DICT,item)
			#多类的话就图成彩色，非多类（两类）的话就是黑白色
        else:
            img=item[:,:,0]
            print(np.max(img),np.min(img))
            img[img>=0.5]=1#此时1是浮点数，下面的0也是
            img[img<0.5]=0
            print(np.max(img),np.min(img))
        io.imsave(os.path.join(save_path,"%d_predict.png"%i),img)

接下来是训练，此处将训练和测试分开在两个.py文件中，好调用！

 
# main_train.py
from model import *
from data import *
 
# #os.environ["CUDA_VISIBLE_DEVICES"] = "0"
# data_gen_args = dict() : 为keras自带的图像增强方法
data_gen_args = dict(rotation_range=0.2, #整数。随机旋转的度数范围。
                    width_shift_range=0.05, #浮点数、一维数组或整数
                    height_shift_range=0.05, #浮点数。剪切强度（以弧度逆时针方向剪切角度）。
                    shear_range=0.05, 
                    zoom_range=0.05, #浮点数 或 [lower, upper]。随机缩放范围
                    horizontal_flip=True, 
                    fill_mode='nearest') # {"constant", "nearest", "reflect" or "wrap"} 之一。默认为 'nearest'。输入边界以外的点根据给定的模式填充：		
# 建立测试集，样本和标签分别放在同一个目录下的两个文件夹中，文件夹名字为：'image','label'
#得到一个生成器，以batch=2的速率无限生成增强后的数据		
myGene = trainGenerator(2,'E:\\数据集目录\\Pic','image','label',data_gen_args,save_to_dir = None) # data
 
# 调用模型，默认模型输入图像size=(256,256,1),样本位深为8位
model = unet() # model
# 保存训练的模型参数到指定的文件夹，格式为.hdf5; 检测的值是'loss'使其更小。
model_checkpoint = ModelCheckpoint('E:/模型参数存储目录/unet_membrane.hdf5', monitor='loss',verbose=1, save_best_only=True) # keras
# 开始训练，steps_per_epoch为迭代次数，epochs：
model.fit_generator(myGene,steps_per_epoch=600,epochs=1,callbacks=[model_checkpoint]) # keras
 

接下来就是愉快的测试了：main_test.py

from model import *
from data import *
 
# python main_test.py
"""
注：
	A: target_size()为图片尺寸，要求测试集图像尺寸设置和model输入图像尺寸保持一致，
		如果不设置图片尺寸，会对输入图片做resize为处理，输入网络和输出图像尺寸默认均为（256,256），
	B: 且要求图片位深为8位，24/32的会报错！！
	C: 测试集数据名称需要设置为：0.png……
	D：model.predict_generator( ,n, ):n为测试集中样本数量，需要手动设置，不然会报错！！
"""
# 输入测试数据集，
testGene = testGenerator("E:\\测试集目录\\crackTestimg",target_size = (256,256)) # data
 
# 导入模型
model = unet(input_size = (256,256,1)) # model
 
# 导入训练好的模型
model.load_weights("E:/模型参数存放的目录/unet_membrane.hdf5")
 
# 预测数据
results = model.predict_generator(testGene,20,verbose=1) # keras
print(results)
saveResult("E:\\测试集目录\\crackTestimg",results) # data
print("over")

测试结果部分展示：（裂缝检测）

问题 ：

A：图片要求位深是8位的，24/32测试会报错，实在不知如何修改！

B：对测试数据命名需要时0.png，在源码data.py可以做相应的修改，改成按图片名字输入，，具体需要进一步探寻

C：修改部分：针对输出图片全灰，全黑，全白的修改

  （1）将data.py中所以/255的变成/255.0， 经验证没啥太大用，不过还是修改了；

   （2）将saveResult()函数进行了修改，使得结果变成二值图像

    （3）正确导入保存的模型参数： model.load_weights("E:/模型参数存放的目录/unet_membrane.hdf5")；做了此处修改后，代码正常输出。其实还是没找到具体输出全灰或全黑/白的问题在哪！！