Train U-Net from the beginning

教你如何使用网站API将IP地址转化成具体国家、城市或经纬度！

数据准备

我们所采用的数据是人的胚胎卵细胞的数据，下图中展示的就是我们所用到的胚胎图像的样例。从图中可以看到的是，一个是图像的噪声比较大，另一个是受到曝光不均匀的影响，左侧光圈内部的图像可能无法正常被识别。

所以我们首先要解决光照不均匀的问题。这个问题的解决方法网上有很多，我们这里采用了一种比较简单的方式。首先我们将图像用最小值滤波产生一个背景板，随后对背景板进行均值滤波，然后让原始图像减去背景板，得到了下图。

可以看到的是我们的光照不均匀的问题已经基本解决了，还顺带解决了一部分的噪声问题。
随后我们对图片进行了打标。我们标注了370张左右的图片，为了避免过拟合，选取了其中的70张作为验证集，剩余的图片用于训练。

模型搭建

我们采用keras搭建了我们的模型，代码如下：

def unet(pretrained_weights=None, input_size=(224, 224, 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'])

    if (pretrained_weights):
        model.load_weights(pretrained_weights)

    return model

数据增强

数据增强部分，我们采用了keras自带的image preprocessing。要保证的是，img_sub_folder和mask_sub_folder在root的文件下，图片都采用从0开始的数字序列命名，而且要保证相同名字的图片和mask对应。
在adjustData中，只做了将图片除以255的操作。

import numpy as np
import os
from keras.preprocessing.image import ImageDataGenerator
import cv2
def get_train_generator(root, img_sub_folder,
                        mask_sub_folder, aug_dict,
                        target_size=224, batch_size=5,
                        save_to_dir=False, img_save_prefix='image',
                        mask_save_prefix='mask',
                        seed=1007):
    image_datagen = ImageDataGenerator(**aug_dict)
    mask_datagen = ImageDataGenerator(**aug_dict)
    image_generator = image_datagen.flow_from_directory(
        root,
        classes=[img_sub_folder],
        class_mode=None,
        color_mode='grayscale',
        target_size=(target_size, target_size),
        batch_size=batch_size,
        save_to_dir=save_to_dir,
        save_prefix=img_save_prefix,
        seed=seed
    )
    mask_generator = mask_datagen.flow_from_directory(
        root,
        classes=[mask_sub_folder],
        class_mode=None,
        color_mode='grayscale',
        target_size=(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)
    for (img, mask) in train_generator:
        img, mask = adjustData(img, mask)
        yield (img, mask)


def adjustData(img, mask):
    # mask = mask[:, :, :, 0] if (len(mask.shape) == 4) else mask[:, :, 0]
    mask = mask[:, :, :, 0]
    mask = mask[:, :, :, np.newaxis]
    mask /= 255.
    img /= 255.
    mask = np.where(mask <= 0.5, 0, 1)
    # print('percentage: {}'.format(np.sum(mask.flatten() == 1)/np.sum(mask.flatten() == 0)))
    return (img, mask)

模型训练

为了避免过拟合，我们用验证集进行了早停。patience设置为5。

model = unet()
checkpoint = ModelCheckpoint('checkpoints/unet_duco.h5', monitor='val_loss', verbose=1, save_best_only=True)
early = EarlyStopping(patience=5, verbose=1)
model.fit_generator(generator, steps_per_epoch=20, epochs=100, callbacks=[checkpoint], validation_data=(x_val, y_val))

训练结果

验证集正确率达到了98.65%，训练集正确率达到了98.53%。