基于resnet+unet的皮肤病变分割

import numpy as np import pandas as pd import os import random import matplotlib.pyplot as plt plt.style.use('seaborn-white') import cv2 from sklearn.model_selection import StratifiedKFold from keras.models import Model from keras.layers import Input, Activation from keras.layers.core import Lambda from keras.layers.convolutional import Conv2D, Conv2DTranspose from keras.layers.pooling import MaxPooling2D from keras.layers.merge import concatenate from keras.callbacks import Callback, LearningRateScheduler, EarlyStopping, ModelCheckpoint, ReduceLROnPlateau from keras import backend as K from keras import optimizers from keras import layers from keras.losses import binary_crossentropy import tensorflow as tf from keras.preprocessing.image import array_to_img, img_to_array, load_img def convBn2d(input_tensor, filters, stage, block, kernel_size=(3, 3)): bn_axis = 3 bn_name_base = 'bn' + str(stage) + block + '_branch' conv_name_base = 'conv2d' + str(stage) + block + '_branch' x = layers.BatchNormalization(axis=bn_axis, name=bn_name_base)(input_tensor) x = layers.Activation('relu')(x) x = layers.Conv2D(filters, kernel_size, padding='same', kernel_initializer='he_normal', name=conv_name_base)(x) return x def identity_block(input_tensor, filters, stage, block, kernel_size=(3, 3)): """The identity block is the block that has no conv layer at shortcut. # Arguments input_tensor: input tensor kernel_size: default 3, the kernel size of middle conv layer at main path filters: list of integers, the filters of 3 conv layer at main path stage: integer, current stage label, used for generating layer names block: 'a','b'..., current block label, used for generating layer names # Returns Output tensor for the block. """ filters1, filters2 = [filters] * 2 bn_axis = 3 conv_name_base = 'res' + str(stage) + block + '_branch' bn_name_base = 'bn' + str(stage) + block + '_branch' x = layers.BatchNormalization(axis=bn_axis, name=bn_name_base + '2a')(input_tensor) x = layers.Activation('relu')(x) x = layers.Conv2D(filters1, kernel_size, padding='same', kernel_initializer='he_normal', name=conv_name_base + '2a')(x) x = layers.BatchNormalization(axis=bn_axis, name=bn_name_base + '2b')(x) x = layers.Activation('relu')(x) # x = layers.Dropout(0.3)(x) x = layers.Conv2D(filters2, kernel_size, padding='same', kernel_initializer='he_normal', name=conv_name_base + '2b')(x) x = layers.add([x, input_tensor]) return x def conv_block(input_tensor, filters, stage, block, strides=(2, 2), kernel_size=(3, 3)): """A block that has a conv layer at shortcut. # Arguments input_tensor: input tensor kernel_size: default 3, the kernel size of middle conv layer at main path filters: list of integers, the filters of 3 conv layer at main path stage: integer, current stage label, used for generating layer names block: 'a','b'..., current block label, used for generating layer names strides: Strides for the first conv layer in the block. # Returns Output tensor for the block. Note that from stage 3, the first conv layer at main path is with strides=(2, 2) And the shortcut should have strides=(2, 2) as well """ filters1, filters2, filters3 = [filters] * 3 bn_axis = 3 conv_name_base = 'res' + str(stage) + block + '_branch' bn_name_base = 'bn' + str(stage) + block + '_branch' x = layers.BatchNormalization(axis=bn_axis, name=bn_name_base + '2a')(input_tensor) x = layers.Activation('relu')(x) x = layers.Conv2D(filters1, (2, 2), strides=strides, padding='valid', kernel_initializer='he_normal', name=conv_name_base + '2a')(x) x = layers.BatchNormalization(axis=bn_axis, name=bn_name_base + '2b')(x) x = layers.Activation('relu')(x) # x = layers.Dropout(0.3)(x) x = layers.Conv2D(filters2, kernel_size, padding='same', kernel_initializer='he_normal', name=conv_name_base + '2b')(x) shortcut = layers.BatchNormalization( axis=bn_axis, name=bn_name_base + '1')(input_tensor) shortcut = layers.Activation('relu')(shortcut) shortcut = layers.Conv2D(filters3, (2, 2), strides=strides, padding='valid', kernel_initializer='he_normal', name=conv_name_base + '1')(shortcut) x = layers.add([x, shortcut]) return x def cse_block(prevlayer, prefix): mean = layers.Lambda(lambda xin: K.mean(xin, axis=[1, 2]))(prevlayer) # mean = layers.Dropout(0.1)(mean) lin1 = layers.Dense(K.int_shape(prevlayer)[3] // 2, name=prefix + 'cse_lin1', activation='relu')(mean) # lin1 = layers.Dropout(0.1)(lin1) lin2 = layers.Dense(K.int_shape(prevlayer)[3], name=prefix + 'cse_lin2', activation='sigmoid')(lin1) x = layers.Multiply()([prevlayer, lin2]) return x def sse_block(prevlayer, prefix): # conv = layers.Conv2D(K.int_shape(prevlayer)[3], (1, 1), padding="same", kernel_initializer="he_normal", activation='sigmoid', strides=(1, 1), # name=prefix + "_conv")(prevlayer) conv = layers.Conv2D(1, (1, 1), padding="same", kernel_initializer="he_normal", activation='sigmoid', strides=(1, 1), name=prefix + "_conv")(prevlayer) conv = layers.Multiply(name=prefix + "_mul")([prevlayer, conv]) return conv def csse_block(x, prefix): ''' Implementation of Concurrent Spatial and Channel ‘Squeeze & Excitation’ in Fully Convolutional Networks https://arxiv.org/abs/1803.02579 ''' cse = cse_block(x, prefix) sse = sse_block(x, prefix) x = layers.Add(name=prefix + "_csse_mul")([cse, sse]) return x def UResNet(input_shape, start_neurons,classes,hc=True,scse=False,block_num=5): ''' input_layer is designed to (128,128,3) ''' # 128 -> 128 input_layer=Input(input_shape) conv1 = Conv2D(start_neurons * 1, (3, 3), activation=None, padding="same")(input_layer) conv1 = identity_block(conv1, filters=start_neurons * 1, stage=1, block='a') conv1 = identity_block(conv1, filters=start_neurons * 1, stage=1, block='b') conv1 = identity_block(conv1, filters=start_neurons * 1, stage=1, block='c') if(scse==True): conv1 = csse_block(conv1, 'stage1') # 128 -> 64 conv2 = conv_block(conv1, filters=start_neurons * 2, stage=2, block='a') conv2 = identity_block(conv2, filters=start_neurons * 2, stage=2, block='b') conv2 = identity_block(conv2, filters=start_neurons * 2, stage=2, block='c') conv2 = identity_block(conv2, filters=start_neurons * 2, stage=2, block='d') if(scse==True): conv2 = csse_block(conv2, 'stage2') # 64 -> 32 conv3 = conv_block(conv2, filters=start_neurons * 4, stage=3, block='a') conv3 = identity_block(conv3, filters=start_neurons * 4, stage=3, block='b') conv3 = identity_block(conv3, filters=start_neurons * 4, stage=3, block='c') conv3 = identity_block(conv3, filters=start_neurons * 4, stage=3, block='d') conv3 = identity_block(conv3, filters=start_neurons * 4, stage=3, block='e') conv3 = identity_block(conv3, filters=start_neurons * 4, stage=3, block='f') if(scse==True): conv3 = csse_block(conv3,