CFResNet鸟类识别：原网络基础上改进算法
本文为为🔗365天深度学习训练营内部文章
原作者：K同学啊
先放一张ResNet50模型的鸟类识别结果图
一 ResNet+SE-Net+BN
import matplotlib.pyplot as plt
import tensorflow as tf
import warnings as w
w.filterwarnings('ignore')
# 支持中文
plt.rcParams['font.sans-serif'] = ['SimHei'] # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False # 用来正常显示负号
import os,PIL,pathlib
#隐藏警告
import warnings
warnings.filterwarnings('ignore')
data_dir = "./bird_photos"
data_dir = pathlib.Path(data_dir)
image_count = len(list(data_dir.glob('*/*')))
print("图片数为：",image_count)
batch_size = 8
img_height = 224
img_width = 224
train_ds = tf.keras.preprocessing.image_dataset_from_directory(
data_dir,
validation_split=0.2,
subset="training",
seed=123,
image_size=(img_height, img_width),
batch_size=batch_size)
val_ds = tf.keras.preprocessing.image_dataset_from_directory(
data_dir,
validation_split=0.2,
subset="validation",
seed=123,
image_size=(img_height, img_width),
batch_size=batch_size)
class_names = train_ds.class_names
print(class_names)
for image_batch, labels_batch in train_ds:
print(image_batch.shape)
print(labels_batch.shape)
break
AUTOTUNE = tf.data.AUTOTUNE
def preprocess_image(image,label):
return (image/255.0,label)
# 归一化处理
train_ds = train_ds.map(preprocess_image, num_parallel_calls=AUTOTUNE)
val_ds = val_ds.map(preprocess_image, num_parallel_calls=AUTOTUNE)
train_ds = train_ds.cache().shuffle(1000).prefetch(buffer_size=AUTOTUNE)
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)
plt.figure(figsize=(15, 10)) # 图形的宽为15高为10
for images, labels in train_ds.take(1):
for i in range(8):
ax = plt.subplot(2, 4, i + 1)
plt.imshow(images[i])
plt.title(class_names[labels[i]])
plt.axis("off")
from keras import layers
from keras.layers import Input, Activation, BatchNormalization, Flatten, Dropout,Reshape
from keras.layers import Dense, Conv2D, MaxPooling2D, ZeroPadding2D, AveragePooling2D, GlobalAveragePooling2D
from keras.models import Model
import tensorflow as tf
from keras.layers import Add, UpSampling2D, Conv2D
from keras.layers import Multiply, Concatenate
def squeeze_excite_block(input_tensor, ratio=16):
'''
Squeeze-and-Excitation Block
:param input_tensor: 输入张量
:param ratio: 压缩比，控制激励层中间层的维度。通常选择较小的值，如16。
:return: 加权后的张量
'''
channel_axis = -1 # 通道轴通常在最后一维
channels = input_tensor.shape[channel_axis] # 获取通道数
# Squeeze：全局平均池化
x = GlobalAveragePooling2D()(input_tensor)
x = Reshape((1, 1, channels))(x)
# Excite：两个全连接层生成通道权重
x = Dense(channels // ratio, activation='relu', kernel_initializer='he_normal', use_bias=False)(x)
x = Dense(channels, activation='sigmoid', kernel_initializer='he_normal', use_bias=False)(x)
# 将生成的权重与输入张量相乘
x = Multiply()([input_tensor, x])
return x
def identity_block(input_tensor, kernel_size, filters, stage, block):
filters1, filters2, filters3 = filters
name_base = str(stage) + block + '_identity_block_'
# 第一卷积层
x = Conv2D(filters1, (1, 1), name=name_base + 'conv1')(input_tensor)
x = BatchNormalization(name=name_base + 'bn1')(x)
x = Activation('relu', name=name_base + 'relu1')(x)
# 第二卷积层
x = Conv2D(filters2, kernel_size, padding='same', name=name_base + 'conv2')(x)
x = BatchNormalization(name=name_base + 'bn2')(x)
x = Activation('relu', name=name_base + 'relu2')(x)
# 第三卷积层
x = Conv2D(filters3, (1, 1), name=name_base + 'conv3')(x)
x = BatchNormalization(name=name_base + 'bn3')(x)
# SE-Net通道注意力机制
x = squeeze_excite_block(x)
# 残差连接
x = layers.add([x, input_tensor], name=name_base + 'add')
x = Activation('relu', name=name_base + 'relu4')(x)
return x
def conv_block(input_tensor, kernel_size, filters, stage, block, strides=(2, 2)):
filters1, filters2, filters3 = filters
res_name_base = str(stage) + block + '_conv_block_res_'
name_base = str(stage) + block + '_conv_block_'
# 主卷积层
x = Conv2D(filters1, (1, 1), strides=strides, name=name_base + 'conv1')(input_tensor)
x = BatchNormalization(name=name_base + 'bn1')(x)
x = Activation('relu', name=name_base + 'relu1')(x)
x = Conv2D(filters2, kernel_size, padding='same', name=name_base + 'conv2')(x)
x = BatchNormalization(name=name_base + 'bn2')(x)
x = Activation('relu', name=name_base + 'relu2')(x)
x = Conv2D(filters3, (1, 1), name=name_base + 'conv3')(x)
x = BatchNormalization(name=name_base + 'bn3')(x)
# 残差连接的卷积
shortcut = Conv2D(filters3, (1, 1), strides=strides, name=res_name_base + 'conv')(input_tensor)
shortcut = BatchNormalization(name=res_name_base + 'bn')(shortcut)
# SE-Net通道注意力机制
x = squeeze_excite_block(x)
# 残差连接加和
x = layers.add([x, shortcut], name=name_base + 'add')
x = Activation('relu', name=name_base + 'relu4')(x)
return x
def ResNet50(input_shape=[224,224,3], classes=4):
img_input = Input(shape=input_shape)
x = ZeroPadding2D((3, 3))(img_input)
x = Conv2D(64, (7, 7), strides=(2, 2), name='conv1')(x)
x = BatchNormalization(name='bn_conv1')(x)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)
x = conv_block(x, 3, [64, 64, 256], stage=2, block='a', strides=(1, 1))
x = identity_block(x, 3, [64, 64, 256], stage=2, block='b')
x = identity_block(x, 3, [64, 64, 256], stage=2, block='c')
x = conv_block(x, 3, [128, 128, 512], stage=3, block='a')
x = identity_block(x, 3, [128, 128, 512], stage=3, block='b')
x = identity_block(x, 3, [128, 128, 512], stage=3, block='c')
x = identity_block(x, 3, [128, 128, 512], stage=3, block='d')
x = conv_block(x, 3, [256, 256, 1024], stage=4, block='a')
x = identity_block(x, 3, [256, 256, 1024], stage=4, block='b')
x = identity_block(x, 3, [256, 256, 1024], stage=4, block='c')
x = identity_block(x, 3, [256, 256, 1024], stage=4, block='d')
x = identity_block(x, 3, [256, 256, 1024], stage=4, block='e')
x = identity_block(x, 3, [256, 256, 1024], stage=4, block='f')
x = conv_block(x, 3, [512, 512, 2048], stage=5, block='a')
x = identity_block(x, 3, [512, 512, 2048], stage=5, block='b')
x = identity_block(x, 3, [512, 512, 2048], stage=5, block='c')
x = AveragePooling2D((7, 7), name='avg_pool')(x)
x = Flatten()(x)
x = Dropout(0.5)(x)
x = Dense(classes, activation='softmax', name='fc2')(x)
model = Model(img_input, x, name='resnet50')
return model
model = ResNet50()
model.summary()
Model: "resnet50"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 224, 224, 3 0 []
)]
zero_padding2d (ZeroPadding2D) (None, 230, 230, 3) 0 ['input_1[0][0]']
conv1 (Conv2D) (None, 112, 112, 64 9472 ['zero_padding2d[0][0]']
)
bn_conv1 (BatchNormalization) (None, 112, 112, 64 256 ['conv1[0][0]']
)
activation (Activation) (None, 112, 112, 64 0 ['bn_conv1[0][0]']
)
max_pooling2d (MaxPooling2D) (None, 55, 55, 64) 0 ['activation[0][0]']
2a_conv_block_conv1 (Conv2D) (None, 55, 55, 64) 4160 ['max_pooling2d[0][0]']
2a_conv_block_bn1 (BatchNormal (None, 55, 55, 64) 256 ['2a_conv_block_conv1[0][0]']
ization)
2a_conv_block_relu1 (Activatio (None, 55, 55, 64) 0 ['2a_conv_block_bn1[0][0]']
n)
2a_conv_block_conv2 (Conv2D) (None, 55, 55, 64) 36928 ['2a_conv_block_relu1[0][0]']
2a_conv_block_bn2 (BatchNormal (None, 55, 55, 64) 256 ['2a_conv_block_conv2[0][0]']
ization)
2a_conv_block_relu2 (Activatio (None, 55, 55, 64) 0 ['2a_conv_block_bn2[0][0]']
n)
2a_conv_block_conv3 (Conv2D) (None, 55, 55, 256) 16640 ['2a_conv_block_relu2[0][0]']
2a_conv_block_bn3 (BatchNormal (None, 55, 55, 256) 1024 ['2a_conv_block_conv3[0][0]']
ization)
global_average_pooling2d (Glob (None, 256) 0 ['2a_conv_block_bn3[0][0]']
alAveragePooling2D)
reshape (Reshape) (None, 1, 1, 256) 0 ['global_average_pooling2d[0][0]'
]
dense (Dense) (None, 1, 1, 16) 4096 ['reshape[0][0]']
dense_1 (Dense) (None, 1, 1, 256) 4096 ['dense[0][0]']
2a_conv_block_res_conv (Conv2D (None, 55, 55, 256) 16640 ['max_pooling2d[0][0]']
)
multiply (Multiply) (None, 55, 55, 256) 0 ['2a_conv_block_bn3[0][0]',
'dense_1[0][0]']
2a_conv_block_res_bn (BatchNor (None, 55, 55, 256) 1024 ['2a_conv_block_res_conv[0][0]']
malization)
2a_conv_block_add (Add) (None, 55, 55, 256) 0 ['multiply[0][0]',
'2a_conv_block_res_bn[0][0]']
2a_conv_block_relu4 (Activatio (None, 55, 55, 256) 0 ['2a_conv_block_add[0][0]']
n)
2b_identity_block_conv1 (Conv2 (None, 55, 55, 64) 16448 ['2a_conv_block_relu4[0][0]']
D)
2b_identity_block_bn1 (BatchNo (None, 55, 55, 64) 256 ['2b_identity_block_conv1[0][0]']
rmalization)
2b_identity_block_relu1 (Activ (None, 55, 55, 64) 0 ['2b_identity_block_bn1[0][0]']
ation)
2b_identity_block_conv2 (Conv2 (None, 55, 55, 64) 36928 ['2b_identity_block_relu1[0][0]']
D)
2b_identity_block_bn2 (BatchNo (None, 55, 55, 64) 256 ['2b_identity_block_conv2[0][0]']
rmalization)
2b_identity_block_relu2 (Activ (None, 55, 55, 64) 0 ['2b_identity_block_bn2[0][0]']
ation)
2b_identity_block_conv3 (Conv2 (None, 55, 55, 256) 16640 ['2b_identity_block_relu2[0][0]']
D)
2b_identity_block_bn3 (BatchNo (None, 55, 55, 256) 1024 ['2b_identity_block_conv3[0][0]']
rmalization)
global_average_pooling2d_1 (Gl (None, 256) 0 ['2b_identity_block_bn3[0][0]']
obalAveragePooling2D)
reshape_1 (Reshape) (None, 1, 1, 256) 0 ['global_average_pooling2d_1[0][0
]']
dense_2 (Dense) (None, 1, 1, 16) 4096 ['reshape_1[0][0]']
dense_3 (Dense) (None, 1, 1, 256) 4096 ['dense_2[0][0]']
multiply_1 (Multiply) (None, 55, 55, 256) 0 ['2b_identity_block_bn3[0][0]',
'dense_3[0][0]']
2b_identity_block_add (Add) (None, 55, 55, 256) 0 ['multiply_1[0][0]',
'2a_conv_block_relu4[0][0]']
2b_identity_block_relu4 (Activ (None, 55, 55, 256) 0 ['2b_identity_block_add[0][0]']
ation)
2c_identity_block_conv1 (Conv2 (None, 55, 55, 64) 16448 ['2b_identity_block_relu4[0][0]']
D)
2c_identity_block_bn1 (BatchNo (None, 55, 55, 64) 256 ['2c_identity_block_conv1[0][0]']
rmalization)
2c_identity_block_relu1 (Activ (None, 55, 55, 64) 0 ['2c_identity_block_bn1[0][0]']
ation)
2c_identity_block_conv2 (Conv2 (None, 55, 55, 64) 36928 ['2c_identity_block_relu1[0][0]']
D)
2c_identity_block_bn2 (BatchNo (None, 55, 55, 64) 256 ['2c_identity_block_conv2[0][0]']
rmalization)
2c_identity_block_relu2 (Activ (None, 55, 55, 64) 0 ['2c_identity_block_bn2[0][0]']
ation)
2c_identity_block_conv3 (Conv2 (None, 55, 55, 256) 16640 ['2c_identity_block_relu2[0][0]']
D)
2c_identity_block_bn3 (BatchNo (None, 55, 55, 256) 1024 ['2c_identity_block_conv3[0][0]']
rmalization)
global_average_pooling2d_2 (Gl (None, 256) 0 ['2c_identity_block_bn3[0][0]']
obalAveragePooling2D)
reshape_2 (Reshape) (None, 1, 1, 256) 0 ['global_average_pooling2d_2[0][0
]']
dense_4 (Dense) (None, 1, 1, 16) 4096 ['reshape_2[0][0]']
dense_5 (Dense) (None, 1, 1, 256) 4096 ['dense_4[0][0]']
multiply_2 (Multiply) (None, 55, 55, 256) 0 ['2c_identity_block_bn3[0][0]',
'dense_5[0][0]']
2c_identity_block_add (Add) (None, 55, 55, 256) 0 ['multiply_2[0][0]',
'2b_identity_block_relu4[0][0]']
2c_identity_block_relu4 (Activ (None, 55, 55, 256) 0 ['2c_identity_block_add[0][0]']
ation)
3a_conv_block_conv1 (Conv2D) (None, 28, 28, 128) 32896 ['2c_identity_block_relu4[0][0]']
3a_conv_block_bn1 (BatchNormal (None, 28, 28, 128) 512 ['3a_conv_block_conv1[0][0]']
ization)
3a_conv_block_relu1 (Activatio (None, 28, 28, 128) 0 ['3a_conv_block_bn1[0][0]']
n)
3a_conv_block_conv2 (Conv2D) (None, 28, 28, 128) 147584 ['3a_conv_block_relu1[0][0]']
3a_conv_block_bn2 (BatchNormal (None, 28, 28, 128) 512 ['3a_conv_block_conv2[0][0]']
ization)
3a_conv_block_relu2 (Activatio (None, 28, 28, 128) 0 ['3a_conv_block_bn2[0][0]']
n)
3a_conv_block_conv3 (Conv2D) (None, 28, 28, 512) 66048 ['3a_conv_block_relu2[0][0]']
3a_conv_block_bn3 (BatchNormal (None, 28, 28, 512) 2048 ['3a_conv_block_conv3[0][0]']
ization)
global_average_pooling2d_3 (Gl (None, 512) 0 ['3a_conv_block_bn3[0][0]']
obalAveragePooling2D)
reshape_3 (Reshape) (None, 1, 1, 512) 0 ['global_average_pooling2d_3[0][0
]']
dense_6 (Dense) (None, 1, 1, 32) 16384 ['reshape_3[0][0]']
dense_7 (Dense) (None, 1, 1, 512) 16384 ['dense_6[0][0]']
3a_conv_block_res_conv (Conv2D (None, 28, 28, 512) 131584 ['2c_identity_block_relu4[0][0]']
)
multiply_3 (Multiply) (None, 28, 28, 512) 0 ['3a_conv_block_bn3[0][0]',
'dense_7[0][0]']
3a_conv_block_res_bn (BatchNor (None, 28, 28, 512) 2048 ['3a_conv_block_res_conv[0][0]']
malization)
3a_conv_block_add (Add) (None, 28, 28, 512) 0 ['multiply_3[0][0]',
'3a_conv_block_res_bn[0][0]']
3a_conv_block_relu4 (Activatio (None, 28, 28, 512) 0 ['3a_conv_block_add[0][0]']
n)
3b_identity_block_conv1 (Conv2 (None, 28, 28, 128) 65664 ['3a_conv_block_relu4[0][0]']
D)
3b_identity_block_bn1 (BatchNo (None, 28, 28, 128) 512 ['3b_identity_block_conv1[0][0]']
rmalization)
3b_identity_block_relu1 (Activ (None, 28, 28, 128) 0 ['3b_identity_block_bn1[0][0]']
ation)
3b_identity_block_conv2 (Conv2 (None, 28, 28, 128) 147584 ['3b_identity_block_relu1[0][0]']
D)
3b_identity_block_bn2 (BatchNo (None, 28, 28, 128) 512 ['3b_identity_block_conv2[0][0]']
rmalization)
3b_identity_block_relu2 (Activ (None, 28, 28, 128) 0 ['3b_identity_block_bn2[0][0]']
ation)
3b_identity_block_conv3 (Conv2 (None, 28, 28, 512) 66048 ['3b_identity_block_relu2[0][0]']
D)
3b_identity_block_bn3 (BatchNo (None, 28, 28, 512) 2048 ['3b_identity_block_conv3[0][0]']
rmalization)
global_average_pooling2d_4 (Gl (None, 512) 0 ['3b_identity_block_bn3[0][0]']
obalAveragePooling2D)
reshape_4 (Reshape) (None, 1, 1, 512) 0 ['global_average_pooling2d_4[0][0
]']
dense_8 (Dense) (None, 1, 1, 32) 16384 ['reshape_4[0][0]']
dense_9 (Dense) (None, 1, 1, 512) 16384 ['dense_8[0][0]']
multiply_4 (Multiply) (None, 28, 28, 512) 0 ['3b_identity_block_bn3[0][0]',
'dense_9[0][0]']
3b_identity_block_add (Add) (None, 28, 28, 512) 0 ['multiply_4[0][0]',
'3a_conv_block_relu4[0][0]']
3b_identity_block_relu4 (Activ (None, 28, 28, 512) 0 ['3b_identity_block_add[0][0]']
ation)
3c_identity_block_conv1 (Conv2 (None, 28, 28, 128) 65664 ['3b_identity_block_relu4[0][0]']
D)
3c_identity_block_bn1 (BatchNo (None, 28, 28, 128) 512 ['3c_identity_block_conv1[0][0]']
rmalization)
3c_identity_block_relu1 (Activ (None, 28, 28, 128) 0 ['3c_identity_block_bn1[0][0]']
ation)
3c_identity_block_conv2 (Conv2 (None, 28, 28, 128) 147584 ['3c_identity_block_relu1[0][0]']
D)
3c_identity_block_bn2 (BatchNo (None, 28, 28, 128) 512 ['3c_identity_block_conv2[0][0]']
rmalization)
3c_identity_block_relu2 (Activ (None, 28, 28, 128) 0 ['3c_identity_block_bn2[0][0]']
ation)
3c_identity_block_conv3 (Conv2 (None, 28, 28, 512) 66048 ['3c_identity_block_relu2[0][0]']
D)
3c_identity_block_bn3 (BatchNo (None, 28, 28, 512) 2048 ['3c_identity_block_conv3[0][0]']
rmalization)
global_average_pooling2d_5 (Gl (None, 512) 0 ['3c_identity_block_bn3[0][0]']
obalAveragePooling2D)
reshape_5 (Reshape) (None, 1, 1, 512) 0 ['global_average_pooling2d_5[0][0
]']
dense_10 (Dense) (None, 1, 1, 32) 16384 ['reshape_5[0][0]']
dense_11 (Dense) (None, 1, 1, 512) 16384 ['dense_10[0][0]']
multiply_5 (Multiply) (None, 28, 28, 512) 0 ['3c_identity_block_bn3[0][0]',
'dense_11[0][0]']
3c_identity_block_add (Add) (None, 28, 28, 512) 0 ['multiply_5[0][0]',
'3b_identity_block_relu4[0][0]']
3c_identity_block_relu4 (Activ (None, 28, 28, 512) 0 ['3c_identity_block_add[0][0]']
ation)
3d_identity_block_conv1 (Conv2 (None, 28, 28, 128) 65664 ['3c_identity_block_relu4[0][0]']
D)
3d_identity_block_bn1 (BatchNo (None, 28, 28, 128) 512 ['3d_identity_block_conv1[0][0]']
rmalization)
3d_identity_block_relu1 (Activ (None, 28, 28, 128) 0 ['3d_identity_block_bn1[0][0]']
ation)
3d_identity_block_conv2 (Conv2 (None, 28, 28, 128) 147584 ['3d_identity_block_relu1[0][0]']
D)
3d_identity_block_bn2 (BatchNo (None, 28, 28, 128) 512 ['3d_identity_block_conv2[0][0]']
rmalization)
3d_identity_block_relu2 (Activ (None, 28, 28, 128) 0 ['3d_identity_block_bn2[0][0]']
ation)
3d_identity_block_conv3 (Conv2 (None, 28, 28, 512) 66048 ['3d_identity_block_relu2[0][0]']
D)
3d_identity_block_bn3 (BatchNo (None, 28, 28, 512) 2048 ['3d_identity_block_conv3[0][0]']
rmalization)
global_average_pooling2d_6 (Gl (None, 512) 0 ['3d_identity_block_bn3[0][0]']
obalAveragePooling2D)
reshape_6 (Reshape) (None, 1, 1, 512) 0 ['global_average_pooling2d_6[0][0
]']
dense_12 (Dense) (None, 1, 1, 32) 16384 ['reshape_6[0][0]']
dense_13 (Dense) (None, 1, 1, 512) 16384 ['dense_12[0][0]']
multiply_6 (Multiply) (None, 28, 28, 512) 0 ['3d_identity_block_bn3[0][0]',
'dense_13[0][0]']
3d_identity_block_add (Add) (None, 28, 28, 512) 0 ['multiply_6[0][0]',
'3c_identity_block_relu4[0][0]']
3d_identity_block_relu4 (Activ (None, 28, 28, 512) 0 ['3d_identity_block_add[0][0]']
ation)
4a_conv_block_conv1 (Conv2D) (None, 14, 14, 256) 131328 ['3d_identity_block_relu4[0][0]']
4a_conv_block_bn1 (BatchNormal (None, 14, 14, 256) 1024 ['4a_conv_block_conv1[0][0]']
ization)
4a_conv_block_relu1 (Activatio (None, 14, 14, 256) 0 ['4a_conv_block_bn1[0][0]']
n)
4a_conv_block_conv2 (Conv2D) (None, 14, 14, 256) 590080 ['4a_conv_block_relu1[0][0]']
4a_conv_block_bn2 (BatchNormal (None, 14, 14, 256) 1024 ['4a_conv_block_conv2[0][0]']
ization)
4a_conv_block_relu2 (Activatio (None, 14, 14, 256) 0 ['4a_conv_block_bn2[0][0]']
n)
4a_conv_block_conv3 (Conv2D) (None, 14, 14, 1024 263168 ['4a_conv_block_relu2[0][0]']
)
4a_conv_block_bn3 (BatchNormal (None, 14, 14, 1024 4096 ['4a_conv_block_conv3[0][0]']
ization) )
global_average_pooling2d_7 (Gl (None, 1024) 0 ['4a_conv_block_bn3[0][0]']
obalAveragePooling2D)
reshape_7 (Reshape) (None, 1, 1, 1024) 0 ['global_average_pooling2d_7[0][0
]']
dense_14 (Dense) (None, 1, 1, 64) 65536 ['reshape_7[0][0]']
dense_15 (Dense) (None, 1, 1, 1024) 65536 ['dense_14[0][0]']
4a_conv_block_res_conv (Conv2D (None, 14, 14, 1024 525312 ['3d_identity_block_relu4[0][0]']
) )
multiply_7 (Multiply) (None, 14, 14, 1024 0 ['4a_conv_block_bn3[0][0]',
) 'dense_15[0][0]']
4a_conv_block_res_bn (BatchNor (None, 14, 14, 1024 4096 ['4a_conv_block_res_conv[0][0]']
malization) )
4a_conv_block_add (Add) (None, 14, 14, 1024 0 ['multiply_7[0][0]',
) '4a_conv_block_res_bn[0][0]']
4a_conv_block_relu4 (Activatio (None, 14, 14, 1024 0 ['4a_conv_block_add[0][0]']
n) )
4b_identity_block_conv1 (Conv2 (None, 14, 14, 256) 262400 ['4a_conv_block_relu4[0][0]']
D)
4b_identity_block_bn1 (BatchNo (None, 14, 14, 256) 1024 ['4b_identity_block_conv1[0][0]']
rmalization)
4b_identity_block_relu1 (Activ (None, 14, 14, 256) 0 ['4b_identity_block_bn1[0][0]']
ation)
4b_identity_block_conv2 (Conv2 (None, 14, 14, 256) 590080 ['4b_identity_block_relu1[0][0]']
D)
4b_identity_block_bn2 (BatchNo (None, 14, 14, 256) 1024 ['4b_identity_block_conv2[0][0]']
rmalization)
4b_identity_block_relu2 (Activ (None, 14, 14, 256) 0 ['4b_identity_block_bn2[0][0]']
ation)
4b_identity_block_conv3 (Conv2 (None, 14, 14, 1024 263168 ['4b_identity_block_relu2[0][0]']
D) )
4b_identity_block_bn3 (BatchNo (None, 14, 14, 1024 4096 ['4b_identity_block_conv3[0][0]']
rmalization) )
global_average_pooling2d_8 (Gl (None, 1024) 0 ['4b_identity_block_bn3[0][0]']
obalAveragePooling2D)
reshape_8 (Reshape) (None, 1, 1, 1024) 0 ['global_average_pooling2d_8[0][0
]']
dense_16 (Dense) (None, 1, 1, 64) 65536 ['reshape_8[0][0]']
dense_17 (Dense) (None, 1, 1, 1024) 65536 ['dense_16[0][0]']
multiply_8 (Multiply) (None, 14, 14, 1024 0 ['4b_identity_block_bn3[0][0]',
) 'dense_17[0][0]']
4b_identity_block_add (Add) (None, 14, 14, 1024 0 ['multiply_8[0][0]',
) '4a_conv_block_relu4[0][0]']
4b_identity_block_relu4 (Activ (None, 14, 14, 1024 0 ['4b_identity_block_add[0][0]']
ation) )
4c_identity_block_conv1 (Conv2 (None, 14, 14, 256) 262400 ['4b_identity_block_relu4[0][0]']
D)
4c_identity_block_bn1 (BatchNo (None, 14, 14, 256) 1024 ['4c_identity_block_conv1[0][0]']
rmalization)
4c_identity_block_relu1 (Activ (None, 14, 14, 256) 0 ['4c_identity_block_bn1[0][0]']
ation)
4c_identity_block_conv2 (Conv2 (None, 14, 14, 256) 590080 ['4c_identity_block_relu1[0][0]']
D)
4c_identity_block_bn2 (BatchNo (None, 14, 14, 256) 1024 ['4c_identity_block_conv2[0][0]']
rmalization)
4c_identity_block_relu2 (Activ (None, 14, 14, 256) 0 ['4c_identity_block_bn2[0][0]']
ation)
4c_identity_block_conv3 (Conv2 (None, 14, 14, 1024 263168 ['4c_identity_block_relu2[0][0]']
D) )
4c_identity_block_bn3 (BatchNo (None, 14, 14, 1024 4096 ['4c_identity_block_conv3[0][0]']
rmalization) )
global_average_pooling2d_9 (Gl (None, 1024) 0 ['4c_identity_block_bn3[0][0]']
obalAveragePooling2D)
reshape_9 (Reshape) (None, 1, 1, 1024) 0 ['global_average_pooling2d_9[0][0
]']
dense_18 (Dense) (None, 1, 1, 64) 65536 ['reshape_9[0][0]']
dense_19 (Dense) (None, 1, 1, 1024) 65536 ['dense_18[0][0]']
multiply_9 (Multiply) (None, 14, 14, 1024 0 ['4c_identity_block_bn3[0][0]',
) 'dense_19[0][0]']
4c_identity_block_add (Add) (None, 14, 14, 1024 0 ['multiply_9[0][0]',
) '4b_identity_block_relu4[0][0]']
4c_identity_block_relu4 (Activ (None, 14, 14, 1024 0 ['4c_identity_block_add[0][0]']
ation) )
4d_identity_block_conv1 (Conv2 (None, 14, 14, 256) 262400 ['4c_identity_block_relu4[0][0]']
D)
4d_identity_block_bn1 (BatchNo (None, 14, 14, 256) 1024 ['4d_identity_block_conv1[0][0]']
rmalization)
4d_identity_block_relu1 (Activ (None, 14, 14, 256) 0 ['4d_identity_block_bn1[0][0]']
ation)
4d_identity_block_conv2 (Conv2 (None, 14, 14, 256) 590080 ['4d_identity_block_relu1[0][0]']
D)
4d_identity_block_bn2 (BatchNo (None, 14, 14, 256) 1024 ['4d_identity_block_conv2[0][0]']
rmalization)
4d_identity_block_relu2 (Activ (None, 14, 14, 256) 0 ['4d_identity_block_bn2[0][0]']
ation)
4d_identity_block_conv3 (Conv2 (None, 14, 14, 1024 263168 ['4d_identity_block_relu2[0][0]']
D) )
4d_identity_block_bn3 (BatchNo (None, 14, 14, 1024 4096 ['4d_identity_block_conv3[0][0]']
rmalization) )
global_average_pooling2d_10 (G (None, 1024) 0 ['4d_identity_block_bn3[0][0]']
lobalAveragePooling2D)
reshape_10 (Reshape) (None, 1, 1, 1024) 0 ['global_average_pooling2d_10[0][
0]']
dense_20 (Dense) (None, 1, 1, 64) 65536 ['reshape_10[0][0]']
dense_21 (Dense) (None, 1, 1, 1024) 65536 ['dense_20[0][0]']
multiply_10 (Multiply) (None, 14, 14, 1024 0 ['4d_identity_block_bn3[0][0]',
) 'dense_21[0][0]']
4d_identity_block_add (Add) (None, 14, 14, 1024 0 ['multiply_10[0][0]',
) '4c_identity_block_relu4[0][0]']
4d_identity_block_relu4 (Activ (None, 14, 14, 1024 0 ['4d_identity_block_add[0][0]']
ation) )
4e_identity_block_conv1 (Conv2 (None, 14, 14, 256) 262400 ['4d_identity_block_relu4[0][0]']
D)
4e_identity_block_bn1 (BatchNo (None, 14, 14, 256) 1024 ['4e_identity_block_conv1[0][0]']
rmalization)
4e_identity_block_relu1 (Activ (None, 14, 14, 256) 0 ['4e_identity_block_bn1[0][0]']
ation)
4e_identity_block_conv2 (Conv2 (None, 14, 14, 256) 590080 ['4e_identity_block_relu1[0][0]']
D)
4e_identity_block_bn2 (BatchNo (None, 14, 14, 256) 1024 ['4e_identity_block_conv2[0][0]']
rmalization)
4e_identity_block_relu2 (Activ (None, 14, 14, 256) 0 ['4e_identity_block_bn2[0][0]']
ation)
4e_identity_block_conv3 (Conv2 (None, 14, 14, 1024 263168 ['4e_identity_block_relu2[0][0]']
D) )
4e_identity_block_bn3 (BatchNo (None, 14, 14, 1024 4096 ['4e_identity_block_conv3[0][0]']
rmalization) )
global_average_pooling2d_11 (G (None, 1024) 0 ['4e_identity_block_bn3[0][0]']
lobalAveragePooling2D)
reshape_11 (Reshape) (None, 1, 1, 1024) 0 ['global_average_pooling2d_11[0][
0]']
dense_22 (Dense) (None, 1, 1, 64) 65536 ['reshape_11[0][0]']
dense_23 (Dense) (None, 1, 1, 1024) 65536 ['dense_22[0][0]']
multiply_11 (Multiply) (None, 14, 14, 1024 0 ['4e_identity_block_bn3[0][0]',
) 'dense_23[0][0]']
4e_identity_block_add (Add) (None, 14, 14, 1024 0 ['multiply_11[0][0]',
) '4d_identity_block_relu4[0][0]']
4e_identity_block_relu4 (Activ (None, 14, 14, 1024 0 ['4e_identity_block_add[0][0]']
ation) )
4f_identity_block_conv1 (Conv2 (None, 14, 14, 256) 262400 ['4e_identity_block_relu4[0][0]']
D)
4f_identity_block_bn1 (BatchNo (None, 14, 14, 256) 1024 ['4f_identity_block_conv1[0][0]']
rmalization)
4f_identity_block_relu1 (Activ (None, 14, 14, 256) 0 ['4f_identity_block_bn1[0][0]']
ation)
4f_identity_block_conv2 (Conv2 (None, 14, 14, 256) 590080 ['4f_identity_block_relu1[0][0]']
D)
4f_identity_block_bn2 (BatchNo (None, 14, 14, 256) 1024 ['4f_identity_block_conv2[0][0]']
rmalization)
4f_identity_block_relu2 (Activ (None, 14, 14, 256) 0 ['4f_identity_block_bn2[0][0]']
ation)
4f_identity_block_conv3 (Conv2 (None, 14, 14, 1024 263168 ['4f_identity_block_relu2[0][0]']
D) )
4f_identity_block_bn3 (BatchNo (None, 14, 14, 1024 4096 ['4f_identity_block_conv3[0][0]']
rmalization) )
global_average_pooling2d_12 (G (None, 1024) 0 ['4f_identity_block_bn3[0][0]']
lobalAveragePooling2D)
reshape_12 (Reshape) (None, 1, 1, 1024) 0 ['global_average_pooling2d_12[0][
0]']
dense_24 (Dense) (None, 1, 1, 64) 65536 ['reshape_12[0][0]']
dense_25 (Dense) (None, 1, 1, 1024) 65536 ['dense_24[0][0]']
multiply_12 (Multiply) (None, 14, 14, 1024 0 ['4f_identity_block_bn3[0][0]',
) 'dense_25[0][0]']
4f_identity_block_add (Add) (None, 14, 14, 1024 0 ['multiply_12[0][0]',
) '4e_identity_block_relu4[0][0]']
4f_identity_block_relu4 (Activ (None, 14, 14, 1024 0 ['4f_identity_block_add[0][0]']
ation) )
5a_conv_block_conv1 (Conv2D) (None, 7, 7, 512) 524800 ['4f_identity_block_relu4[0][0]']
5a_conv_block_bn1 (BatchNormal (None, 7, 7, 512) 2048 ['5a_conv_block_conv1[0][0]']
ization)
5a_conv_block_relu1 (Activatio (None, 7, 7, 512) 0 ['5a_conv_block_bn1[0][0]']
n)
5a_conv_block_conv2 (Conv2D) (None, 7, 7, 512) 2359808 ['5a_conv_block_relu1[0][0]']
5a_conv_block_bn2 (BatchNormal (None, 7, 7, 512) 2048 ['5a_conv_block_conv2[0][0]']
ization)
5a_conv_block_relu2 (Activatio (None, 7, 7, 512) 0 ['5a_conv_block_bn2[0][0]']
n)
5a_conv_block_conv3 (Conv2D) (None, 7, 7, 2048) 1050624 ['5a_conv_block_relu2[0][0]']
5a_conv_block_bn3 (BatchNormal (None, 7, 7, 2048) 8192 ['5a_conv_block_conv3[0][0]']
ization)
global_average_pooling2d_13 (G (None, 2048) 0 ['5a_conv_block_bn3[0][0]']
lobalAveragePooling2D)
reshape_13 (Reshape) (None, 1, 1, 2048) 0 ['global_average_pooling2d_13[0][
0]']
dense_26 (Dense) (None, 1, 1, 128) 262144 ['reshape_13[0][0]']
dense_27 (Dense) (None, 1, 1, 2048) 262144 ['dense_26[0][0]']
5a_conv_block_res_conv (Conv2D (None, 7, 7, 2048) 2099200 ['4f_identity_block_relu4[0][0]']
)
multiply_13 (Multiply) (None, 7, 7, 2048) 0 ['5a_conv_block_bn3[0][0]',
'dense_27[0][0]']
5a_conv_block_res_bn (BatchNor (None, 7, 7, 2048) 8192 ['5a_conv_block_res_conv[0][0]']
malization)
5a_conv_block_add (Add) (None, 7, 7, 2048) 0 ['multiply_13[0][0]',
'5a_conv_block_res_bn[0][0]']
5a_conv_block_relu4 (Activatio (None, 7, 7, 2048) 0 ['5a_conv_block_add[0][0]']
n)
5b_identity_block_conv1 (Conv2 (None, 7, 7, 512) 1049088 ['5a_conv_block_relu4[0][0]']
D)
5b_identity_block_bn1 (BatchNo (None, 7, 7, 512) 2048 ['5b_identity_block_conv1[0][0]']
rmalization)
5b_identity_block_relu1 (Activ (None, 7, 7, 512) 0 ['5b_identity_block_bn1[0][0]']
ation)
5b_identity_block_conv2 (Conv2 (None, 7, 7, 512) 2359808 ['5b_identity_block_relu1[0][0]']
D)
5b_identity_block_bn2 (BatchNo (None, 7, 7, 512) 2048 ['5b_identity_block_conv2[0][0]']
rmalization)
5b_identity_block_relu2 (Activ (None, 7, 7, 512) 0 ['5b_identity_block_bn2[0][0]']
ation)
5b_identity_block_conv3 (Conv2 (None, 7, 7, 2048) 1050624 ['5b_identity_block_relu2[0][0]']
D)
5b_identity_block_bn3 (BatchNo (None, 7, 7, 2048) 8192 ['5b_identity_block_conv3[0][0]']
rmalization)
global_average_pooling2d_14 (G (None, 2048) 0 ['5b_identity_block_bn3[0][0]']
lobalAveragePooling2D)
reshape_14 (Reshape) (None, 1, 1, 2048) 0 ['global_average_pooling2d_14[0][
0]']
dense_28 (Dense) (None, 1, 1, 128) 262144 ['reshape_14[0][0]']
dense_29 (Dense) (None, 1, 1, 2048) 262144 ['dense_28[0][0]']
multiply_14 (Multiply) (None, 7, 7, 2048) 0 ['5b_identity_block_bn3[0][0]',
'dense_29[0][0]']
5b_identity_block_add (Add) (None, 7, 7, 2048) 0 ['multiply_14[0][0]',
'5a_conv_block_relu4[0][0]']
5b_identity_block_relu4 (Activ (None, 7, 7, 2048) 0 ['5b_identity_block_add[0][0]']
ation)
5c_identity_block_conv1 (Conv2 (None, 7, 7, 512) 1049088 ['5b_identity_block_relu4[0][0]']
D)
5c_identity_block_bn1 (BatchNo (None, 7, 7, 512) 2048 ['5c_identity_block_conv1[0][0]']
rmalization)
5c_identity_block_relu1 (Activ (None, 7, 7, 512) 0 ['5c_identity_block_bn1[0][0]']
ation)
5c_identity_block_conv2 (Conv2 (None, 7, 7, 512) 2359808 ['5c_identity_block_relu1[0][0]']
D)
5c_identity_block_bn2 (BatchNo (None, 7, 7, 512) 2048 ['5c_identity_block_conv2[0][0]']
rmalization)
5c_identity_block_relu2 (Activ (None, 7, 7, 512) 0 ['5c_identity_block_bn2[0][0]']
ation)
5c_identity_block_conv3 (Conv2 (None, 7, 7, 2048) 1050624 ['5c_identity_block_relu2[0][0]']
D)
5c_identity_block_bn3 (BatchNo (None, 7, 7, 2048) 8192 ['5c_identity_block_conv3[0][0]']
rmalization)
global_average_pooling2d_15 (G (None, 2048) 0 ['5c_identity_block_bn3[0][0]']
lobalAveragePooling2D)
reshape_15 (Reshape) (None, 1, 1, 2048) 0 ['global_average_pooling2d_15[0][
0]']
dense_30 (Dense) (None, 1, 1, 128) 262144 ['reshape_15[0][0]']
dense_31 (Dense) (None, 1, 1, 2048) 262144 ['dense_30[0][0]']
multiply_15 (Multiply) (None, 7, 7, 2048) 0 ['5c_identity_block_bn3[0][0]',
'dense_31[0][0]']
5c_identity_block_add (Add) (None, 7, 7, 2048) 0 ['multiply_15[0][0]',
'5b_identity_block_relu4[0][0]']
5c_identity_block_relu4 (Activ (None, 7, 7, 2048) 0 ['5c_identity_block_add[0][0]']
ation)
avg_pool (AveragePooling2D) (None, 1, 1, 2048) 0 ['5c_identity_block_relu4[0][0]']
flatten (Flatten) (None, 2048) 0 ['avg_pool[0][0]']
dropout (Dropout) (None, 2048) 0 ['flatten[0][0]']
fc2 (Dense) (None, 4) 8196 ['dropout[0][0]']
==================================================================================================
Total params: 26,110,852
Trainable params: 26,057,732
Non-trainable params: 53,120
__________________________________________________________________________________________________
model.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
epochs = 30
history = model.fit(
train_ds,
validation_data=val_ds,
epochs=epochs,
)
Epoch 1/30 57/57 [==============================] - 120s 2s/step - loss: 1.6071 - accuracy: 0.5022 - val_loss: 1.6805 - val_accuracy: 0.3628 Epoch 2/30 57/57 [==============================] - 101s 2s/step - loss: 1.1572 - accuracy: 0.5973 - val_loss: 2.9043 - val_accuracy: 0.3628 Epoch 3/30 57/57 [==============================] - 103s 2s/step - loss: 1.0106 - accuracy: 0.6438 - val_loss: 3.5882 - val_accuracy: 0.2655 Epoch 4/30 57/57 [==============================] - 104s 2s/step - loss: 0.6960 - accuracy: 0.7345 - val_loss: 3.8824 - val_accuracy: 0.2920 Epoch 5/30 57/57 [==============================] - 103s 2s/step - loss: 0.6722 - accuracy: 0.8009 - val_loss: 1.9140 - val_accuracy: 0.3805 Epoch 6/30 57/57 [==============================] - 102s 2s/step - loss: 0.5720 - accuracy: 0.8009 - val_loss: 1.3526 - val_accuracy: 0.5133 Epoch 7/30 57/57 [==============================] - 103s 2s/step - loss: 0.5234 - accuracy: 0.8252 - val_loss: 1.6950 - val_accuracy: 0.6195 Epoch 8/30 57/57 [==============================] - 103s 2s/step - loss: 0.5409 - accuracy: 0.8186 - val_loss: 1.0905 - val_accuracy: 0.5752 Epoch 9/30 57/57 [==============================] - 102s 2s/step - loss: 0.4960 - accuracy: 0.8230 - val_loss: 1.0269 - val_accuracy: 0.5664 Epoch 10/30 57/57 [==============================] - 106s 2s/step - loss: 0.3521 - accuracy: 0.8761 - val_loss: 0.7942 - val_accuracy: 0.7965 Epoch 11/30 57/57 [==============================] - 103s 2s/step - loss: 0.2162 - accuracy: 0.9204 - val_loss: 0.9084 - val_accuracy: 0.7168 Epoch 12/30 57/57 [==============================] - 103s 2s/step - loss: 0.3159 - accuracy: 0.9093 - val_loss: 1.8489 - val_accuracy: 0.7611 Epoch 13/30 57/57 [==============================] - 102s 2s/step - loss: 0.2727 - accuracy: 0.8960 - val_loss: 2.2825 - val_accuracy: 0.7345 Epoch 14/30 57/57 [==============================] - 104s 2s/step - loss: 0.1902 - accuracy: 0.9381 - val_loss: 1.1050 - val_accuracy: 0.7257 Epoch 15/30 57/57 [==============================] - 105s 2s/step - loss: 0.2085 - accuracy: 0.9292 - val_loss: 0.3290 - val_accuracy: 0.9027 Epoch 16/30 57/57 [==============================] - 105s 2s/step - loss: 0.1697 - accuracy: 0.9358 - val_loss: 1.0470 - val_accuracy: 0.7965 Epoch 17/30 57/57 [==============================] - 106s 2s/step - loss: 0.1955 - accuracy: 0.9381 - val_loss: 9.2690 - val_accuracy: 0.3540 Epoch 18/30 57/57 [==============================] - 103s 2s/step - loss: 0.3337 - accuracy: 0.8960 - val_loss: 1.6920 - val_accuracy: 0.7699 Epoch 19/30 57/57 [==============================] - 103s 2s/step - loss: 0.1869 - accuracy: 0.9292 - val_loss: 9.7153 - val_accuracy: 0.3628 Epoch 20/30 57/57 [==============================] - 103s 2s/step - loss: 0.2506 - accuracy: 0.9049 - val_loss: 0.9142 - val_accuracy: 0.7876 Epoch 21/30 57/57 [==============================] - 107s 2s/step - loss: 0.1941 - accuracy: 0.9358 - val_loss: 0.7740 - val_accuracy: 0.8142 Epoch 22/30 57/57 [==============================] - 103s 2s/step - loss: 0.0971 - accuracy: 0.9690 - val_loss: 0.5248 - val_accuracy: 0.8230 Epoch 23/30 57/57 [==============================] - 104s 2s/step - loss: 0.0549 - accuracy: 0.9845 - val_loss: 2.8425 - val_accuracy: 0.6637 Epoch 24/30 57/57 [==============================] - 107s 2s/step - loss: 0.0251 - accuracy: 0.9934 - val_loss: 0.1835 - val_accuracy: 0.9292 Epoch 25/30 57/57 [==============================] - 110s 2s/step - loss: 0.0131 - accuracy: 1.0000 - val_loss: 0.2147 - val_accuracy: 0.9469 Epoch 26/30 57/57 [==============================] - 103s 2s/step - loss: 0.0043 - accuracy: 1.0000 - val_loss: 0.1484 - val_accuracy: 0.9469 Epoch 27/30 57/57 [==============================] - 106s 2s/step - loss: 0.0029 - accuracy: 1.0000 - val_loss: 0.1338 - val_accuracy: 0.9558 Epoch 28/30 57/57 [==============================] - 107s 2s/step - loss: 0.0013 - accuracy: 1.0000 - val_loss: 0.1428 - val_accuracy: 0.9469 Epoch 29/30 57/57 [==============================] - 108s 2s/step - loss: 6.1017e-04 - accuracy: 1.0000 - val_loss: 0.1412 - val_accuracy: 0.9469 Epoch 30/30 57/57 [==============================] - 111s 2s/step - loss: 5.0776e-04 - accuracy: 1.0000 - val_loss: 0.1457 - val_accuracy: 0.9381
# 获取实际训练轮数
actual_epochs = len(history.history['accuracy'])
acc = history.history['accuracy']
val_acc = history.history['val_accuracy']
loss = history.history['loss']
val_loss = history.history['val_loss']
epochs_range = range(actual_epochs)
plt.figure(figsize=(12, 4))
# 绘制准确率
plt.subplot(1, 2, 1)
plt.plot(epochs_range, acc, label='Training Accuracy')
plt.plot(epochs_range, val_acc, label='Validation Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')
# 绘制损失
plt.subplot(1, 2, 2)
plt.plot(epochs_range, loss, label='Training Loss')
plt.plot(epochs_range, val_loss, label='Validation Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()
二 CFResNet算法
是在ResNet基础上加上BN、FPN和SE-Net的组合神经网络
我们下面直接看模型代码和结果
from keras import layers
from keras.layers import Input, Activation, BatchNormalization, Flatten, Dropout, Reshape
from keras.layers import Dense, Conv2D, MaxPooling2D, ZeroPadding2D, AveragePooling2D, GlobalAveragePooling2D
from keras.models import Model
from keras.layers import Add, UpSampling2D, Multiply, Concatenate
# Squeeze-and-Excitation Block
def squeeze_excite_block(input_tensor, ratio=16):
channel_axis = -1 # 通道轴通常在最后一维
channels = input_tensor.shape[channel_axis] # 获取通道数
# Squeeze：全局平均池化
x = GlobalAveragePooling2D()(input_tensor)
x = Reshape((1, 1, channels))(x)
# Excite：两个全连接层生成通道权重
x = Dense(channels // ratio, activation='relu', kernel_initializer='he_normal', use_bias=False)(x)
x = Dense(channels, activation='sigmoid', kernel_initializer='he_normal', use_bias=False)(x)
# 将生成的权重与输入张量相乘
x = Multiply()([input_tensor, x])
return x
# Identity Block
def identity_block(input_tensor, kernel_size, filters, stage, block):
filters1, filters2, filters3 = filters
name_base = str(stage) + block + '_identity_block_'
# 第一卷积层
x = Conv2D(filters1, (1, 1), name=name_base + 'conv1')(input_tensor)
x = BatchNormalization(name=name_base + 'bn1')(x)
x = Activation('relu', name=name_base + 'relu1')(x)
# 第二卷积层
x = Conv2D(filters2, kernel_size, padding='same', name=name_base + 'conv2')(x)
x = BatchNormalization(name=name_base + 'bn2')(x)
x = Activation('relu', name=name_base + 'relu2')(x)
# 第三卷积层
x = Conv2D(filters3, (1, 1), name=name_base + 'conv3')(x)
x = BatchNormalization(name=name_base + 'bn3')(x)
# SE-Net通道注意力机制
x = squeeze_excite_block(x)
# 残差连接
x = layers.add([x, input_tensor], name=name_base + 'add')
x = Activation('relu', name=name_base + 'relu4')(x)
return x
# Conv Block
def conv_block(input_tensor, kernel_size, filters, stage, block, strides=(2, 2)):
filters1, filters2, filters3 = filters
res_name_base = str(stage) + block + '_conv_block_res_'
name_base = str(stage) + block + '_conv_block_'
# 主卷积层
x = Conv2D(filters1, (1, 1), strides=strides, name=name_base + 'conv1')(input_tensor)
x = BatchNormalization(name=name_base + 'bn1')(x)
x = Activation('relu', name=name_base + 'relu1')(x)
x = Conv2D(filters2, kernel_size, padding='same', name=name_base + 'conv2')(x)
x = BatchNormalization(name=name_base + 'bn2')(x)
x = Activation('relu', name=name_base + 'relu2')(x)
x = Conv2D(filters3, (1, 1), name=name_base + 'conv3')(x)
x = BatchNormalization(name=name_base + 'bn3')(x)
# 残差连接的卷积
shortcut = Conv2D(filters3, (1, 1), strides=strides, name=res_name_base + 'conv')(input_tensor)
shortcut = BatchNormalization(name=res_name_base + 'bn')(shortcut)
# SE-Net通道注意力机制
x = squeeze_excite_block(x)
# 残差连接加和
x = layers.add([x, shortcut], name=name_base + 'add')
x = Activation('relu', name=name_base + 'relu4')(x)
return x
# Feature Pyramid Network (FPN)
def fpn_block(input_tensor, filters, name):
# 上采样
x = UpSampling2D(size=(2, 2), name=name + '_upsample')(input_tensor)
x = Conv2D(filters, (1, 1), padding='same', name=name + '_conv')(x)
x = BatchNormalization(name=name + '_bn')(x)
return x
def ResNet50_with_FPN(input_shape=[224,224,3], classes=4):
img_input = Input(shape=input_shape)
x = ZeroPadding2D((3, 3))(img_input)
# ResNet50 Backbone with SE Block
x = Conv2D(64, (7, 7), strides=(2, 2), name='conv1')(x)
x = BatchNormalization(name='bn_conv1')(x)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)
# ResNet50 Backbone with SE Block
c2 = conv_block(x, 3, [64, 64, 256], stage=2, block='a', strides=(1, 1))
c2 = identity_block(c2, 3, [64, 64, 256], stage=2, block='b')
c2 = identity_block(c2, 3, [64, 64, 256], stage=2, block='c')
c3 = conv_block(c2, 3, [128, 128, 512], stage=3, block='a')
c3 = identity_block(c3, 3, [128, 128, 512], stage=3, block='b')
c3 = identity_block(c3, 3, [128, 128, 512], stage=3, block='c')
c3 = identity_block(c3, 3, [128, 128, 512], stage=3, block='d')
c4 = conv_block(c3, 3, [256, 256, 1024], stage=4, block='a')
c4 = identity_block(c4, 3, [256, 256, 1024], stage=4, block='b')
c4 = identity_block(c4, 3, [256, 256, 1024], stage=4, block='c')
c4 = identity_block(c4, 3, [256, 256, 1024], stage=4, block='d')
c4 = identity_block(c4, 3, [256, 256, 1024], stage=4, block='e')
c4 = identity_block(c4, 3, [256, 256, 1024], stage=4, block='f')
c5 = conv_block(c4, 3, [512, 512, 2048], stage=5, block='a')
c5 = identity_block(c5, 3, [512, 512, 2048], stage=5, block='b')
c5 = identity_block(c5, 3, [512, 512, 2048], stage=5, block='c')
# FPN Layer: create top-down and lateral connections
p5 = Conv2D(256, (1, 1), padding='same', name='fpn_p5')(c5)
p4 = fpn_block(p5, 256, 'fpn_p4')
p3 = fpn_block(p4, 256, 'fpn_p3')
p2 = fpn_block(p3, 256, 'fpn_p2')
# Final classification
x = AveragePooling2D((7, 7), name='avg_pool')(c5)
x = Flatten()(x)
x = Dropout(0.5)(x)
x = Dense(classes, activation='softmax', name='fc2')(x)
model = Model(img_input, x, name='resnet50_with_fpn')
return model
# Instantiate the model
model = ResNet50_with_FPN()
model.summary()
Epoch 1/30 57/57 [==============================] - 124s 2s/step - loss: 1.7040 - accuracy: 0.5465 - val_loss: 4.1759 - val_accuracy: 0.2655 Epoch 2/30 57/57 [==============================] - 108s 2s/step - loss: 1.0829 - accuracy: 0.6615 - val_loss: 6.2262 - val_accuracy: 0.2655 Epoch 3/30 57/57 [==============================] - 110s 2s/step - loss: 0.8514 - accuracy: 0.7102 - val_loss: 7.9029 - val_accuracy: 0.2655 Epoch 4/30 57/57 [==============================] - 108s 2s/step - loss: 0.6071 - accuracy: 0.7743 - val_loss: 5.6822 - val_accuracy: 0.2655 Epoch 5/30 57/57 [==============================] - 107s 2s/step - loss: 0.6340 - accuracy: 0.7854 - val_loss: 2.8168 - val_accuracy: 0.2832 Epoch 6/30 57/57 [==============================] - 107s 2s/step - loss: 0.3986 - accuracy: 0.8562 - val_loss: 2.2164 - val_accuracy: 0.3894 Epoch 7/30 57/57 [==============================] - 108s 2s/step - loss: 0.4435 - accuracy: 0.8518 - val_loss: 1.9530 - val_accuracy: 0.5841 Epoch 8/30 57/57 [==============================] - 112s 2s/step - loss: 0.3334 - accuracy: 0.8650 - val_loss: 0.8318 - val_accuracy: 0.7522 Epoch 9/30 57/57 [==============================] - 109s 2s/step - loss: 0.2631 - accuracy: 0.9027 - val_loss: 0.3123 - val_accuracy: 0.8938 Epoch 10/30 57/57 [==============================] - 121s 2s/step - loss: 0.1918 - accuracy: 0.9336 - val_loss: 0.3432 - val_accuracy: 0.8584 Epoch 11/30 57/57 [==============================] - 111s 2s/step - loss: 0.2280 - accuracy: 0.9270 - val_loss: 3.3803 - val_accuracy: 0.6018 Epoch 12/30 57/57 [==============================] - 112s 2s/step - loss: 0.2377 - accuracy: 0.9137 - val_loss: 1.7401 - val_accuracy: 0.6283 Epoch 13/30 57/57 [==============================] - 112s 2s/step - loss: 0.2143 - accuracy: 0.9204 - val_loss: 1.9620 - val_accuracy: 0.6460 Epoch 14/30 57/57 [==============================] - 108s 2s/step - loss: 0.1334 - accuracy: 0.9602 - val_loss: 1.1494 - val_accuracy: 0.8142 Epoch 15/30 57/57 [==============================] - 107s 2s/step - loss: 0.2434 - accuracy: 0.9226 - val_loss: 1.5473 - val_accuracy: 0.7699 Epoch 16/30 57/57 [==============================] - 107s 2s/step - loss: 0.1815 - accuracy: 0.9425 - val_loss: 0.6820 - val_accuracy: 0.8230 Epoch 17/30 57/57 [==============================] - 107s 2s/step - loss: 0.1237 - accuracy: 0.9580 - val_loss: 0.5633 - val_accuracy: 0.8938 Epoch 18/30 57/57 [==============================] - 107s 2s/step - loss: 0.0653 - accuracy: 0.9779 - val_loss: 0.7569 - val_accuracy: 0.8496 Epoch 19/30 57/57 [==============================] - 106s 2s/step - loss: 0.1139 - accuracy: 0.9558 - val_loss: 5.6673 - val_accuracy: 0.4956 Epoch 20/30 57/57 [==============================] - 108s 2s/step - loss: 0.1208 - accuracy: 0.9535 - val_loss: 10.7703 - val_accuracy: 0.3628 Epoch 21/30 57/57 [==============================] - 107s 2s/step - loss: 0.1212 - accuracy: 0.9668 - val_loss: 184.4987 - val_accuracy: 0.2743 Epoch 22/30 57/57 [==============================] - 106s 2s/step - loss: 0.3250 - accuracy: 0.8938 - val_loss: 44.9149 - val_accuracy: 0.3451 Epoch 23/30 57/57 [==============================] - 106s 2s/step - loss: 0.1613 - accuracy: 0.9381 - val_loss: 3.1342 - val_accuracy: 0.7522 Epoch 24/30 57/57 [==============================] - 108s 2s/step - loss: 0.1241 - accuracy: 0.9558 - val_loss: 1.4571 - val_accuracy: 0.8142 Epoch 25/30 57/57 [==============================] - 108s 2s/step - loss: 0.1052 - accuracy: 0.9558 - val_loss: 0.5071 - val_accuracy: 0.8673 Epoch 26/30 57/57 [==============================] - 108s 2s/step - loss: 0.1868 - accuracy: 0.9403 - val_loss: 0.5222 - val_accuracy: 0.8584 Epoch 27/30 57/57 [==============================] - 107s 2s/step - loss: 0.0699 - accuracy: 0.9801 - val_loss: 1.3800 - val_accuracy: 0.6991 Epoch 28/30 57/57 [==============================] - 106s 2s/step - loss: 0.0975 - accuracy: 0.9712 - val_loss: 4.8408 - val_accuracy: 0.6372 Epoch 29/30 57/57 [==============================] - 108s 2s/step - loss: 0.0452 - accuracy: 0.9801 - val_loss: 0.3862 - val_accuracy: 0.8761 Epoch 30/30 57/57 [==============================] - 107s 2s/step - loss: 0.0164 - accuracy: 0.9978 - val_loss: 0.2110 - val_accuracy: 0.9292
