-
Notifications
You must be signed in to change notification settings - Fork 9
/
model.py
186 lines (148 loc) · 5.92 KB
/
model.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
# @Date: 2019-12-23T11:32:33+01:00
# @Last modified time: 2020-04-24T10:31:27+02:00
# @Date: 2019-12-23T11:24:52+01:00
# @Last modified time: 2020-04-24T10:31:27+02:00
"""
sen2LCZ, the exact architecture can be found in ./modelFig
Reference: https://ieeexplore.ieee.org/document/9103196
Multilevel Feature Fusion-Based CNN for Local Climate Zone Classification From Sentinel-2 Images: Benchmark Results on the So2Sat LCZ42 Dataset
"""
from __future__ import print_function
import keras
from keras.layers import *
from keras.regularizers import l2
from keras.models import Model
def sen2LCZ_drop_core(inputs, num_classes=17, bn=1, depth=5, dim=16, dropRate=0.1, fusion=0):
# Start model definition.
inc_rate = 2
lay_per_block=int((depth-1)/4)
'32*32'
conv0 = Conv2D(dim, (3, 3), padding='same', kernel_initializer = 'he_normal')(inputs)
if bn==1:
print('with BN')
conv0 = BatchNormalization(axis=-1)(conv0)
conv0 = Activation('relu')(conv0)
for i in np.arange(lay_per_block-1):
print(str(i) +'in' +str(lay_per_block-1), '!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!')
conv0 = Conv2D(dim, (3, 3), padding='same', kernel_initializer = 'he_normal')(conv0)
if bn==1:
print('with BN')
conv0 = BatchNormalization(axis=-1)(conv0)
conv0 = Activation('relu')(conv0)
"how to pooling?!"
#############################################
# merge0 = MaxPooling2D((2, 2))(conv0)
"original idea"
pool0 = MaxPooling2D((2, 2))(conv0)
pool1 = AveragePooling2D(pool_size=2)(conv0)
merge0 = Concatenate()([pool0,pool1])
#############################################
if fusion==1:
'prediction'
x = GlobalAveragePooling2D()(merge0)#Flatten
print(x.shape)
outputs_32 = Dense(num_classes,
activation='softmax',
kernel_initializer='he_normal')(x)
'16*16'
dim=dim*inc_rate
conv1 = Conv2D(dim, (3, 3), padding='same', kernel_initializer = 'he_normal')(merge0)
if bn==1:
print('with BN')
conv1 = BatchNormalization(axis=-1)(conv1)
conv1 = Activation('relu')(conv1)
for i in np.arange(lay_per_block-1):
print(str(i) +'in' +str(lay_per_block-1), '!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!')
conv1 = Conv2D(dim, (3, 3), padding='same', kernel_initializer = 'he_normal')(conv1)
if bn==1:
print('with BN')
conv1 = BatchNormalization(axis=-1)(conv1)
conv1 = Activation('relu')(conv1)
"how to pooling?!"
#############################################
# merge1 = MaxPooling2D((2, 2))(conv1)
"original idea"
pool0 = MaxPooling2D((2, 2))(conv1)
pool1 = AveragePooling2D(pool_size=2)(conv1)
merge1 = Concatenate()([pool0,pool1])
#############################################
'dropOut'
merge1 = Dropout(dropRate)(merge1)
if fusion==1:
'prediction'
x = GlobalAveragePooling2D()(merge1)#Flatten
print(x.shape)
outputs_16 = Dense(num_classes,
activation='softmax',
kernel_initializer='he_normal')(x)
'8*8'
dim=dim*inc_rate
conv2 = Conv2D(dim, (3, 3), padding='same', kernel_initializer = 'he_normal')(merge1)
if bn==1:
print('with BN')
conv2 = BatchNormalization(axis=-1)(conv2)
conv2 = Activation('relu')(conv2)
for i in np.arange(lay_per_block-1):
print(str(i) +'in' +str(lay_per_block-1), '!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!')
conv2 = Conv2D(dim, (3, 3), padding='same', kernel_initializer = 'he_normal')(conv2)
if bn==1:
print('with BN')
conv2 = BatchNormalization(axis=-1)(conv2)
conv2 = Activation('relu')(conv2)
"how to pooling?!"
#############################################
# merge2 = MaxPooling2D((2, 2))(conv2)
"original idea"
pool0 = MaxPooling2D((2, 2))(conv2)
pool1 = AveragePooling2D(pool_size=2)(conv2)
merge2 = Concatenate()([pool0,pool1])
#############################################
'dropOut'
merge2 = Dropout(dropRate)(merge2)
if fusion==1:
'prediction'
x = GlobalAveragePooling2D()(merge2)#Flatten
print(x.shape)
outputs_8 = Dense(num_classes,
activation='softmax',
kernel_initializer='he_normal')(x)
'4*4'
dim=dim*inc_rate
conv3 = Conv2D(dim, (3, 3), padding='same', kernel_initializer = 'he_normal')(merge2)
if bn==1:
print('with BN')
conv3 = BatchNormalization(axis=-1)(conv3)
conv3 = Activation('relu')(conv3)
for i in np.arange(lay_per_block-1):
print(str(i) +'in' +str(lay_per_block-1), '!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!')
conv3 = Conv2D(dim, (3, 3), padding='same', kernel_initializer = 'he_normal')(conv3)
if bn==1:
print('with BN')
conv3 = BatchNormalization(axis=-1)(conv3)
conv3 = Activation('relu')(conv3)
'prediction'
x = GlobalAveragePooling2D()(conv3)#Flatten
print(x.shape)
outputs = Dense(num_classes,
activation='softmax',
kernel_initializer='he_normal')(x)
if fusion==1:
'final prediction'
# x = GlobalAveragePooling2D()(merge2)#Flatten
# print(x.shape)
# outputs_8 = Dense(num_classes,
# activation='softmax',
# kernel_initializer='he_normal')(x)
o=outputs=Average()([outputs, outputs_32, outputs_16, outputs_8])
else:
o=outputs
return o
def sen2LCZ_drop(input_shape=(32,32,10), num_classes=17, bn=1, depth=5, dim=16, dropRate=0.1, fusion=0):
"""
# Arguments
# Returns
model (Model): Keras model instance
"""
inputs = Input(shape=input_shape)
o=sen2LCZ_drop_core(inputs, num_classes=num_classes, bn=bn, depth=depth, dim=dim, dropRate=dropRate, fusion=fusion)
return Model(inputs=inputs, outputs=o)