-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathModel.py
128 lines (123 loc) · 6.78 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
# 卞庆朝
# 开发时间：2024/4/17 11:59
import torch
import torch.nn as nn
# 一维卷积模块
class CNN_1D(nn.Module):
def __init__(self):
super(CNN_1D, self).__init__()
self.conv1 = nn.Conv1d(in_channels=7, out_channels=32, kernel_size=2, stride=1,padding=1)
self.pool1 = nn.MaxPool1d(kernel_size=3, stride=1,padding=1)
self.conv2 = nn.Conv1d(in_channels=32, out_channels=32, kernel_size=3, stride=1,padding=1)
self.pool2 = nn.MaxPool1d(kernel_size=4, stride=1,padding=1)
self.conv3 = nn.Conv1d(in_channels=32, out_channels=7, kernel_size=3, stride=1,padding=1)
self.pool3 = nn.MaxPool1d(kernel_size=3, stride=1,padding=1)
# self.conv4 = nn.Conv1d(in_channels=32, out_channels=16, kernel_size=2, stride=1)
# self.pool4 = nn.MaxPool1d(kernel_size=2, stride=2)
# self.conv5 = nn.Conv1d(in_channels=16, out_channels=16, kernel_size=2, stride=1)
# self.pool5 = nn.MaxPool1d(kernel_size=3, stride=3)
# self.conv6 = nn.Conv1d(in_channels=16, out_channels=16, kernel_size=2, stride=1)
# self.pool6 = nn.MaxPool1d(kernel_size=3, stride=3)
self.net = nn.Sequential(self.conv1,nn.ReLU(),self.pool1,
self.conv2,nn.ReLU(),self.pool2,
self.conv3,nn.ReLU(),self.pool3)
# self.fc = nn.Sequential(nn.Linear(in_features=224, out_features=1), nn.ReLU())
def forward(self, x):
print('Conv1D输入的x是:',x.shape)
x = self.net(x)
print('CNN_1D输出x的形状:',x.shape)
# print('x.size(0)的形状:',type(x.size(0)),x.size(0))
# print('x.view(x.size(0), -1)的形状:',x.view(x.size(0), -1).shape)
# view()的作用相当于numpy中的reshape，重新定义矩阵的形状
# out = self.fc(x.view(x.size(0), -1))#view中一个参数定为-1，代表动态调整这个维度上的元素个数，以保证元素的总数不变
# print('out的形状:',out.shape)
return x
# 通道注意力模块
class ChannelAttentionModule(nn.Module):
# def __init__(self, channel=256, reduction=16):
# def __init__(self, channel=10, reduction=2):
def __init__(self, channel=10, reduction=2):
print('传参成功了吗？',channel)
super(ChannelAttentionModule, self).__init__()
mid_channel = channel // reduction #mid_channel==5
self.avg_pool = nn.AdaptiveAvgPool2d(1)#平均池化，使用自适应池化缩减map的大小，保持通道不变
self.max_pool = nn.AdaptiveMaxPool2d(1)#最大池化，
# 两个卷积层用于从池化后的特征中学习注意力权重
self.shared_MLP = nn.Sequential(
nn.Linear(in_features=channel, out_features=mid_channel),
nn.ReLU(),
nn.Linear(in_features=mid_channel, out_features=channel)
)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
print('channel的输入:',x.shape)
avgout = self.shared_MLP(self.avg_pool(x).view(x.size(0), -1)).unsqueeze(2).unsqueeze(3)#对平均池化的特征进行处理
print('avg',avgout.size())
maxout = self.shared_MLP(self.max_pool(x).view(x.size(0), -1)).unsqueeze(2).unsqueeze(3)#对最大池化的特征进行处理
print('max',maxout.size())
return self.sigmoid(avgout + maxout) # 将两种池化的特征加权和作为输出
# 空间注意力模块
class SpatialAttentionModule(nn.Module):
def __init__(self):
super(SpatialAttentionModule, self).__init__()
self.conv2d = nn.Conv2d(in_channels=2, out_channels=1, kernel_size=7, stride=1, padding=3)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
print('**********************')
print('spatial的输入:', x.shape)
avgout = torch.mean(x, dim=1, keepdim=True) # 平均池化，map尺寸不变，缩减通道
print('savg',avgout.size()) # [16, 1, 384, 384]
maxout, _ = torch.max(x, dim=1, keepdim=True) # 最大池化
print('smax',maxout.size()) # [16, 1, 384, 384]
out = torch.cat([avgout, maxout], dim=1) # 将两种池化的特征图连接起来
print('平均+Max:',out.size()) # [16, 2, 384, 384]
out = self.sigmoid(self.conv2d(out)) # 通过卷积层处理连接后的特征图+++使用sigmoid激活函数计算注意力权重
print('ssigmoid',out.size())
return out
# CBAM模块由上述CAM和SAM模块两个组合而成
class CBAM(nn.Module):
# def __init__(self, in_channel=256, out_channels=64):
def __init__(self,in_channel=7, out_channels=7):
super(CBAM, self).__init__()
# self.channel=channel
self.cnn_1d=CNN_1D()
self.channel_attention = ChannelAttentionModule()
self.spatial_attention = SpatialAttentionModule()
self.relu = nn.ReLU(inplace=True)
# self.conv = nn.Conv2d(in_channel, out_channels, kernel_size=3, stride=1, padding=1)
self.conv1 = nn.Conv1d(in_channels=7, out_channels=32, kernel_size=7, stride=3)
self.pool1 = nn.MaxPool1d(kernel_size=3, stride=2)
self.conv2 = nn.Conv1d(in_channels=32, out_channels=out_channels, kernel_size=3, stride=2)
self.pool2 = nn.MaxPool1d(kernel_size=3, stride=2)
self.bn = nn.BatchNorm1d(num_features=out_channels)
self.fc = nn.Sequential(nn.Linear(in_features=1449, out_features=1), nn.ReLU())
def forward(self, x):
print('CBAM输入的x是:', x.shape)
out=self.cnn_1d(x)
out=out.unsqueeze(0)# 增加一个维度
print('cnn1D增加一个维度的输出out是:', out.shape)
print('cnn1D增加一个维度的输出out.size(1)是:',out.size(1))
out1 = self.channel_attention(out)
print('channel_attention输出的out1是:', out1.shape)
out = out1 * out
print('out1*out乘积的输出:',out.shape)
# out = self.channel_attention(out) * out
out = self.spatial_attention(out) * out
print('spatial_attention(out)和channel_attention乘积的输出:',out.shape)
residual = x
residual = residual.unsqueeze(0) # 增加一个维度
print('Residual增加一个维度的Residual是:', residual.shape)
out = residual + out
print('res+channel+spatial=out的输出:',out.size())
out=torch.squeeze(out, dim=0)
print('新的维度:',out.shape)
out = self.relu(out)
out = self.conv1(out)
out=self.pool1(out)
out = self.conv2(out)
out=self.pool2(out)
print('conv12(out)+pool12(out)输出后的维度:',out.shape)
print('out.view(x.size(0), -1)输出后的维度:',out.view(x.size(0), -1).shape)
out = self.fc(out.view(x.size(0), -1))#view中一个参数定为-1，代表动态调整这个维度上的元素个数，以保证元素的总数不变
# out = self.bn(out)
return out