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
| """
CNN-LSTM 混合架构疲劳检测
"""
import torch
import torch.nn as nn
import torchvision.models as models
class CNNLSTMFatigueDetector(nn.Module):
"""
CNN-LSTM混合疲劳检测器
架构：
1. CNN提取每帧空间特征
2. LSTM建模时序依赖
3. 注意力机制聚焦关键时刻
"""
def __init__(self,
cnn_backbone: str = 'resnet18',
lstm_hidden: int = 256,
lstm_layers: int = 2,
num_classes: int = 2,
pretrained: bool = True):
super().__init__()
if cnn_backbone == 'resnet18':
self.cnn = models.resnet18(pretrained=pretrained)
feature_dim = 512
elif cnn_backbone == 'mobilenetv2':
self.cnn = models.mobilenet_v2(pretrained=pretrained)
feature_dim = 1280
else:
raise ValueError(f"不支持的骨干网络: {cnn_backbone}")
self.cnn.fc = nn.Identity()
self.lstm = nn.LSTM(
input_size=feature_dim,
hidden_size=lstm_hidden,
num_layers=lstm_layers,
batch_first=True,
bidirectional=True,
dropout=0.3
)
self.attention = nn.Sequential(
nn.Linear(lstm_hidden * 2, 64),
nn.Tanh(),
nn.Linear(64, 1)
)
self.classifier = nn.Sequential(
nn.Linear(lstm_hidden * 2, 64),
nn.ReLU(),
nn.Dropout(0.3),
nn.Linear(64, num_classes)
)
def forward(self, x: torch.Tensor) -> torch.Tensor:
"""
前向传播
Args:
x: 视频帧序列 (B, T, 3, H, W)
Returns:
logits: 分类输出 (B, num_classes)
"""
batch_size, seq_len = x.size(0), x.size(1)
x_flat = x.view(batch_size * seq_len, *x.size()[2:])
cnn_features = self.cnn(x_flat)
cnn_features = cnn_features.view(batch_size, seq_len, -1)
lstm_out, _ = self.lstm(cnn_features)
attn_scores = self.attention(lstm_out)
attn_weights = torch.softmax(attn_scores, dim=1)
context = torch.sum(lstm_out * attn_weights, dim=1)
logits = self.classifier(context)
return logits
def get_temporal_attention(self, x: torch.Tensor) -> torch.Tensor:
"""获取时间注意力权重（可视化用）"""
batch_size, seq_len = x.size(0), x.size(1)
x_flat = x.view(batch_size * seq_len, *x.size()[2:])
cnn_features = self.cnn(x_flat)
cnn_features = cnn_features.view(batch_size, seq_len, -1)
lstm_out, _ = self.lstm(cnn_features)
attn_scores = self.attention(lstm_out)
attn_weights = torch.softmax(attn_scores, dim=1)
return attn_weights.squeeze(-1)
if __name__ == "__main__":
model = CNNLSTMFatigueDetector(
cnn_backbone='resnet18',
lstm_hidden=256,
lstm_layers=2
)
x = torch.randn(4, 10, 3, 224, 224)
logits = model(x)
attn = model.get_temporal_attention(x)
print(f"输入形状: {x.shape}")
print(f"输出形状: {logits.shape}")
print(f"注意力权重形状: {attn.shape}")
print(f"注意力权重样本: {attn[0].detach().numpy()}")
total_params = sum(p.numel() for p in model.parameters())
print(f"总参数量: {total_params/1e6:.2f}M")
|