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
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
| """
VOC 红外光谱深度学习识别系统
论文核心：
1. FTIR实验采集9种VOC光谱
2. CVAE合成增强训练数据
3. CNN分类器：识别VOC类别 + 预测浓度
4. 两头设计：分类头 + 回归头
"""
import torch
import torch.nn as nn
import numpy as np
from typing import Tuple, Optional
class VOCSpectraClassifier(nn.Module):
"""
VOC红外光谱分类器
两头设计：
- 分类头：9类VOC识别
- 回归头：浓度预测
输入：FTIR吸收光谱（400-5000 cm^-1, 2301点）
"""
def __init__(
self,
input_length: int = 2301,
num_classes: int = 9,
hidden_dim: int = 256
):
super().__init__()
self.feature_extractor = nn.Sequential(
nn.Conv1d(1, 32, kernel_size=7, stride=2, padding=3),
nn.BatchNorm1d(32),
nn.ReLU(),
nn.MaxPool1d(2),
nn.Conv1d(32, 64, kernel_size=5, stride=1, padding=2),
nn.BatchNorm1d(64),
nn.ReLU(),
nn.MaxPool1d(2),
nn.Conv1d(64, 128, kernel_size=3, stride=1, padding=1),
nn.BatchNorm1d(128),
nn.ReLU(),
nn.MaxPool1d(2),
nn.Conv1d(128, hidden_dim, kernel_size=3, stride=1, padding=1),
nn.BatchNorm1d(hidden_dim),
nn.ReLU(),
nn.AdaptiveAvgPool1d(1)
)
self.classification_head = nn.Sequential(
nn.Linear(hidden_dim, hidden_dim // 2),
nn.ReLU(),
nn.Dropout(0.3),
nn.Linear(hidden_dim // 2, num_classes)
)
self.regression_head = nn.Sequential(
nn.Linear(hidden_dim, hidden_dim // 2),
nn.ReLU(),
nn.Dropout(0.3),
nn.Linear(hidden_dim // 2, 1),
nn.Softplus()
)
def forward(self, spectra: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
"""
Args:
spectra: (B, 1, L) FTIR吸收光谱
Returns:
logits: (B, 9) VOC分类
concentration: (B, 1) 浓度预测 (ppm)
"""
features = self.feature_extractor(spectra)
features = features.squeeze(-1)
logits = self.classification_head(features)
concentration = self.regression_head(features)
return logits, concentration
class ConditionalVAE(nn.Module):
"""
条件变分自编码器 (CVAE)
生成合成VOC光谱用于数据增强
条件：VOC类别 + 浓度
"""
def __init__(
self,
input_length: int = 2301,
latent_dim: int = 128,
num_classes: int = 9,
cond_dim: int = 32
):
super().__init__()
self.latent_dim = latent_dim
self.class_embed = nn.Embedding(num_classes, cond_dim)
self.conc_embed = nn.Sequential(
nn.Linear(1, cond_dim),
nn.ReLU()
)
self.encoder = nn.Sequential(
nn.Conv1d(1, 32, 7, 2, 3),
nn.ReLU(),
nn.Conv1d(32, 64, 5, 2, 2),
nn.ReLU(),
nn.Conv1d(64, 128, 3, 2, 1),
nn.ReLU(),
nn.AdaptiveAvgPool1d(1),
)
total_cond = cond_dim * 2
self.fc_mu = nn.Linear(128 + total_cond, latent_dim)
fc_logvar = nn.Linear(128 + total_cond, latent_dim)
self.decoder_input = nn.Linear(latent_dim + total_cond, 128 * (input_length // 8))
self.decoder = nn.Sequential(
nn.ConvTranspose1d(128, 64, 4, 2, 1),
nn.ReLU(),
nn.ConvTranspose1d(64, 32, 4, 2, 1),
nn.ReLU(),
nn.ConvTranspose1d(32, 1, 4, 2, 1),
)
def encode(self, x, class_idx, conc):
h = self.encoder(x).squeeze(-1)
c = torch.cat([
self.class_embed(class_idx),
self.conc_embed(conc.unsqueeze(-1))
], dim=-1)
h = torch.cat([h, c], dim=-1)
return self.fc_mu(h), self.fc_logvar(h)
def reparameterize(self, mu, logvar):
std = torch.exp(0.5 * logvar)
eps = torch.randn_like(std)
return mu + eps * std
def decode(self, z, class_idx, conc):
c = torch.cat([
self.class_embed(class_idx),
self.conc_embed(conc.unsqueeze(-1))
], dim=-1)
h = self.decoder_input(torch.cat([z, c], dim=-1))
h = h.view(-1, 128, -1)
return self.decoder(h)
def forward(self, x, class_idx, conc):
mu, logvar = self.encode(x, class_idx, conc)
z = self.reparameterize(mu, logvar)
recon = self.decode(z, class_idx, conc)
return recon, mu, logvar
def generate(self, class_idx, conc, device='cpu'):
"""生成指定类别和浓度的合成光谱"""
z = torch.randn(class_idx.size(0), self.latent_dim).to(device)
return self.decode(z, class_idx, conc)
if __name__ == "__main__":
torch.manual_seed(42)
spectra = torch.randn(4, 1, 2301)
classifier = VOCSpectraClassifier()
logits, conc = classifier(spectra)
print("VOC分类器:")
print(f" 输入: {spectra.shape}")
print(f" 分类logits: {logits.shape}")
print(f" 浓度预测: {conc.shape}")
cvae = ConditionalVAE()
class_idx = torch.tensor([0, 1, 2, 3])
conc_vals = torch.tensor([5.0, 10.0, 3.0, 20.0])
synthetic = cvae.generate(class_idx, conc_vals)
print(f"\nCVAE合成光谱: {synthetic.shape}")
print(f" 可生成指定VOC类别+浓度的合成光谱")
params = sum(p.numel() for p in classifier.parameters())
print(f"\n分类器参数: {params:,} ({params/1e6:.2f}M)")
|
