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
| """
Driver-Intention Prediction CNN
论文: arXiv 2601.05084 (2026年1月)
核心方法：
1. 64通道原始EEG（512Hz采样）
2. 无传统预处理（无滤波/无ICA/无特征提取）
3. CNN直接从时域EEG学习
4. 3类分类：左转/右转/直行
关键发现：
- 83.7%准确率（3类）
- 右转分类最准确（空间偏侧效应）
- CNN能从噪声EEG中提取有效特征
- 深度学习优于传统频谱特征方法
"""
import torch
import torch.nn as nn
import numpy as np
from typing import Tuple
class EEGDriverIntentionCNN(nn.Module):
"""
驾驶员意图预测 CNN
论文架构复现：
- 输入：64通道 × T时间步原始EEG
- 网络结构：Conv1D → Conv1D → Pool → FC → Softmax
- 输出：3类（左转/右转/直行）
关键设计决策（来自论文）：
1. 不使用传统EEG预处理（滤波/ICA）
2. CNN直接学习时域模式
3. 原始信号→深度学习端到端
4. 64通道全保留（不选通道）
"""
def __init__(self, num_channels: int = 64, num_classes: int = 3,
sampling_rate: int = 512, segment_sec: float = 2.0):
super().__init__()
self.num_channels = num_channels
self.segment_samples = int(sampling_rate * segment_sec)
self.conv1 = nn.Conv1d(
in_channels=num_channels,
out_channels=32,
kernel_size=64,
stride=4,
padding=30
)
self.bn1 = nn.BatchNorm1d(32)
self.relu1 = nn.ReLU()
self.pool1 = nn.MaxPool1d(kernel_size=4, stride=4)
self.conv2 = nn.Conv1d(32, 64, kernel_size=16, stride=2, padding=7)
self.bn2 = nn.BatchNorm1d(64)
self.relu2 = nn.ReLU()
self.pool2 = nn.MaxPool1d(kernel_size=4, stride=4)
self.conv3 = nn.Conv1d(64, 128, kernel_size=8, stride=1, padding=3)
self.bn3 = nn.BatchNorm1d(128)
self.relu3 = nn.ReLU()
self.pool3 = nn.AdaptiveAvgPool1d(1)
self.fc1 = nn.Linear(128, 64)
self.relu4 = nn.ReLU()
self.dropout = nn.Dropout(0.5)
self.fc2 = nn.Linear(64, num_classes)
def forward(self, x):
"""
Args:
x: (B, 64, T) 原始EEG信号
Returns:
logits: (B, 3) 类别logits
"""
x = self.pool1(self.relu1(self.bn1(self.conv1(x))))
x = self.pool2(self.relu2(self.bn2(self.conv2(x))))
x = self.pool3(self.relu3(self.bn3(self.conv3(x))))
x = x.view(x.size(0), -1)
x = self.dropout(self.relu4(self.fc1(x)))
x = self.fc2(x)
return x
def predict_intention(self, eeg_segment: np.ndarray) -> dict:
"""
实时意图预测
Args:
eeg_segment: (64, T) EEG段
Returns:
prediction: 意图分类+置信度
"""
x = torch.FloatTensor(eeg_segment).unsqueeze(0)
with torch.no_grad():
logits = self.forward(x)
probs = torch.softmax(logits, dim=1)
predicted = torch.argmax(probs, dim=1).item()
confidence = probs[0, predicted].item()
intentions = ['left_turn', 'right_turn', 'straight']
return {
'intention': intentions[predicted],
'confidence': confidence,
'all_probs': {
intentions[i]: probs[0, i].item()
for i in range(len(intentions))
},
}
if __name__ == "__main__":
np.random.seed(42)
model = EEGDriverIntentionCNN(
num_channels=64, num_classes=3,
sampling_rate=512, segment_sec=2.0
)
eeg_segment = np.random.randn(64, 1024) * 10
prediction = model.predict_intention(eeg_segment)
print(f"=== EEG 驾驶意图预测 ===")
print(f"输入: 64通道 × 1024样本 (2秒@512Hz)")
print(f"预测意图: {prediction['intention']}")
print(f"置信度: {prediction['confidence']:.3f}")
print(f"各类概率:")
for intent, prob in prediction['all_probs'].items():
print(f" {intent}: {prob:.3f}")
print(f"\n=== 论文报告性能 ===")
print(f"总准确率: 83.7% (3类分类)")
print(f"右转准确率: 最高 (空间偏侧效应)")
print(f"左转准确率: 中等")
print(f"直行准确率: 最低 (基线状态难区分)")
print(f"被试: 10名 (1男9女, 22-44岁)")
print(f"EEG: 64通道 gel cap, 512Hz")
print(f"预处理: 极少 (CNN端到端)")
total_params = sum(p.numel() for p in model.parameters())
print(f"\n模型参数量: {total_params:,}")
print(f"模型大小: ~{total_params * 4 / 1024 / 1024:.1f} MB (float32)")
|
