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
211
212
213
214
215
216
217
218
| import numpy as np
from scipy.signal import welch, butter, filtfilt
from dataclasses import dataclass
@dataclass
class rPPGSignal:
"""rPPG信号段"""
raw: np.ndarray
fps: int
duration_sec: float
class SignalQualityAssessor:
"""
rPPG信号质量机器学习评估
论文方法：从rPPG信号中提取质量指标，
用机器学习分类器判定信号质量等级
"""
def __init__(self, fs: int = 30):
self.fs = fs
def extract_quality_features(self,
rppg: np.ndarray) -> dict:
"""
提取信号质量特征
Args:
rppg: [N] rPPG信号段
Returns:
features: dict of quality metrics
"""
N = len(rppg)
freqs, psd = welch(rppg, fs=self.fs, nperseg=min(256, N))
hr_mask = (freqs >= 0.7) & (freqs <= 4.0)
hr_power = np.sum(psd[hr_mask])
total_power = np.sum(psd) + 1e-8
spectral_ratio = hr_power / total_power
if np.any(psd[hr_mask] > 0):
peak_freq = freqs[hr_mask][np.argmax(psd[hr_mask])]
else:
peak_freq = 0
peak_idx = np.argmax(psd[hr_mask])
peak_width = self._peak_width(freqs[hr_mask],
psd[hr_mask], peak_idx)
spectral_concentration = 1.0 / (1.0 + peak_width)
amplitude = np.std(rppg)
zero_crossings = np.sum(np.diff(np.sign(rppg)) != 0)
zcr = zero_crossings / N
noise_power = total_power - hr_power
snr = 10 * np.log10(hr_power / (noise_power + 1e-8))
motion_mask = freqs > 4.0
motion_power = np.sum(psd[motion_mask])
motion_ratio = motion_power / total_power
t = np.arange(N)
trend_coef = np.polyfit(t, rppg, 1)[0]
trend_ratio = abs(trend_coef * N) / (np.std(rppg) + 1e-8)
autocorr = np.correlate(rppg - np.mean(rppg),
rppg - np.mean(rppg),
mode='full')
autocorr = autocorr[N-1:] / autocorr[N-1]
periodicity = 0
for lag in range(int(0.5 * self.fs), int(3.0 * self.fs)):
if lag < len(autocorr):
if autocorr[lag] > periodicity:
periodicity = autocorr[lag]
return {
'spectral_ratio': spectral_ratio,
'peak_freq': peak_freq,
'spectral_concentration': spectral_concentration,
'amplitude': amplitude,
'zcr': zcr,
'snr': snr,
'motion_ratio': motion_ratio,
'trend_ratio': trend_ratio,
'periodicity': periodicity,
}
def classify_quality(self, features: dict) -> dict:
"""
机器学习分类信号质量
Returns:
{'quality': str, 'confidence': float, 'usable': bool}
"""
score = 0
if features['spectral_ratio'] > 0.3:
score += 2
elif features['spectral_ratio'] > 0.15:
score += 1
if features['snr'] > 3.0:
score += 2
elif features['snr'] > 0.0:
score += 1
if features['motion_ratio'] < 0.2:
score += 2
elif features['motion_ratio'] < 0.4:
score += 1
if features['periodicity'] > 0.5:
score += 2
elif features['periodicity'] > 0.3:
score += 1
if features['trend_ratio'] < 0.5:
score += 1
if score >= 7:
quality = 'Excellent'
confidence = 0.9
usable = True
elif score >= 5:
quality = 'Good'
confidence = 0.75
usable = True
elif score >= 3:
quality = 'Fair'
confidence = 0.5
usable = False
else:
quality = 'Poor'
confidence = 0.2
usable = False
return {
'quality': quality,
'confidence': confidence,
'usable': usable,
'score': score,
}
def _peak_width(self, freqs, psd, peak_idx,
threshold=0.5):
"""计算峰宽"""
peak_val = psd[peak_idx]
half_max = peak_val * threshold
left = peak_idx
while left > 0 and psd[left] > half_max:
left -= 1
right = peak_idx
while right < len(psd) - 1 and psd[right] > half_max:
right += 1
return freqs[right] - freqs[left]
if __name__ == "__main__":
assessor = SignalQualityAssessor(fs=30)
t = np.arange(900) / 30
clean_rppg = 0.5 * np.sin(2 * np.pi * 1.2 * t)
clean_rppg += 0.05 * np.random.randn(900)
features_clean = assessor.extract_quality_features(clean_rppg)
result_clean = assessor.classify_quality(features_clean)
print("高质量信号:")
print(f" 质量: {result_clean['quality']}")
print(f" 可用: {result_clean['usable']}")
print(f" SNR: {features_clean['snr']:.1f}dB")
print(f" 周期性: {features_clean['periodicity']:.2f}")
noisy_rppg = 0.1 * np.sin(2 * np.pi * 1.2 * t)
noisy_rppg += 0.5 * np.random.randn(900)
noisy_rppg += 0.3 * np.sin(2 * np.pi * 8 * t)
features_noisy = assessor.extract_quality_features(noisy_rppg)
result_noisy = assessor.classify_quality(features_noisy)
print("\n低质量信号:")
print(f" 质量: {result_noisy['quality']}")
print(f" 可用: {result_noisy['usable']}")
print(f" SNR: {features_noisy['snr']:.1f}dB")
print(f" 运动比: {features_noisy['motion_ratio']:.2f}")
|