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
| class TemporalStatisticalFeatures:
"""
RadarMind 时序统计特征提取器
论文核心：从微多普勒谱图中提取
统计特征而非深度学习特征
优势：
1. 可解释——每个特征有物理意义
2. 轻量——无需GPU，CPU即可
3. 泛化——统计特征跨用户鲁棒
"""
def extract_all(self, spectrogram: np.ndarray) -> dict:
"""
从微多普勒谱图提取完整特征集
Args:
spectrogram: (freq_bins, time_bins)
Returns:
features: dict of feature_name → value
"""
features = {}
features.update(self._temporal_stats(spectrogram))
features.update(self._spectral_stats(spectrogram))
features.update(self._joint_tf_stats(spectrogram))
features.update(self._morphology_stats(spectrogram))
return features
def _temporal_stats(self, spec: np.ndarray) -> dict:
"""时域统计特征"""
temporal_envelope = np.sum(spec, axis=0)
return {
'temporal_mean': np.mean(temporal_envelope),
'temporal_std': np.std(temporal_envelope),
'temporal_max': np.max(temporal_envelope),
'temporal_min': np.min(temporal_envelope),
'temporal_range': np.ptp(temporal_envelope),
'temporal_skew': float(self._skewness(temporal_envelope)),
'temporal_kurtosis': float(self._kurtosis(temporal_envelope)),
'zero_crossing_rate': self._zcr(temporal_envelope),
'energy_ratio': np.sum(temporal_envelope ** 2) / (
np.sum(spec ** 2) + 1e-8
),
}
def _spectral_stats(self, spec: np.ndarray) -> dict:
"""频域统计特征"""
spectrum = np.sum(spec, axis=1)
freqs = np.arange(len(spectrum))
centroid = np.sum(freqs * spectrum) / (np.sum(spectrum) + 1e-8)
bandwidth = np.sqrt(
np.sum(((freqs - centroid) ** 2) * spectrum) /
(np.sum(spectrum) + 1e-8)
)
geometric_mean = np.exp(np.mean(np.log(spec + 1e-10)))
arithmetic_mean = np.mean(spec)
flatness = geometric_mean / (arithmetic_mean + 1e-10)
return {
'spectral_centroid': float(centroid),
'spectral_bandwidth': float(bandwidth),
'spectral_flatness': float(flatness),
'spectral_roll_off': float(self._roll_off(spectrum, 0.85)),
'spectral_flux': float(self._spectral_flux(spec)),
}
def _joint_tf_stats(self, spec: np.ndarray) -> dict:
"""时频联合特征"""
peak_idx = np.unravel_index(np.argmax(spec), spec.shape)
peak_freq, peak_time = peak_idx
spec_norm = spec / (np.sum(spec) + 1e-10)
entropy = -np.sum(spec_norm * np.log2(spec_norm + 1e-10))
sparsity = np.sum(np.abs(spec)) ** 2 / (
np.sum(spec ** 2) + 1e-10
)
return {
'tf_peak_frequency': float(peak_freq),
'tf_peak_time': float(peak_time),
'tf_entropy': float(entropy),
'tf_sparsity': float(sparsity),
'freq_modulation_rate': float(
self._frequency_modulation_rate(spec)
),
}
def _morphology_stats(self, spec: np.ndarray) -> dict:
"""微多普勒形态特征"""
binary = (spec > np.mean(spec) + np.std(spec)).astype(float)
from scipy.ndimage import label, center_of_mass
labeled, num_features = label(binary)
if num_features > 0:
areas = [np.sum(labeled == i) for i in range(1, num_features + 1)]
max_area = max(areas)
total_area = sum(areas)
centroids = center_of_mass(binary, labeled,
range(1, num_features + 1))
else:
max_area = 0
total_area = 0
centroids = []
return {
'morph_num_components': int(num_features),
'morph_max_area': float(max_area),
'morph_total_area': float(total_area),
'morph_area_ratio': float(max_area / (total_area + 1e-8)),
}
def _skewness(self, x):
m = np.mean(x)
s = np.std(x)
return np.mean(((x - m) / s) ** 3) if s > 0 else 0
def _kurtosis(self, x):
m = np.mean(x)
s = np.std(x)
return np.mean(((x - m) / s) ** 4) - 3 if s > 0 else 0
def _zcr(self, x):
return np.mean(np.diff(np.sign(x)) != 0)
def _roll_off(self, spectrum, ratio=0.85):
cumsum = np.cumsum(spectrum)
total = cumsum[-1]
idx = np.searchsorted(cumsum, ratio * total)
return idx / len(spectrum)
def _spectral_flux(self, spec):
diff = np.diff(spec, axis=1)
return np.sum(diff ** 2) / (spec.shape[1] - 1)
def _frequency_modulation_rate(self, spec):
peak_freqs = np.argmax(spec, axis=0)
return np.std(np.diff(peak_freqs)) if len(peak_freqs) > 2 else 0
def extract_features_from_radar(adc_cube: np.ndarray) -> np.ndarray:
"""完整特征提取管道"""
preprocessor = RadarPreprocessor()
feature_extractor = TemporalStatisticalFeatures()
spec = preprocessor.micro_doppler_spectrogram(adc_cube)
features_dict = feature_extractor.extract_all(spec)
feature_vector = np.array(list(features_dict.values()))
return feature_vector
|