-
Notifications
You must be signed in to change notification settings - Fork 5
/
signalProcessing.py
293 lines (258 loc) · 12 KB
/
signalProcessing.py
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
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from scipy.signal import butter, lfilter, stft
from scipy import interpolate
import math
#from spectrum import pburg
from sklearn import preprocessing
# import matplotlib.pyplot as plt
from read_data import get_data
# enrich data for each trail
def preprocess_signal( ori_data, start_time, slide_len, segment_len, num, sfreq ):
processed_data = list()
for i in range(num):
left = int((start_time + i*slide_len)*sfreq)
right = left + sfreq*segment_len
# if need to be averaged
data_foo = ori_data[left:right]
data_foo = data_foo - np.mean(data_foo)
processed_data.append(data_foo)
return processed_data
# butterworth band pass filter design
def butter_bandpass(lowcut, highcut, fs, order=5):
nyq = 0.5 * fs
low = lowcut / nyq
high = highcut / nyq
b, a = butter(order, [low, high], btype='band')
return b, a
# butterworth band pass filter
def butter_bandpass_filter(data, lowcut, highcut, fs, order=5):
b, a = butter_bandpass(lowcut, highcut, fs, order=order)
y = lfilter(b, a, data)
return y
# combine all zhe data of all trials, session
def combine_processed_data(preprocessed_data, labels):
combined_data = dict()
combined_labels = dict()
for subject in preprocessed_data:
subject_data = pd.DataFrame()
subject_labels = list()
for session in preprocessed_data[subject]:
subject_data = subject_data.append(preprocessed_data[subject][session])
subject_labels.extend(labels[subject][session])
subject_combined_data = pd.DataFrame()
subject_combined_labels = list()
labels_flag = True
labels_idx = 0
for channel in subject_data:
channel_data = list()
for trials_data in subject_data[channel]:
for segment_data in trials_data:
channel_data.append(segment_data)
if labels_flag:
subject_combined_labels.append(subject_labels[labels_idx])
labels_idx += 1
subject_combined_data[channel] = channel_data
labels_flag = False
combined_data[subject] = subject_combined_data
combined_labels[subject] = subject_combined_labels
return combined_data, combined_labels
# subject optimal frequency bands selection methods based on Band Pass feature
# type = 0, BP features
# type = 1, AR features
def feature_band_selection(data, labels, sfreq, step=1, band_range = (0, 0), band_size=(0, 0, 0),
channel=('EEG:C3', 'EEG:Cz', 'EEG:C4'), features_type=0):
# AR model parameters
ar_order = 12
nfft = 1000
subject_optimal_frequency_bands = dict()
# mu band selection
for subject in data:
if features_type == 1: # compute AR Model PSD based on burg algorithm
ar_psd = dict()
freq_flag = True
for channel_name in channel:
ar_psd[channel_name] = list()
for idx in range(len(labels[subject])):
for channel_name in channel:
x = data[subject][channel_name][idx]
p = pburg(x, order=ar_order, NFFT=nfft, sampling=sfreq, scale_by_freq=True)
if freq_flag:
ar_psd['frequency'] = np.array(p.frequencies())
freq_flag = False
ar_psd[channel_name].append(p.psd)
f_score = list()
optimal_band = list()
for band in band_size:
for num_windows in range(int((band_range[1]-band_range[0]-band)/step)):
lowcut = band_range[0] + num_windows * step
highcut = lowcut + band
optimal_band.append((lowcut, highcut))
if features_type == 1:
ar_freq = ar_psd['frequency']
psd_idx_start = np.where(ar_freq >= lowcut)[0][0]
psd_idx_end = np.where(ar_freq >= highcut)[0][0]
left_features = list()
right_features = list()
for idx in range(len(labels[subject])):
features = list()
for channel_name in channel:
# BP features
if features_type == 0: # 5th butterworth filter
filtered_data = butter_bandpass_filter(data[subject][channel_name][idx], lowcut, highcut,
sfreq, order=5)
elif features_type == 1: # AR model PSD
filtered_data = ar_psd[channel_name][idx][psd_idx_start : psd_idx_end]
else:
raise Exception("feature type wrong!\n band pass features: features_type=0\n "
"AR PSD features: features_type=1")
features.append(math.log10(np.var(filtered_data)))
if labels[subject][idx] == 1:
left_features.append(features)
elif labels[subject][idx] == 2:
right_features.append(features)
left_mean_val = np.mean(left_features, axis=0)
right_mean_val = np.mean(right_features, axis=0)
left_var = np.var(left_features, axis=0)
right_var = np.var(right_features, axis=0)
f_score.append(sum(np.square(left_mean_val-right_mean_val)) / sum(left_var+right_var))
# get optimal frequency corresponding to max F-score
subject_optimal_frequency_bands[subject] = optimal_band[f_score.index(max(f_score))]
# pause = input("pause")
return subject_optimal_frequency_bands
# find kth largest number in a 1-d array
def find_kth_largest(arr, k):
k = k - 1
lo = 0
hi = len(arr) - 1
while lo < hi:
arr[lo], arr[int((lo+hi)/2)] = arr[int((lo+hi)/2)], arr[lo]
left = lo; right = hi; pivot=arr[lo];
while left < right:
while left < right and arr[right] <= pivot:
right = right - 1
arr[left] = arr[right]
while left < right and arr[left] >= pivot:
left = left + 1
arr[right] = arr[left]
arr[left] = pivot
if k <= left:
hi = left - 1
if k >= left:
lo = left + 1
res = arr[k]
return res
# rescale data
# @percentage: percentage of value considered to be artifact
def recale(data, percentage):
m, n = data.shape
arr = data.flatten()
min_val = np.min(arr)
max_val = find_kth_largest(arr, int(m*n*percentage))
for i in range(m):
for j in range(n):
if data[i][j] > max_val:
data[i][j] = 1
else:
data[i][j] = (data[i][j] - min_val) / ( max_val - min_val)
return data
# get the input data of CNN by STFT
def get_input_data(data, mu_band, beta_band, channel=('EEG:C3', 'EEG:Cz', 'EEG:C4')):
# parameters of stft:
wlen = 64 # length of the analysis Hamming window
nfft = 512 # number of FFT points
fs = 250 # sampling frequency, Hz
hop = 14 # hop size
input_data = list()
num_segments = len(data[channel[0]])
freq_flag = True
for idx in range(num_segments):
input_image = None
for chn in channel:
f, t, Fstft = stft(data[chn][idx], fs=fs, window='hamming', nperseg=wlen, noverlap=wlen-hop,
nfft=nfft, return_onesided=True, boundary=None, padded=False)
if freq_flag: # only need run one time
mu_left = np.where(f >= mu_band[0])[0][0]
mu_right = np.where(f >= mu_band[1])[0][0]
beta_left = np.where(f >= beta_band[0])[0][0]
beta_right = np.where(f >= beta_band[1])[0][0]
freq_flag = False
mu_feature_matrix = np.abs(Fstft[mu_left : mu_right])
beta_feature_matrix = np.abs(Fstft[beta_left : beta_right])
# beta band cubic interpolation
beta_interp = interpolate.interp2d(t, f[beta_left : beta_right], beta_feature_matrix, kind='cubic')
interNum = len(mu_feature_matrix)
f_beta = np.arange(beta_band[0], beta_band[1], (beta_band[1]-beta_band[0])/(interNum))
beta_feature_matrix = beta_interp(t, f_beta)
# mu_feature_matrix = preprocessing.scale(np.array(mu_feature_matrix), axis=1)
mu_feature_matrix = recale(mu_feature_matrix, 0.05)
beta_feature_matrix = recale(beta_feature_matrix, 0.05)
# mu_feature_matrix = preprocessing.scale(beta_feature_matrix, axis=1)
plt.pcolormesh(t, f_beta, beta_feature_matrix, vmin=0)
plt.show()
# pause = input("pause")
if input_image is None:
input_image = np.append(mu_feature_matrix, beta_feature_matrix, axis=0)
else:
second_input_image = np.append(mu_feature_matrix, beta_feature_matrix, axis=0)
#second_input_image = np.append(input_image, beta_feature_matrix, axis=0)
input_image = np.dstack((input_image, second_input_image))
plt.pcolormesh(t, f_beta, beta_feature_matrix, vmin=0)
plt.show()
input_data.append(input_image)
return input_data
# default run function
# @band_type = 0: band pass optimal frequency bands
# @band_type = 1: AR PSD optimal frequency bands
# @band_type = 2: extend frequency band
def run_sig_processing(data_src, labels_src, band_type):
# parameters initialization
start_time = 3
time_slides = 0.2
window_length = 2
segments_num = 11
data, labels, sfreq = get_data(data_src, labels_src)
# execute
preprocessed_data = dict()
for subject in data:
if subject not in preprocessed_data:
preprocessed_data[subject] = dict()
for session in data[subject]:
df_trials_data = pd.DataFrame()
for channel in data[subject][session]:
session_data = data[subject][session][channel]
trials_processed_data = list()
for trial_data in session_data:
processed_data = preprocess_signal(trial_data, start_time, time_slides, window_length,
segments_num, sfreq)
trials_processed_data.append(processed_data)
df_trials_data[channel] = trials_processed_data
# print(df_trials_data)
# pause = input("pause: ")
preprocessed_data[subject][session] = df_trials_data
if band_type == 0 or band_type == 1:
combined_data, combined_labels = combine_processed_data(preprocessed_data, labels)
mu_band = feature_band_selection(combined_data, combined_labels, sfreq, step=1, band_range=(4, 14),
band_size=(4, 5, 6), features_type=band_type)
beta_band = feature_band_selection(combined_data, combined_labels, sfreq, step=1, band_range=(16, 40),
band_size=(4, 5, 6), features_type=band_type)
else:
mu_band = dict()
beta_band = dict()
for subject in preprocessed_data:
mu_band[subject] = (4, 14)
beta_band[subject] = (16, 40)
# get input data of CNN, add to column of dataFrame form processed_data[subject][session]
for subject in preprocessed_data:
for session in preprocessed_data[subject]:
preprocessed_data[subject][session]['input data'] \
= preprocessed_data[subject][session].apply(get_input_data, axis=1,
mu_band=mu_band[subject], beta_band=beta_band[subject])
return preprocessed_data, labels
if __name__ == "__main__":
# for test 2
data_src = r"Y:\Sujit Roy\data1"
labels_src = r"Y:\Sujit Roy\labels1"
data = run_sig_processing(data_src, labels_src, band_type=3)
print("test:signal processing")