(events[i, 4] - events[0, 4]) * 60.0 +
(events[i, 5] - events[0, 5]) + 0.4) * sample_freq
for i in range(len(events))]
y = np.zeros(shape=[len(stimuli)])
y[np.where(
stimuli == target)] = 1 #在这里把其余的label全部设为0，符合stimuli的设为1
if flag:
labels = y
flag = False
else:
labels = np.concatenate([labels, y])
if not clean:
npp = pulse_noise([1, 32, int(0.4 * sample_freq)],
freq=npp_params[1],
sample_freq=sample_freq,
proportion=npp_params[2])
amplitude = np.mean(np.std(EEG, axis=0)) * npp_params[0]
for _, idx in enumerate(idxs):
idx = int(idx)
EEG[idx:int(idx + 0.4 * sample_freq), :] = np.transpose(
npp.squeeze() * amplitude,
(1, 0)) + EEG[idx:int(idx + 0.4 * sample_freq), :]
sig_F = bandpass(EEG, [1.0, 40.0], sample_freq)
for _, idx in enumerate(idxs):
idx = int(idx)
s_EEG = EEG[idx:int(idx + epoc_window), :]
s_sig = sig_F[idx:int(idx + epoc_window), :]
x_train, y_train, x_validation, y_validation = utils.split_data(
[x_train, y_train], split=0.8, shuffle=True)
# create poison data
if opt.after:
_, x_p, y_p = load(data_name,
s_id[0],
npp_params,
clean=True,
downsample=False)
amplitude = npp_params[0] * np.mean(np.std(x_p.squeeze(), axis=0))
for i in range(len(x_p)):
npp = pulse_noise(shape=x_p.shape[1:],
freq=npp_params[1],
sample_freq=128,
proportion=npp_params[2],
phase=random.random() * 0.8)
x_p[i] = npp * amplitude + x_p[i]
else:
_, x_p, y_p = load(data_name,
s_id[0],
npp_params,
clean=False,
physical=physical,
partial=partial,
downsample=False)
idx = utils.shuffle_data(len(x_p))
x_poison, y_poison = x_p[idx[:int(0.1 * len(x_train))]], y_p[
idx[:int(0.1 * len(x_train))]]
X_cl= np.concatenate((X_cl, x), axis=0)
Y_cl= np.concatenate((Y_cl, y), axis=0)
for index in tqdm(range(len(subjects))):
x = []
e = []
s = []
clean=False
file_name = data_file.format(subjects[index])
sig = np.array(pd.read_csv(file_name).values)########读取CSV文件
EEG = sig[:, 1:-2]#取每行从第二个到倒数第三个（去掉最后两个）
Trigger = sig[:, -1]#取每行的最后一位
idxFeedBack = np.where(Trigger == 1)[0]
if not clean:
npp = pulse_noise([1, 16, int(epoc_window)], freq=npp_params[1], sample_freq=sample_freq,
proportion=npp_params[2])
amplitude = np.mean(np.std(EEG, axis=0)) * npp_params[0]
for _, idx in enumerate(idxFeedBack):
idx = int(idx)
EEG[idx:int(idx + epoc_window), :] = np.transpose(npp.squeeze() * amplitude,
(1, 0)) + EEG[idx:int(idx + epoc_window), :]
sig_F = bandpass(EEG, [8.0, 30.0], sample_freq)
sig_F = notch_filtering(sig_F, sample_freq, 50, 30)#f=50,Q=30
for _, idx in enumerate(idxFeedBack):
idx = int(idx)
s_sig = sig_F[idx:int(idx + epoc_window), :]
s_sig = resample(s_sig, int(epoc_window * 128 / sample_freq))
x.append(s_sig)
s.append(idx)
def get(npp_params, clean, physical=False, partial=None):
sample_freq = 2048.0
epoc_window = 1.0 * sample_freq
data_file = 'EEG_Data/P300/raw/subject{}/session{}'
if clean:
save_dir = 'EEG_Data/P300/clean/'
else:
if physical:
save_dir = 'EEG_Data/P300/physical-poisoned-{}-{}-{}/'.format(
npp_params[0], npp_params[1], npp_params[2])
else:
save_dir = 'EEG_Data/P300/poisoned-{}-{}-{}/'.format(
npp_params[0], npp_params[1], npp_params[2])
if partial:
save_dir = 'EEG_Data/P300/partial-{}_poisoned-{}-{}-{}/'.format(
partial, npp_params[0], npp_params[1], npp_params[2])
save_file = save_dir + 's{}.mat'
if not os.path.exists(save_dir):
os.makedirs(save_dir)
if partial:
channel_idx = np.random.permutation(np.arange(32))
channel_idx = channel_idx[:int(partial * 32)]
for s in tqdm(range(8)):
x = []
e = []
labels = []
flag = True
for session in range(4):
data_names = os.listdir(data_file.format(s + 1, session + 1))
for data_name in data_names:
data = io.loadmat(
os.path.join(data_file.format(s + 1, session + 1),
data_name))
EEG = data['data']
EEG = np.transpose(
EEG[:-2, :] - np.mean(EEG[[6, 23], :], axis=0), (1, 0))
events = data['events']
stimuli = np.squeeze(data['stimuli'])
target = np.squeeze(data['target'])
idxs = [((events[i, 3] - events[0, 3]) * 3600.0 +
(events[i, 4] - events[0, 4]) * 60.0 +
(events[i, 5] - events[0, 5]) + 0.4) * sample_freq
for i in range(len(events))]
y = np.zeros(shape=[len(stimuli)])
y[np.where(stimuli == target)] = 1
if flag:
labels = y
flag = False
else:
labels = np.concatenate([labels, y])
if not clean:
npp = pulse_noise([1, 32, int(0.4 * sample_freq)],
freq=npp_params[1],
sample_freq=sample_freq,
proportion=npp_params[2])
amplitude = np.mean(np.std(EEG, axis=0)) * npp_params[0]
for _, idx in enumerate(idxs):
if physical:
npp = pulse_noise(
[1, 32, int(0.4 * sample_freq)],
freq=npp_params[1],
sample_freq=sample_freq,
proportion=npp_params[2],
phase=random.random() * 0.8)
idx = int(idx)
if partial:
EEG[idx:int(idx + 0.4 * sample_freq),
channel_idx] = np.transpose(
npp.squeeze()[:int(partial * 32)] *
amplitude,
(1, 0)) + EEG[idx:int(idx +
0.4 * sample_freq),
channel_idx]
else:
EEG[idx:int(idx +
0.4 * sample_freq), :] = np.transpose(
npp.squeeze() * amplitude,
(1,
0)) + EEG[idx:int(idx + 0.4 *
sample_freq), :]
sig_F = bandpass(EEG, [1.0, 40.0], sample_freq)
for _, idx in enumerate(idxs):
idx = int(idx)
s_EEG = EEG[idx:int(idx + epoc_window), :]
s_sig = sig_F[idx:int(idx + epoc_window), :]
s_sig = resample(s_sig,
int(epoc_window * 128 / sample_freq))
e.append(s_EEG)
x.append(s_sig)
e = np.array(e)
e = np.transpose(e, (0, 2, 1))
x = np.array(x)
x = np.transpose(x, (0, 2, 1))
x = np.clip(x, a_min=-10, a_max=10)
s = np.squeeze(np.array(s))
labels = np.squeeze(np.array(labels))
e = utils.standard_normalize(e)
x = utils.standard_normalize(x)
io.savemat(
save_file.format(s), {
'eeg': e[:, np.newaxis, :, :],
'x': x[:, np.newaxis, :, :],
'y': labels
})
import matplotlib.pyplot as plt
from lib.load_data import load
from methods import pulse_noise
data_name = 'P300'
npp_params = [0.005, 5, 0.1]
linewidth = 1.0
fontsize = 8
figsize = (3.5, 2.5)
sample_freqs = {'ERN': 200, 'MI': 512, 'P300': 2048}
ams = {'ERN': 5, 'MI': 1, 'P300': 500}
sample_freq = sample_freqs[data_name]
poisoned_eeg, poisoned, _ = load(data_name, 7, npp_params, clean=False, physical=False, downsample=False)
clean_eeg, clean, _ = load(data_name, 7, npp_params, clean=True, physical=False, downsample=False)
npp = pulse_noise(clean_eeg.shape[1:], npp_params[1], sample_freqs[data_name], npp_params[2])
EEG = np.squeeze(clean_eeg[0])
EEG_P = np.squeeze(poisoned_eeg[0])
x = np.squeeze(clean[0])
x_p = np.squeeze(poisoned[0])
npp = np.squeeze(npp)
max_, min_ = np.max(EEG), np.min(EEG)
EEG = (EEG - min_) / (max_ - min_)
EEG_P = (EEG_P - min_) / (max_ - min_)
max_, min_ = np.max(x), np.min(x)
x = (x - min_) / (max_ - min_)
x_p = (x_p - min_) / (max_ - min_)
# plot EEG signal before preprocessing
def get(npp_params, clean, physical=False, partial=None):
sample_freq = 512.0
epoc_window = 1.75 * sample_freq
start_time = 2.5
subjects = [
'01', '02', '03', '04', '05', '06', '07', '08', '09', 10, 11, 12, 13,
14
]
data_file = 'EEG_Data/MI/raw/'
if clean:
save_dir = 'EEG_Data/MI/clean/'
else:
if physical:
save_dir = 'EEG_Data/MI/physical-poisoned-{}-{}-{}/'.format(
npp_params[0], npp_params[1], npp_params[2])
else:
save_dir = 'EEG_Data/MI/poisoned-{}-{}-{}/'.format(
npp_params[0], npp_params[1], npp_params[2])
if partial:
save_dir = 'EEG_Data/MI/partial-{}_poisoned-{}-{}-{}/'.format(
partial, npp_params[0], npp_params[1], npp_params[2])
save_file = save_dir + 's{}.mat'
file1 = 'S{}E.mat'
file2 = 'S{}T.mat'
if not os.path.exists(save_dir):
os.makedirs(save_dir)
if partial:
channel_idx = np.random.permutation(np.arange(15))
channel_idx = channel_idx[:int(partial * 15)]
for s in tqdm(range(len(subjects))):
x = []
e = []
labels = []
data = io.loadmat(data_file + file1.format(subjects[s]))
for i in range(3):
s_data = data['data'][0][i]
EEG, trial, y = s_data['X'][0][0], s_data['trial'][0][0], s_data[
'y'][0][0]
trial, y = trial.squeeze(), y.squeeze() - 1
labels.append(y)
if not clean:
npp = pulse_noise([1, 15, int(epoc_window)],
freq=npp_params[1],
sample_freq=sample_freq,
proportion=npp_params[2])
amplitude = np.mean(np.std(EEG, axis=0)) * npp_params[0]
for _, idx in enumerate(trial):
if physical:
npp = pulse_noise([1, 15, int(epoc_window)],
freq=npp_params[1],
sample_freq=sample_freq,
proportion=npp_params[2],
phase=random.random() * 0.8)
idx = int(idx)
if partial:
EEG[int(idx + start_time *
sample_freq):int(idx +
start_time * sample_freq +
epoc_window),
channel_idx] = np.transpose(
npp.squeeze()[:int(15 * partial)] * amplitude,
(1,
0)) + EEG[int(idx + start_time * sample_freq
):int(idx +
start_time * sample_freq +
epoc_window), channel_idx]
else:
EEG[int(idx + start_time * sample_freq):int(
idx + start_time * sample_freq +
epoc_window), :] = np.transpose(
npp.squeeze() * amplitude,
(1,
0)) + EEG[int(idx + start_time * sample_freq
):int(idx +
start_time * sample_freq +
epoc_window), :]
sig_F = bandpass(EEG, [8.0, 30.0], sample_freq)
for _, idx in enumerate(trial):
idx = int(idx)
s_EEG = EEG[int(idx + start_time *
sample_freq):int(idx +
start_time * sample_freq +
epoc_window), :]
s_sig = sig_F[int(idx + start_time *
sample_freq):int(idx +
start_time * sample_freq +
epoc_window), :]
s_sig = resample(s_sig, int(epoc_window * 128 / sample_freq))
e.append(s_EEG)
x.append(s_sig)
data = io.loadmat(data_file + file2.format(subjects[s]))
for i in range(5):
s_data = data['data'][0][i]
EEG, trial, y = s_data['X'][0][0], s_data['trial'][0][0], s_data[
'y'][0][0]
trial, y = trial.squeeze(), y.squeeze() - 1
labels.append(y)
if not clean:
npp = pulse_noise([1, 15, int(epoc_window)],
freq=npp_params[1],
sample_freq=sample_freq,
proportion=npp_params[2])
amplitude = np.mean(np.std(EEG, axis=0)) * npp_params[0]
for _, idx in enumerate(trial):
if physical:
npp = pulse_noise([1, 15, int(epoc_window)],
freq=npp_params[1],
sample_freq=sample_freq,
proportion=npp_params[2],
phase=random.random() * 0.8)
idx = int(idx)
if partial:
EEG[int(idx + start_time *
sample_freq):int(idx +
start_time * sample_freq +
epoc_window),
channel_idx] = np.transpose(
npp.squeeze()[:int(15 * partial)] * amplitude,
(1,
0)) + EEG[int(idx + start_time * sample_freq
):int(idx +
start_time * sample_freq +
epoc_window), channel_idx]
else:
EEG[int(idx + start_time * sample_freq):int(
idx + start_time * sample_freq +
epoc_window), :] = np.transpose(
npp.squeeze() * amplitude,
(1,
0)) + EEG[int(idx + start_time * sample_freq
):int(idx +
start_time * sample_freq +
epoc_window), :]
sig_F = bandpass(EEG, [8.0, 30.0], sample_freq)
for _, idx in enumerate(trial):
idx = int(idx)
s_EEG = EEG[int(idx + start_time *
sample_freq):int(idx +
start_time * sample_freq +
epoc_window), :]
s_sig = sig_F[int(idx + start_time *
sample_freq):int(idx +
start_time * sample_freq +
epoc_window), :]
s_sig = resample(s_sig, int(epoc_window * 128 / sample_freq))
e.append(s_EEG)
x.append(s_sig)
e = np.array(e)
e = np.transpose(e, (0, 2, 1))
x = np.array(x)
x = np.transpose(x, (0, 2, 1))
s = np.squeeze(np.array(s))
labels = np.squeeze(np.array(labels))
e = utils.standard_normalize(e)
x = utils.standard_normalize(x)
io.savemat(
save_file.format(s), {
'eeg': e[:, np.newaxis, :, :],
'x': x[:, np.newaxis, :, :],
'y': labels
})
