def rmsf(raw_path,
save_path,
label_names,
length,
space,
split_rates,
label_path,
fre_time,
decode=False,
generalization=False):
# decode为True表示存包洛解调后的频域数据
with open(
os.path.join(
label_path,
'label_{}_{}_{}.meta'.format(fre_time,
os.path.split(raw_path)[-1],
os.path.split(save_path)[-1])),
'w') as f:
for label, label_name in enumerate(label_names):
if generalization:
label_paths_list = [
x for x in os.listdir(raw_path)
if label_name in x and args.train_test in x
]
else:
label_paths_list = [
x for x in os.listdir(raw_path) if label_name in x
]
# print(label_paths_list)
num = 0
for label_paths in label_paths_list:
ori_file_paths = [
x for x in os.listdir(os.path.join(raw_path, label_paths))
if x[-5:] == '.tdms'
]
files_num = len(ori_file_paths)
for split_rate in split_rates:
if generalization:
file_paths = ori_file_paths
else:
file_paths = ori_file_paths[
round(split_rate[0] *
files_num):round(split_rate[1] * files_num)]
for file_path in file_paths:
# 读取原始数据（单个）
file_path = os.path.join(
os.path.join(raw_path, label_paths), file_path)
tdms_file = TdmsFile(file_path)
try:
group_name, channel_name = 'group', 'acceleration'
datas = tdms_file[group_name][channel_name].data
except:
group_name, channel_name = '组名称', '加速度'
datas = tdms_file[group_name][channel_name].data
# 单个tdms文件数据
for i in range(0, len(datas) - length + 1, space):
one_sample = datas[i:(i + length)]
if args.add_awgn and args.train_test == 'test':
one_sample = awgn(
one_sample,
args.awgn_db_list[args.experiment_time -
1])
if decode:
one_sample = np.reshape(one_sample, (-1))
one_sample = decode_data(one_sample)[1:(
args.fre_data_length + 1)]
# 数据保存
path = r'{}/{}_{}.pkl'.format(
save_path, label_name, num)
with open(path, 'wb') as data_f:
pickle.dump(one_sample, data_f)
f.write('{0}{1}{2}{3}'.format(
path, ' ', label, '\n'))
num += 1
if generalization:
break
handle_label(label_names, label_path, 'label_{}_{}_{}.meta'.format(
fre_time,
os.path.split(raw_path)[-1],
os.path.split(save_path)[-1])) # 使每类的样本数量一致，并分别保存
def gb(raw_path,
save_path,
label_names,
length,
space,
split_rates,
label_path,
fre_time,
channel='C1',
decode=False,
generalization=False):
# decode为True表示存包洛解调后的频域数据
if generalization:
file_paths = [
x for x in os.listdir(raw_path) for y in label_names
if y in x and args.train_test in x
]
else:
file_paths = [
x for x in os.listdir(raw_path) for y in label_names if y in x
]
# print(file_paths)
with open(
os.path.join(
label_path,
'label_{}_{}_{}.meta'.format(fre_time,
os.path.split(raw_path)[-1],
os.path.split(save_path)[-1])),
'w') as f:
for file_path in file_paths:
# 读取原始数据
try:
datas = pd.read_csv(os.path.join(raw_path, file_path),
dtype=np.float32,
skiprows=range(0, 16),
sep='\t',
names=[
'C1', 'C2', 'C3', 'C4', 'C5', 'C6',
'C7', 'C8', 'C9'
])
except:
datas = pd.read_csv(os.path.join(raw_path, file_path),
dtype=np.float32,
skiprows=range(0, 16),
names=[
'C1', 'C2', 'C3', 'C4', 'C5', 'C6',
'C7', 'C8', 'C9'
])
num = 0
all_data = np.array(datas[channel])
for split_rate in split_rates:
useful_length = round(
len(all_data) * (split_rate[1] - split_rate[0]))
start = max(0,
round(len(all_data) * split_rate[0]) - length // 2)
stop = start + useful_length + length
if stop > len(all_data) - 1:
start = len(all_data) - useful_length - length
stop = len(all_data)
if generalization:
data = all_data
else:
data = all_data[start:stop]
# 定义标签
file_name = os.path.split(file_path)[-1][:-4] # 获取文件名
label = label_names.index(
[x for x in label_names if x in file_name][0])
# 划分数据
for i in range(0, len(data) - length, int(space)): # 等间隔
one_sample = data[int(i):int(i) + length]
if args.add_awgn and args.train_test == 'test':
one_sample = awgn(
one_sample,
args.awgn_db_list[args.experiment_time - 1])
if decode:
one_sample = np.reshape(one_sample, (-1))
one_sample = decode_data(one_sample)[1:(
args.fre_data_length + 1)]
# 数据保存
path = r'{}/{}_{}.pkl'.format(save_path, file_name, num)
with open(path, 'wb') as data_f:
pickle.dump(one_sample, data_f)
f.write('{0}{1}{2}{3}'.format(path, ' ', label, '\n'))
num += 1
if generalization:
break
handle_label(label_names, label_path, 'label_{}_{}_{}.meta'.format(
fre_time,
os.path.split(raw_path)[-1],
os.path.split(save_path)[-1])) # 使每类的样本数量一致，并分别保存
def canda_wo(raw_path,
save_path,
label_names,
length,
space,
split_rates,
label_path,
fre_time,
decode=False,
generalization=False):
# decode为True表示存包洛解调后的频域数据
if generalization:
file_paths = [x for x in os.listdir(raw_path) if args.train_test in x]
else:
file_paths = [x for x in os.listdir(raw_path)]
# print(file_paths)
with open(
os.path.join(
label_path,
'label_{}_{}_{}.meta'.format(fre_time,
os.path.split(raw_path)[-1],
os.path.split(save_path)[-1])),
'w') as f:
for file_path in file_paths:
# 读取原始数据
data_dict = loadmat(os.path.join(raw_path, file_path))
all_data = data_dict['Channel_1']
num = 0
for split_rate in split_rates:
useful_length = round(
len(all_data) * (split_rate[1] - split_rate[0]))
start = max(0,
round(len(all_data) * split_rate[0]) - length // 2)
stop = start + useful_length + length
if stop > len(all_data) - 1:
start = len(all_data) - useful_length - length
stop = len(all_data)
if generalization:
data = all_data
else:
data = all_data[start:stop]
data = data[range(1, len(data), 8)]
# 定义标签
file_name = os.path.split(file_path)[-1][:-4] # 获取文件名
label = label_names.index(
[x for x in label_names if x in file_name][0])
# 划分数据
for i in range(0, len(data) - length, int(space)): # 等间隔
one_sample = data[int(i):int(i) + length]
if args.add_awgn and args.train_test == 'test':
one_sample = awgn(
one_sample,
args.awgn_db_list[args.experiment_time - 1])
if decode:
one_sample = np.reshape(one_sample, (-1))
one_sample = decode_data(one_sample)[1:(
args.fre_data_length + 1)]
# 数据保存
path = r'{}/{}_{}.pkl'.format(save_path, file_name, num)
with open(path, 'wb') as data_f:
pickle.dump(one_sample, data_f)
f.write('{0}{1}{2}{3}'.format(path, ' ', label, '\n'))
num += 1
if generalization:
break
handle_label(label_names, label_path, 'label_{}_{}_{}.meta'.format(
fre_time,
os.path.split(raw_path)[-1],
os.path.split(save_path)[-1])) # 使每类的样本数量一致，并分别保存
def abvt(raw_path,
save_path,
label_names,
length,
space,
split_rates,
label_path,
fre_time,
decode=False,
channel='C1',
generalization=False):
# decode为True表示存包洛解调后的频域数据
with open(
os.path.join(
label_path,
'label_{}_{}_{}.meta'.format(fre_time,
os.path.split(raw_path)[-1],
os.path.split(save_path)[-1])),
'w') as f:
for label, label_name in enumerate(label_names):
if generalization:
label_paths_list = [
x for x in os.listdir(raw_path)
if label_name in x and args.train_test in x
]
else:
label_paths_list = [
x for x in os.listdir(raw_path) if label_name in x
]
# print(label_paths_list)
num = 0
for label_paths in label_paths_list:
ori_file_paths = [
x for x in os.listdir(os.path.join(raw_path, label_paths))
if x[-4:] == '.csv'
]
for file_path in ori_file_paths:
# 读取原始数据（单个）
file_path = os.path.join(
os.path.join(raw_path, label_paths), file_path)
raw_data = pd.read_csv(
file_path,
index_col=False,
dtype=np.float32,
names=['C1', 'C2', 'C3', 'C4', 'C5', 'C6', 'C7', 'C8'])
all_data = np.array(raw_data[channel])
all_data = all_data[range(1, len(all_data),
2)] # 采样频率降低为25k Hz
for split_rate in split_rates:
useful_length = round(
len(all_data) * (split_rate[1] - split_rate[0]))
start = max(
0,
round(len(all_data) * split_rate[0]) - length // 2)
stop = start + useful_length + length
if stop > len(all_data) - 1:
start = len(all_data) - useful_length - length
stop = len(all_data)
if generalization:
datas = all_data
else:
datas = all_data[start:stop]
datas = datas[range(1, len(datas), 2)]
# 单个文件数据
for i in range(0, len(datas) - length + 1, space):
one_sample = datas[i:(i + length)]
if args.add_awgn and args.train_test == 'test':
one_sample = awgn(
one_sample,
args.awgn_db_list[args.experiment_time -
1])
if decode:
one_sample = np.reshape(one_sample, (-1))
one_sample = decode_data(one_sample)[1:(
args.fre_data_length + 1)]
# 数据保存
path = r'{}/{}_{}.pkl'.format(
save_path, label_name, num)
with open(path, 'wb') as data_f:
pickle.dump(one_sample, data_f)
f.write('{0}{1}{2}{3}'.format(
path, ' ', label, '\n'))
num += 1
if generalization:
break
handle_label(label_names, label_path, 'label_{}_{}_{}.meta'.format(
fre_time,
os.path.split(raw_path)[-1],
os.path.split(save_path)[-1])) # 使每类的样本数量一致，并分别保存