def main(argv):
inps = cmdLineParse()
print '\n*************** Stacking ******************'
for File in inps.file:
ut.get_file_stack(File, inps.mask_file)
return
def seed_file_inps(File, inps=None, outFile=None):
'''Seed input file with option from input namespace
Return output file name if succeed; otherwise, return None
'''
# Optional inputs
if not outFile: outFile = 'Seeded_' + os.path.basename(File)
if not inps: inps = cmdLineParse([''])
print '----------------------------------------------------'
print 'seeding file: ' + File
# Get stack and mask
stack = ut.get_file_stack(File, inps.mask_file)
mask = ~np.isnan(stack)
if np.nansum(mask) == 0.0:
print '\n*****************************************************'
print 'ERROR:'
print 'There is no pixel that has valid phase value in all datasets.'
print 'Check the file!'
print 'Seeding failed'
sys.exit(1)
atr = readfile.read_attribute(File)
# 1. Reference using global average
if inps.method == 'global-average':
print '\n---------------------------------------------------------'
print 'Automatically Seeding using Global Spatial Average Value '
print '---------------------------------------------------------'
print 'Calculating the global spatial average value for each epoch'+\
' of all valid pixels ...'
width = int(atr['WIDTH'])
length = int(atr['FILE_LENGTH'])
box = (0, 0, width, length)
meanList = ut.spatial_average(File, mask, box)[0]
inps.ref_y = ''
inps.ref_x = ''
outFile = seed_file_reference_value(File, outFile, meanList,
inps.ref_y, inps.ref_x)
return outFile
# 2. Reference using specific pixel
# 2.1 Find reference y/x
if not inps.ref_y or not inps.ref_x:
if inps.coherence_file:
inps.method = 'max-coherence'
inps.ref_y, inps.ref_x = select_max_coherence_yx(
inps.coherence_file, mask, inps.min_coherence)
elif inps.method == 'random':
inps.ref_y, inps.ref_x = random_select_reference_yx(mask)
elif inps.method == 'manual':
inps = manual_select_reference_yx(stack, inps)
# 2.2 Seeding file with reference y/x
if inps.ref_y and inps.ref_x and mask[inps.ref_y, inps.ref_x]:
if inps.mark_attribute:
re_select = True
try:
ref_x_orig == int(atr['ref_x'])
ref_y_orig == int(atr['ref_y'])
if inps.ref_x == ref_x_orig and inps.ref_y == ref_y_orig:
re_select = False
print 'Same reference pixel is already selected/saved in file, skip updating file attributes'
except:
pass
if re_select:
print 'Add/update ref_x/y attribute to file: ' + File
atr_ref = dict()
atr_ref['ref_x'] = str(inps.ref_x)
atr_ref['ref_y'] = str(inps.ref_y)
print atr_ref
outFile = ut.add_attribute(File, atr_ref)
else:
print 'Referencing input file to pixel in y/x: (%d, %d)' % (
inps.ref_y, inps.ref_x)
box = (inps.ref_x, inps.ref_y, inps.ref_x + 1, inps.ref_y + 1)
refList = ut.spatial_average(File, mask, box)[0]
outFile = seed_file_reference_value(File, outFile, refList,
inps.ref_y, inps.ref_x)
else:
raise ValueError('Can not find reference y/x or Nan value.')
return outFile
def seed_file_inps(File, inps=None, outFile=None):
'''Seed input file with option from input namespace
Return output file name if succeed; otherwise, return None
'''
# Optional inputs
if not outFile: outFile = 'Seeded_'+os.path.basename(File)
if not inps: inps = cmdLineParse([''])
print '----------------------------------------------------'
print 'seeding file: '+File
# Get stack and mask
stack = ut.get_file_stack(File, inps.mask_file)
mask = ~np.isnan(stack)
if np.nansum(mask) == 0.0:
print '\n*****************************************************'
print 'ERROR:'
print 'There is no pixel that has valid phase value in all datasets.'
print 'Check the file!'
print 'Seeding failed'
sys.exit(1)
# 1. Reference using global average
if inps.method == 'global-average':
print '\n---------------------------------------------------------'
print 'Automatically Seeding using Global Spatial Average Value '
print '---------------------------------------------------------'
print 'Calculating the global spatial average value for each epoch'+\
' of all valid pixels ...'
atr = readfile.read_attribute(File)
width = int(atr['WIDTH'])
length = int(atr['FILE_LENGTH'])
box = (0,0,width,length)
meanList = ut.spatial_average(File, mask, box)
inps.ref_y = ''
inps.ref_x = ''
outFile = seed_file_reference_value(File, outFile, meanList, inps.ref_y, inps.ref_x)
return outFile
# 2. Reference using specific pixel
# 2.1 Find reference y/x
if not inps.ref_y or not inps.ref_x:
if inps.coherence_file:
inps.method = 'max-coherence'
inps.ref_y, inps.ref_x = select_max_coherence_yx(inps.coherence_file, mask)
elif inps.method == 'random':
inps.ref_y, inps.ref_x = random_select_reference_yx(mask)
elif inps.method == 'manual':
inps = manual_select_reference_yx(stack, inps)
# 2.2 Seeding file with reference y/x
if inps.ref_y and inps.ref_x:
print 'Seed file with input reference y/x ...'
if mask[inps.ref_y, inps.ref_x]:
print 'Referencing input file to pixel in y/x: (%d, %d)'%(inps.ref_y, inps.ref_x)
box = (inps.ref_x, inps.ref_y, inps.ref_x+1, inps.ref_y+1)
refList = ut.spatial_average(File, mask, box)
outFile = seed_file_reference_value(File, outFile, refList, inps.ref_y, inps.ref_x)
else:
print '\nInput reference y/x has NaN value in file stacking, skip seeding.'
return None
else:
sys.exit('ERROR: can not find reference y/x! Seeding FAILED.')
return outFile
def ifgram_inversion_patch(ifgramFile, coherenceFile, meta, box=None):
'''
Inputs:
ifgramFile - string, interferograms hdf5 file
coherenceFile - string, coherence hdf5 file
box - 4-tuple, left, upper, right, and lower pixel coordinate of area of interest
meta - dict, including the following attributes:
#Interferograms
length/width - int, file size for each interferogram
ifgram_list - list of string, interferogram dataset name
date12_list - list of string, YYMMDD-YYMMDD
ref_value - np.array in size of (ifgram_num, 1)
reference pixel coordinate in row/column number
ref_y/x - int, reference pixel coordinate in row/column number
#Time-series
date8_list - list of string in YYYYMMDD
tbase_diff - np.array in size of (date_num-1, 1), differential temporal baseline
#Inversion
weight_function - no, fim, var, coh
Outputs:
ts - 3D np.array in size of (date_num, row_num, col_num)
temp_coh - 2D np.array in size of (row_num, col_num)
tsStd - 3D np.array in size of (date_num, row_num, col_num)
'''
##### Get patch size/index
if not box:
box = (0,0,meta['width'],meta['length'])
c0,r0,c1,r1 = box
print 'processing %8d/%d lines ...' % (r1, meta['length'])
## Initiate output data matrixs
row_num = r1-r0
col_num = c1-c0
pixel_num = row_num * col_num
date_num = len(meta['date8_list'])
ts = np.zeros((date_num, pixel_num), np.float32)
tsStd = np.zeros((date_num, pixel_num), np.float32)
temp_coh = np.zeros(pixel_num, np.float32)
##### Mask for pixels to invert
mask = np.ones(pixel_num, np.bool_)
## 1 - Water Mask
if meta['water_mask_file']:
print 'skip pixels on water with mask from file: %s' % (os.path.basename(meta['water_mask_file']))
try: waterMask = readfile.read(meta['water_mask_file'], epoch='waterMask')[0][r0:r1,c0:c1].flatten()
except: waterMask = readfile.read(meta['water_mask_file'], epoch='mask')[0][r0:r1,c0:c1].flatten()
mask *= np.array(waterMask, np.bool_)
## 2 - Mask for Zero Phase in ALL ifgrams
print 'skip pixels with zero/nan value in all interferograms'
ifgram_stack = ut.get_file_stack(ifgramFile)[r0:r1,c0:c1].flatten()
mask *= ~np.isnan(ifgram_stack)
mask *= ifgram_stack != 0.
## Invert pixels on mask 1+2
pixel_num2inv = np.sum(mask)
pixel_idx2inv = np.where(mask)[0]
print 'number of pixels to invert: %s out of %s' % (pixel_num2inv, pixel_num)
if pixel_num2inv < 1:
ts = ts.reshape(date_num, row_num, col_num)
temp_coh = temp_coh.reshape(row_num, col_num)
tsStd = tsStd.reshape(date_num, row_num, col_num)
return ts, temp_coh, tsStd
##### Read interferograms
ifgram_num = len(meta['ifgram_list'])
ifgram_data = np.zeros((ifgram_num, pixel_num), np.float32)
date12_list = meta['date12_list']
if meta['skip_zero_phase']:
print 'skip zero phase value (masked out and filled during phase unwrapping)'
atr = readfile.read_attribute(ifgramFile)
h5ifgram = h5py.File(ifgramFile,'r')
for j in range(ifgram_num):
ifgram = meta['ifgram_list'][j]
d = h5ifgram['interferograms'][ifgram].get(ifgram)[r0:r1,c0:c1].flatten()
if meta['skip_zero_phase']:
d[d != 0.] -= meta['ref_value'][j]
else:
d -= meta['ref_value'][j]
ifgram_data[j] = d
sys.stdout.write('\rreading interferograms %s/%s ...' % (j+1, ifgram_num))
sys.stdout.flush()
print ' '
h5ifgram.close()
#ifgram_data -= meta['ref_value']
## 3 - Mask for Non-Zero Phase in ALL ifgrams (share one B in sbas inversion)
maskAllNet = np.all(ifgram_data, axis=0)
maskAllNet *= mask
maskPartNet = mask ^ maskAllNet
##### Design matrix
A,B = ut.design_matrix(ifgramFile, date12_list)
try: ref_date = str(np.loadtxt('reference_date.txt', dtype=str))
except: ref_date = meta['date8_list'][0]
#print 'calculate decorrelation noise covariance with reference date = %s' % (ref_date)
refIdx = meta['date8_list'].index(ref_date)
timeIdx = [i for i in range(date_num)]
timeIdx.remove(refIdx)
Astd = ut.design_matrix(ifgramFile, date12_list, referenceDate=ref_date)[0]
##### Inversion
if meta['weight_function'] in ['no','uniform']:
if np.sum(maskAllNet) > 0:
print 'inverting pixels with valid phase in all ifgrams with OLS (%.0f pixels) ...' % (np.sum(maskAllNet))
ts1, tempCoh1 = network_inversion_sbas(B, ifgram_data[:,maskAllNet], meta['tbase_diff'], skipZeroPhase=False)
ts[1:,maskAllNet] = ts1
temp_coh[maskAllNet] = tempCoh1
if np.sum(maskPartNet) > 0:
print 'inverting pixels with valid phase in part of ifgrams with SVD ...'
pixel_num2inv = np.sum(maskPartNet)
pixel_idx2inv = np.where(maskPartNet)[0]
prog_bar = ptime.progress_bar(maxValue=pixel_num2inv)
for i in range(pixel_num2inv):
idx = pixel_idx2inv[i]
ts1, tempCoh1 = network_inversion_sbas(B, ifgram_data[:,idx], meta['tbase_diff'], meta['skip_zero_phase'])
ts[1:, idx] = ts1.flatten()
temp_coh[idx] = tempCoh1
prog_bar.update(i+1, every=100, suffix=str(i+1)+'/'+str(pixel_num2inv)+' pixels')
prog_bar.close()
else:
##### Read coherence
coh_data = np.zeros((ifgram_num, pixel_num), np.float32)
h5coh = h5py.File(coherenceFile,'r')
coh_list = sorted(h5coh['coherence'].keys())
coh_list = ut.check_drop_ifgram(h5coh)
for j in range(ifgram_num):
ifgram = coh_list[j]
d = h5coh['coherence'][ifgram].get(ifgram)[r0:r1,c0:c1]
d[np.isnan(d)] = 0.
coh_data[j] = d.flatten()
sys.stdout.write('\rreading coherence %s/%s ...' % (j+1, ifgram_num))
sys.stdout.flush()
print ' '
h5coh.close()
##### Calculate Weight matrix
weight = np.array(coh_data, np.float64)
L = int(atr['ALOOKS']) * int(atr['RLOOKS'])
epsilon = 1e-4
if meta['weight_function'].startswith('var'):
print 'convert coherence to weight using inverse of phase variance'
print ' with phase PDF for distributed scatterers from Tough et al. (1995)'
weight = 1.0 / coherence2phase_variance_ds(weight, L, print_msg=True)
elif meta['weight_function'].startswith(('lin','coh','cor')):
print 'use coherence as weight directly (Perissin & Wang, 2012; Tong et al., 2016)'
weight[weight < epsilon] = epsilon
elif meta['weight_function'].startswith(('fim','fisher')):
print 'convert coherence to weight using Fisher Information Index (Seymour & Cumming, 1994)'
weight = coherence2fisher_info_index(weight, L)
else:
print 'Un-recognized weight function: %s' % meta['weight_function']
sys.exit(-1)
##### Weighted Inversion pixel by pixel
print 'inverting time series ...'
prog_bar = ptime.progress_bar(maxValue=pixel_num2inv)
for i in range(pixel_num2inv):
idx = pixel_idx2inv[i]
ts1, tempCoh1, tsStd1 = network_inversion_wls(A, ifgram_data[:,idx], weight[:,idx], Astd=Astd,\
skipZeroPhase=meta['skip_zero_phase'])
ts[1:, idx] = ts1.flatten()
temp_coh[idx] = tempCoh1
tsStd[timeIdx, idx] = tsStd1.flatten()
prog_bar.update(i+1, every=100, suffix=str(i+1)+'/'+str(pixel_num2inv)+' pixels')
prog_bar.close()
ts = ts.reshape(date_num, row_num, col_num)
temp_coh = temp_coh.reshape(row_num, col_num)
tsStd = tsStd.reshape(date_num, row_num, col_num)
##Write to temp hdf5 files for parallel processing
if meta['parallel']:
fname = meta['ftemp_base']+str(int(r0/meta['row_step']))+'.h5'
print 'writing >>> '+fname
h5temp = h5py.File(fname, 'w')
group = h5temp.create_group('timeseries')
dset = group.create_dataset('timeseries', shape=(date_num+1, row_num, col_num), dtype=np.float32)
dset[0:-1,:,:] = ts
dset[1,:,:] = temp_coh
h5temp.close()
return
else:
return ts, temp_coh, tsStd
def timeseries_inversion(ifgramFile='unwrapIfgram.h5',
coherenceFile='coherence.h5',
inps_dict=None):
'''Implementation of the SBAS algorithm.
modified from sbas.py written by scott baker, 2012
Inputs:
ifgramFile - string, HDF5 file name of the interferograms
coherenceFile - string, HDF5 file name of the coherence
inps_dict - dict, including the following options:
weight_function
min_coherence
max_coherence
Output:
timeseriesFile - string, HDF5 file name of the output timeseries
tempCohFile - string, HDF5 file name of temporal coherence
'''
total = time.time()
if not inps_dict:
inps_dict = vars(cmdLineParse())
weight_func = inps_dict['weight_function']
min_coh = inps_dict['min_coherence']
max_coh = inps_dict['max_coherence']
# Basic Info
atr = readfile.read_attribute(ifgramFile)
length = int(atr['FILE_LENGTH'])
width = int(atr['WIDTH'])
pixel_num = length * width
h5ifgram = h5py.File(ifgramFile, 'r')
ifgram_list = sorted(h5ifgram['interferograms'].keys())
if inps_dict['weight_function'] == 'no':
ifgram_list = ut.check_drop_ifgram(h5ifgram, atr, ifgram_list)
ifgram_num = len(ifgram_list)
# Convert ifgram_list to date12/8_list
date12_list = ptime.list_ifgram2date12(ifgram_list)
m_dates = [i.split('-')[0] for i in date12_list]
s_dates = [i.split('-')[1] for i in date12_list]
date8_list = ptime.yyyymmdd(sorted(list(set(m_dates + s_dates))))
date_num = len(date8_list)
tbase_list = ptime.date_list2tbase(date8_list)[0]
tbase_diff = np.diff(tbase_list).reshape((date_num - 1, 1))
print 'number of interferograms: ' + str(ifgram_num)
print 'number of acquisitions : ' + str(date_num)
print 'number of pixels: ' + str(pixel_num)
# Reference pixel in space
try:
ref_x = int(atr['ref_x'])
ref_y = int(atr['ref_y'])
print 'reference pixel in y/x: [%d, %d]' % (ref_y, ref_x)
except:
print 'ERROR: No ref_x/y found! Can not inverse interferograms without reference in space.'
print 'run seed_data.py ' + ifgramFile + ' --mark-attribute for a quick referencing.'
sys.exit(1)
##### Read Interferograms
print 'reading interferograms ...'
ifgram_data = np.zeros((ifgram_num, pixel_num), np.float32)
prog_bar = ptime.progress_bar(maxValue=ifgram_num)
for j in range(ifgram_num):
ifgram = ifgram_list[j]
d = h5ifgram['interferograms'][ifgram].get(ifgram)[:]
#d[d != 0.] -= d[ref_y, ref_x]
d -= d[ref_y, ref_x]
ifgram_data[j] = d.flatten()
prog_bar.update(j + 1, suffix=date12_list[j])
h5ifgram.close()
prog_bar.close()
#####---------------------- Inversion ----------------------#####
# Design matrix
A, B = ut.design_matrix(ifgramFile, date12_list)
if weight_func == 'no':
print 'generalized inversion using SVD (Berardino et al., 2002, IEEE-TGRS)'
print 'inversing time series ...'
B_inv = np.array(np.linalg.pinv(B), np.float32)
ts_rate = np.dot(B_inv, ifgram_data)
ts1 = ts_rate * np.tile(tbase_diff, (1, pixel_num))
ts0 = np.array([0.] * pixel_num, np.float32)
ts_data = np.vstack((ts0, np.cumsum(ts1, axis=0)))
del ts_rate, ts0, ts1
# Temporal coherence
print 'calculating temporal coherence (Tizzani et al., 2007, RSE)'
temp_coh = np.zeros((1, pixel_num), np.float32) + 0j
prog_bar = ptime.progress_bar(maxValue=ifgram_num)
for i in range(ifgram_num):
ifgram_est = np.dot(A[i, :], ts_data[1:, :])
ifgram_diff = ifgram_data[i, :] - ifgram_est
temp_coh += np.exp(1j * ifgram_diff)
prog_bar.update(i + 1, suffix=date12_list[i])
prog_bar.close()
del ifgram_data, ifgram_est, ifgram_diff
temp_coh = np.array((np.absolute(temp_coh) / ifgram_num).reshape(
(length, width)),
dtype=np.float32)
else:
print 'weighted least square (WLS) inversion using coherence pixel by pixel'
if np.linalg.matrix_rank(A) < date_num - 1:
print 'ERROR: singular design matrix!'
print ' Input network of interferograms is not fully connected!'
print ' Can not inverse the weighted least square solution.'
print 'You could try:'
print ' 1) Add more interferograms to make the network fully connected:'
print ' a.k.a., no multiple subsets nor network islands'
print " 2) Use '-w no' option for non-weighted SVD solution."
sys.exit(-1)
pixel_mask = np.ones(pixel_num, np.bool_)
print 'reading coherence: ' + os.path.basename(coherenceFile)
h5coh = h5py.File(coherenceFile, 'r')
coh_list = sorted(h5coh['coherence'].keys())
coh_data = np.zeros((ifgram_num, pixel_num), np.float32)
prog_bar = ptime.progress_bar(maxValue=ifgram_num)
for j in range(ifgram_num):
ifgram = coh_list[j]
d = h5coh['coherence'][ifgram].get(ifgram)[:].flatten()
d[np.isnan(d)] = 0.
pixel_mask[d == 0.] = 0
coh_data[j] = d
prog_bar.update(j + 1, suffix=date12_list[j])
h5coh.close()
prog_bar.close()
# Get mask of valid pixels to inverse
print 'skip pixels with zero coherence in at least one interferogram'
print 'skip pixels with zero phase in all interferograms'
ifgram_stack = ut.get_file_stack(ifgramFile).flatten()
pixel_mask[ifgram_stack == 0.] = 0
pixel_num2inv = np.sum(pixel_mask)
pixel_idx2inv = np.where(pixel_mask)[0]
ifgram_data = ifgram_data[:, pixel_mask]
coh_data = coh_data[:, pixel_mask]
print 'number of pixels to inverse: %d' % (pixel_num2inv)
##### Calculate Weight matrix
weight = coh_data
if weight_func.startswith('var'):
print 'convert coherence to weight using inverse of variance: x**2/(1-x**2) from Hanssen (2001, for 4.2.32)'
weight[weight > 0.999] = 0.999
if weight_func == 'variance-max-coherence':
print 'constrain the max coherence to %f' % max_coh
weight[weight > max_coh] = max_coh
weight = np.square(weight)
weight *= 1. / (1. - weight)
if weight_func == 'variance-log':
print 'use log(1/variance)+1 as weight'
weight = np.log(weight + 1)
elif weight_func.startswith('lin'):
print 'use coherence as weight directly (Tong et al., 2016, RSE)'
elif weight_func.startswith('norm'):
print 'convert coherence to weight using CDF of normal distribution: N(%f, %f)' % (
mu, std)
mu = (min_coh + max_coh) / 2.0
std = (max_coh - min_coh) / 6.0
chunk_size = 1000
chunk_num = int(pixel_num2inv / chunk_size) + 1
prog_bar = ptime.progress_bar(maxValue=chunk_num)
for i in range(chunk_num):
i0 = (i - 1) * chunk_size
i1 = min([pixel_num2inv, i0 + chunk_size])
weight[:, i0:i1] = norm.cdf(weight[:, i0:i1], mu, std)
prog_bar.update(i + 1, every=10)
prog_bar.close()
#weight = norm.cdf(weight, mu, std)
else:
print 'Un-recognized weight function: %s' % weight_func
sys.exit(-1)
##### Weighted Inversion pixel by pixel
print 'inversing time series ...'
ts_data = np.zeros((date_num, pixel_num), np.float32)
temp_coh = np.zeros(pixel_num, np.float32)
prog_bar = ptime.progress_bar(maxValue=pixel_num2inv)
for i in range(pixel_num2inv):
# Inverse timeseries
ifgram_pixel = ifgram_data[:, i]
weight_pixel = weight[:, i]
W = np.diag(weight_pixel)
ts = np.linalg.inv(A.T.dot(W).dot(A)).dot(
A.T).dot(W).dot(ifgram_pixel)
ts_data[1:, pixel_idx2inv[i]] = ts
# Calculate weighted temporal coherence
ifgram_diff = ifgram_pixel - np.dot(A, ts)
temp_coh_pixel = np.abs(
np.sum(np.multiply(weight_pixel, np.exp(1j * ifgram_diff)),
axis=0)) / np.sum(weight_pixel)
temp_coh[pixel_idx2inv[i]] = temp_coh_pixel
prog_bar.update(i + 1, every=2000, suffix=str(i + 1) + ' pixels')
prog_bar.close()
del ifgram_data, weight
#####---------------------- Outputs ----------------------#####
## 1.1 Convert time-series phase to displacement
print 'converting phase to range'
phase2range = -1 * float(atr['WAVELENGTH']) / (4. * np.pi)
ts_data *= phase2range
## 1.2 Write time-series data matrix
timeseriesFile = 'timeseries.h5'
print 'writing >>> ' + timeseriesFile
print 'number of acquisitions: ' + str(date_num)
h5timeseries = h5py.File(timeseriesFile, 'w')
group = h5timeseries.create_group('timeseries')
prog_bar = ptime.progress_bar(maxValue=date_num)
for i in range(date_num):
date = date8_list[i]
dset = group.create_dataset(date,
data=ts_data[i].reshape(length, width),
compression='gzip')
prog_bar.update(i + 1, suffix=date)
prog_bar.close()
## 1.3 Write time-series attributes
print 'calculating perpendicular baseline timeseries'
pbase, pbase_top, pbase_bottom = ut.perp_baseline_ifgram2timeseries(
ifgramFile, ifgram_list)
pbase = str(pbase.tolist()).translate(
None, '[],') # convert np.array into string separated by white space
pbase_top = str(pbase_top.tolist()).translate(None, '[],')
pbase_bottom = str(pbase_bottom.tolist()).translate(None, '[],')
atr['P_BASELINE_TIMESERIES'] = pbase
atr['P_BASELINE_TOP_TIMESERIES'] = pbase_top
atr['P_BASELINE_BOTTOM_TIMESERIES'] = pbase_bottom
atr['ref_date'] = date8_list[0]
atr['FILE_TYPE'] = 'timeseries'
atr['UNIT'] = 'm'
for key, value in atr.iteritems():
group.attrs[key] = value
h5timeseries.close()
del ts_data
## 2. Write Temporal Coherence File
tempCohFile = 'temporalCoherence.h5'
print 'writing >>> ' + tempCohFile
atr['FILE_TYPE'] = 'temporal_coherence'
atr['UNIT'] = '1'
writefile.write(temp_coh.reshape(length, width), atr, tempCohFile)
print 'Time series inversion took ' + str(time.time() -
total) + ' secs\nDone.'
return timeseriesFile, tempCohFile
def ifgram_inversion_patch(ifgramFile, coherenceFile, meta, box=None):
'''
Inputs:
ifgramFile - string, interferograms hdf5 file
coherenceFile - string, coherence hdf5 file
box - 4-tuple, left, upper, right, and lower pixel coordinate of area of interest
meta - dict, including the following attributes:
#Interferograms
length/width - int, file size for each interferogram
ifgram_list - list of string, interferogram dataset name
date12_list - list of string, YYMMDD-YYMMDD
ref_value - np.array in size of (ifgram_num, 1)
reference pixel coordinate in row/column number
ref_y/x - int, reference pixel coordinate in row/column number
#Time-series
date8_list - list of string in YYYYMMDD
tbase_diff - np.array in size of (date_num-1, 1), differential temporal baseline
#Inversion
weight_function -
'''
##### Get patch size/index
if not box:
box = (0, 0, meta['width'], meta['length'])
c0, r0, c1, r1 = box
print 'processing %8d/%d lines ...' % (r1, meta['length'])
## Initiate output data matrixs
row_num = r1 - r0
col_num = c1 - c0
pixel_num = row_num * col_num
date_num = len(meta['date8_list'])
ts = np.zeros((date_num, pixel_num), np.float32)
temp_coh = np.zeros(pixel_num, np.float32)
##### Get mask of non-zero pixels
print 'skip pixels with zero/nan value in all interferograms'
ifgram_stack = ut.get_file_stack(ifgramFile)[r0:r1, c0:c1].flatten()
mask = ~np.isnan(ifgram_stack)
mask[ifgram_stack == 0.] = 0
pixel_num2inv = np.sum(mask)
pixel_idx2inv = np.where(mask)[0]
print 'number of pixels to inverse: %d' % (pixel_num2inv)
if pixel_num2inv < 1:
ts = ts.reshape(date_num, row_num, col_num)
temp_coh = temp_coh.reshape(row_num, col_num)
return ts, temp_coh
##### Read interferograms
ifgram_num = len(meta['ifgram_list'])
ifgram_data = np.zeros((ifgram_num, pixel_num2inv), np.float32)
date12_list = meta['date12_list']
atr = readfile.read_attribute(ifgramFile)
h5ifgram = h5py.File(ifgramFile, 'r')
for j in range(ifgram_num):
ifgram = meta['ifgram_list'][j]
d = h5ifgram['interferograms'][ifgram].get(ifgram)[r0:r1, c0:c1]
ifgram_data[j] = d.flatten()[mask]
sys.stdout.write('\rreading interferograms %s/%s ...' %
(j + 1, ifgram_num))
sys.stdout.flush()
print ' '
h5ifgram.close()
ifgram_data -= meta['ref_value']
##### Design matrix
A, B = ut.design_matrix(ifgramFile, date12_list)
B_inv = np.array(np.linalg.pinv(B), np.float32)
##### Inverse
if meta['weight_function'] in ['no', 'uniform']:
print 'inversing time-series ...'
ts[1:, mask] = network_inversion_sbas(B,
ifgram_data,
meta['tbase_diff'],
B_inv=B_inv)
print 'calculating temporal coherence ...'
temp_coh[mask] = temporal_coherence(A, ts[1:, mask], ifgram_data)
else:
##### Read coherence
coh_data = np.zeros((ifgram_num, pixel_num2inv), np.float32)
h5coh = h5py.File(coherenceFile, 'r')
coh_list = sorted(h5coh['coherence'].keys())
coh_list = ut.check_drop_ifgram(h5coh)
for j in range(ifgram_num):
ifgram = coh_list[j]
d = h5coh['coherence'][ifgram].get(ifgram)[r0:r1, c0:c1]
d[np.isnan(d)] = 0.
coh_data[j] = d.flatten()[mask]
sys.stdout.write('\rreading coherence %s/%s ...' %
(j + 1, ifgram_num))
sys.stdout.flush()
print ' '
h5coh.close()
##### Calculate Weight matrix
weight = coh_data
if meta['weight_function'].startswith('var'):
print 'convert coherence to weight using inverse of phase variance'
print ' with phase PDF for distributed scatterers from Tough et al. (1995)'
L = int(atr['ALOOKS']) * int(atr['RLOOKS'])
lineStr = ' number of multilooks L=%d' % L
if L > 80:
L = 80
lineStr += ', use L=80 to avoid dividing by 0 in calculation with Negligible effect'
print lineStr
weight = 1.0 / coherence2phase_variance_ds(weight, L)
elif meta['weight_function'].startswith(('lin', 'coh', 'cor')):
print 'use coherence as weight directly (Perissin & Wang, 2012; Tong et al., 2016)'
epsilon = 1e-4
weight[weight < epsilon] = epsilon
else:
print 'Un-recognized weight function: %s' % meta['weight_function']
sys.exit(-1)
##### Weighted Inversion pixel by pixel
print 'inversing time series ...'
prog_bar = ptime.progress_bar(maxValue=pixel_num2inv)
for i in range(pixel_num2inv):
ts[1:, pixel_idx2inv[i]] = network_inversion_wls(
A, ifgram_data[:, i], weight[:, i])[0].flatten()
prog_bar.update(i + 1,
every=100,
suffix=str(i + 1) + '/' + str(pixel_num2inv) +
' pixels')
prog_bar.close()
print 'calculating temporal coherence ...'
#temp_coh[mask] = temporal_coherence(A, ts[1:,mask], ifgram_data, weight)
temp_coh[mask] = temporal_coherence(A, ts[1:, mask], ifgram_data)
ts = ts.reshape(date_num, row_num, col_num)
temp_coh = temp_coh.reshape(row_num, col_num)
##Write to temp hdf5 files for parallel processing
if meta['parallel']:
fname = meta['ftemp_base'] + str(int(r0 / meta['row_step'])) + '.h5'
print 'writing >>> ' + fname
h5temp = h5py.File(fname, 'w')
group = h5temp.create_group('timeseries')
dset = group.create_dataset('timeseries',
shape=(date_num + 1, row_num, col_num),
dtype=np.float32)
dset[0:-1, :, :] = ts
dset[1, :, :] = temp_coh
h5temp.close()
return
else:
return ts, temp_coh
