def gen_abundance_maps(data, U, result_path):
print('Abundance maps generation with NNLS')
nnls = amp.NNLS()
amaps = nnls.map(data, U, normalize=True)
nnls.plot(result_path, colorMap='jet', suffix='gas')
# return an array of abundance maps
return amaps
def NNLS(data, U, umix_source, mask, path):
import pysptools.abundance_maps as amp
print(' Testing NNLS')
nnls = amp.NNLS()
amaps = nnls.map(data, U, normalize=True)
nnls.plot(path, colorMap='jet', suffix=umix_source)
nnls.plot(path, interpolation='spline36', suffix=umix_source + '_spline36')
return amaps
def test_NNLS(data, U, umix_source, mask, path):
print(' Testing NNLS')
nnls = amp.NNLS()
pr = profile()
amap = nnls.map(data, U, normalize=True)
stat(pr)
nnls.plot(path, colorMap='jet', suffix=umix_source)
nnls.plot(path, mask=mask, colorMap='jet', suffix=umix_source + '_mask')
nnls.plot(path, interpolation='spline36', suffix=umix_source + '_spline36')
def __init__(self, hcube, n_em, suffix):
self.suffix = suffix
self.nfindr = eea.NFINDR()
self.U = self.nfindr.extract(hcube,
n_em,
maxit=5,
normalize=False,
ATGP_init=True)
# self.xxls = amp.FCLS()
self.xxls = amp.NNLS()
self.amaps = self.xxls.map(hcube, self.U, normalize=False)
def main():
try:
import pysptools.eea as eea
except ImportError:
gs.fatal(_("Cannot import pysptools \
(https://pypi.python.org/pypi/pysptools) library."
" Please install it (pip install pysptools)"
" or ensure that it is on path"
" (use PYTHONPATH variable)."))
try:
# sklearn is a dependency of used pysptools functionality
import sklearn
except ImportError:
gs.fatal(_("Cannot import sklearn \
(https://pypi.python.org/pypi/scikit-learn) library."
" Please install it (pip install scikit-learn)"
" or ensure that it is on path"
" (use PYTHONPATH variable)."))
try:
from cvxopt import solvers, matrix
except ImportError:
gs.fatal(_("Cannot import cvxopt \
(https://pypi.python.org/pypi/cvxopt) library."
" Please install it (pip install cvxopt)"
" or ensure that it is on path"
" (use PYTHONPATH variable)."))
# Parse input options
input = options['input']
output = options['output']
prefix = options['prefix']
endmember_n = int(options['endmember_n'])
endmembers = options['endmembers']
if options['maxit']:
maxit = options['maxit']
else:
maxit = 0
extraction_method = options['extraction_method']
unmixing_method = options['unmixing_method']
atgp_init = True if not flags['n'] else False
# List maps in imagery group
try:
maps = gs.read_command('i.group', flags='g', group=input,
quiet=True).rstrip('\n').split('\n')
except:
pass
# Validate input
# q and maxit can be None according to manual, but does not work in current pysptools version
if endmember_n <= 0:
gs.fatal('Number of endmembers has to be > 0')
"""if (extraction_method == 'PPI' or
extraction_method == 'NFINDR'):
gs.fatal('Extraction methods PPI and NFINDR require endmember_n >= 2')
endmember_n = None"""
if maxit <= 0:
maxit = 3 * len(maps)
if endmember_n > len(maps) + 1:
gs.warning('More endmembers ({}) requested than bands in \
input imagery group ({})'.format(endmember_n, len(maps)))
if extraction_method != 'PPI':
gs.fatal('Only PPI method can extract more endmembers than number \
of bands in the imagery group')
if not atgp_init and extraction_method != 'NFINDR':
gs.verbose('ATGP is only taken into account in \
NFINDR extraction method...')
# Get metainformation from input bands
band_types = {}
img = None
n = 0
gs.verbose('Reading imagery group...')
for m in maps:
map = m.split('@')
# Build numpy stack from imagery group
raster = r.raster2numpy(map[0], mapset=map[1])
if raster == np.float64:
raster = float32(raster)
gs.warning('{} is of type Float64.\
Float64 is currently not supported.\
Reducing precision to Float32'.format(raster))
# Determine map type
band_types[map[0]] = get_rastertype(raster)
# Create cube and mask from GRASS internal NoData value
if n == 0:
img = raster
# Create mask from GRASS internal NoData value
mask = mask_rasternd(raster)
else:
img = np.dstack((img, raster))
mask = np.logical_and((mask_rasternd(raster)), mask)
n = n + 1
# Read a mask if present and give waringing if not
# Note that otherwise NoData is read as values
gs.verbose('Checking for MASK...')
try:
MASK = r.raster2numpy('MASK', mapset=getenv('MAPSET')) == 1
mask = np.logical_and(MASK, mask)
MASK = None
except:
pass
if extraction_method == 'NFINDR':
# Extract endmembers from valid pixels using NFINDR function from pysptools
gs.verbose('Extracting endmembers using NFINDR...')
nfindr = eea.NFINDR()
E = nfindr.extract(img, endmember_n, maxit=maxit, normalize=False,
ATGP_init=atgp_init, mask=mask)
elif extraction_method == 'PPI':
# Extract endmembers from valid pixels using PPI function from pysptools
gs.verbose('Extracting endmembers using PPI...')
ppi = eea.PPI()
E = ppi.extract(img, endmember_n, numSkewers=10000, normalize=False,
mask=mask)
elif extraction_method == 'FIPPI':
# Extract endmembers from valid pixels using FIPPI function from pysptools
gs.verbose('Extracting endmembers using FIPPI...')
fippi = eea.FIPPI()
# q and maxit can be None according to manual, but does not work
"""if not maxit and not endmember_n:
E = fippi.extract(img, q=None, normalize=False, mask=mask)
if not maxit:
E = fippi.extract(img, q=endmember_n, normalize=False, mask=mask)
if not endmember_n:
E = fippi.extract(img, q=int(), maxit=maxit, normalize=False,
mask=mask)
else:
E = fippi.extract(img, q=endmember_n, maxit=maxit, normalize=False,
mask=mask)"""
E = fippi.extract(img, q=endmember_n, maxit=maxit, normalize=False,
mask=mask)
# Write output file in format required for i.spec.unmix addon
if output:
gs.verbose('Writing spectra file...')
n = 0
with open(output, 'w') as o:
o.write('# Channels: {}\n'.format('\t'.join(band_types.keys())))
o.write('# Wrote {} spectra line wise.\n#\n'.format(endmember_n))
o.write('Matrix: {0} by {1}\n'.format(endmember_n, len(maps)))
for e in E:
o.write('row{0}: {1}\n'.format(n, '\t'.join([str(i) for i in e])))
n = n + 1
# Write vector map with endmember information if requested
if endmembers:
gs.verbose('Writing vector map with endmembers...')
from grass.pygrass import utils as u
from grass.pygrass.gis.region import Region
from grass.pygrass.vector import Vector
from grass.pygrass.vector import VectorTopo
from grass.pygrass.vector.geometry import Point
# Build attribute table
# Deinfe columns for attribute table
cols = [(u'cat', 'INTEGER PRIMARY KEY')]
for b in band_types.keys():
cols.append((b.replace('.','_'), band_types[b]))
# Get region information
reg = Region()
# Create vector map
new = Vector(endmembers)
new.open('w', tab_name=endmembers, tab_cols=cols)
cat = 1
for e in E:
# Get indices
idx = np.where((img[:,:]==e).all(-1))
# Numpy array is ordered rows, columns (y,x)
if len(idx[0]) == 0 or len(idx[1]) == 0:
gs.warning('Could not compute coordinated for endmember {}. \
Please consider rescaling your data to integer'.format(cat))
cat = cat + 1
continue
coords = u.pixel2coor((idx[1][0], idx[0][0]), reg)
point = Point(coords[1] + reg.ewres / 2.0,
coords[0] - reg.nsres / 2.0)
# Get attributes
n = 0
attr = []
for b in band_types.keys():
if band_types[b] == u'INTEGER':
attr.append(int(e[n]))
else:
attr.append(float(e[n]))
n = n + 1
# Write geometry with attributes
new.write(point, cat=cat,
attrs=tuple(attr))
cat = cat + 1
# Close vector map
new.table.conn.commit()
new.close(build=True)
if prefix:
# Run spectral unmixing
import pysptools.abundance_maps as amaps
if unmixing_method == 'FCLS':
fcls = amaps.FCLS()
result = fcls.map(img, E, normalize=False, mask=mask)
elif unmixing_method == 'NNLS':
nnls = amaps.NNLS()
result = nnls.map(img, E, normalize=False, mask=mask)
elif unmixing_method == 'UCLS':
ucls = amaps.UCLS()
result = ucls.map(img, E, normalize=False, mask=mask)
# Write results
for l in range(endmember_n):
rastname = '{0}_{1}'.format(prefix, l + 1)
r.numpy2raster(result[:,:,l], 'FCELL', rastname)
def getWaterFraction(cls,
ds,
cloudThresh=-20,
constrain=True,
maskClouds=True):
if maskClouds:
atmsNoClouds = cls.maskClouds(ds, threshold=cloudThresh)
else:
atmsNoClouds = ds.copy()
dBtr = atmsNoClouds.sel(band='C4').astype(
np.float) - atmsNoClouds.sel(band='C3').astype(np.float)
dBtr.coords['band'] = 'dBtr'
channels = xr.concat([
atmsNoClouds.sel(band=['C3', 'C4', 'C16']).isel(time=0, z=0),
dBtr.isel(time=0, z=0)
],
dim='band')
arr = channels.values
arr[np.isnan(arr)] = -9999
nClasses = 3
nfindr = eea.NFINDR()
U = nfindr.extract(arr,
nClasses,
maxit=100,
normalize=True,
ATGP_init=True)
drop = np.argmin(list(map(lambda x: U[x, :].mean(), range(nClasses))))
waterIdx = np.argmin(
list(
map(lambda x: np.delete(U, drop, axis=1)[x, :],
range(nClasses - 2))))
if waterIdx == 0:
bandList = ['water', 'land', 'mask']
else:
bandList = ['land', 'water', 'mask']
nnls = amp.NNLS()
amaps = nnls.map(arr, U, normalize=True)
drop = np.argmin(
list(map(lambda x: amaps[:, :, x].mean(), range(amaps.shape[2]))))
unmixed = np.delete(amaps, drop, axis=2)
unmixed[unmixed == 0] = np.nan
scaled = np.zeros_like(unmixed)
for i in range(scaled.shape[2]):
summed = unmixed[:, :, i] / unmixed.sum(axis=2)
scaled[:, :,
i] = (summed - np.nanmin(summed)) / (np.nanmax(summed) -
np.nanmin(summed))
scaled = scaled - 0.25
scaled[scaled < 0] = 0
fWater = atmsNoClouds.sel(band=['C1', 'C2', 'mask']).copy()
fWater[:, :, 0, :2, 0] = scaled[:, :, :]
fWater.coords['band'] = bandList
return fWater.raster.updateMask(atmsNoClouds.sel(band='mask'))
# N-FINDR
# useful in situations where the basis spectra are already present in the dataset
# New modules: pysptools
from pysptools.eea import nfindr
import pysptools.abundance_maps as amp
num_endmembers = 3
# finde the endmembers
comps = nfindr.NFINDR(h5_main[:].copy(), num_endmembers)[0]
# calculate abundance maps
nnls = amp.NNLS()
abundances=nnls.map(h5_main[:].copy().reshape(h5_main.pos_dim_sizes[0],\
h5_main.pos_dim_sizes[1], -1),comps)
# plot the components
usid.plot_utils.plot_map_stack(comps.reshape(num_endmembers,\
h5_main.spec_dim_sizes[0], h5_main.spec_dim_sizes[1]),title = '',\
color_bar_mode='each')
plt.savefig("spectral-decomposition-10.jpg", dpi=300)
plt.show()
# plot the abundances
usid.plot_utils.plot_map_stack(abundances.transpose(2, 0, 1),
