def _build_observation_constraint ( self, state_grid, image, image_mask, \
emulator, band_uncertainty, factor, band_pass, bw, per_band=True):
"""This is just a method to build up the observation operator."""
self.the_observations = ObservationOperatorImageGP ( state_grid, \
self.the_state, image, image_mask, emulator, \
bu=band_uncertainty, factor=factor, \
band_pass=band_pass, bw=bw, per_band=per_band )
def _build_observation_constraint ( self, state_grid, image, image_mask, \
emulator, band_uncertainty, factor, band_pass, bw, per_band=True):
"""This is just a method to build up the observation operator."""
self.the_observations = ObservationOperatorImageGP ( state_grid, \
self.the_state, image, image_mask, emulator, \
bu=band_uncertainty, factor=factor, \
band_pass=band_pass, bw=bw, per_band=per_band )
class SingleImageProcessor ( object ):
"""This class sets up and solves a single date/single image using
eoldas_ng"""
def __init__ ( self, state_config, state_grid, image, band_unc, mask, \
bw, band_pass, emulator, prior, regularisation, process_name, \
factor, optimisation_options=None, verbose=False ):
"""The class creator. Takes state configuration and grids (spatial),
band uncertainties, image and associated mask, relevant emulator, a
prior object and a regularisation (spatial) model. We also have a
process name to store the results more or less neatly.
TODO We ought to have private methods to set up the different
objects, as that allows the user to select a different type of
"pre-packaged" state (e.g. not the prosail one).
Parameters
-----------
state_config: dict
A state configuration object
state_grid: array
A state grid (2D)
image: array
A multispectral image (3D)
band_unc: array
Per band uncertainty. Note that I need to check whether you
can have a per pixel uncertainty, or whether that requires
some extra recoding. CHECK
mask: array
A 2D array, indicating pixels which are OK and pixels which
aren't
bw: array?
A BW array. This is needed for somem reason, but i think it's
superfluous CHECK
band_pass: list
A list of band_pass objects. We use this list (and the ``bw``
parameter above) to calculate per band emulators and the
initial inverse emulators.
emulator: list
The emulator for this particular image acquisition geometry.
In principle this is a MultivariateEmulator in a single element
list, but could be per band emulators
prior: Prior
An ``eoldas_ng`` prior object
regularisation: SpatialSmoother
An ``eoldas_ng`` SpatialSmoother object.
process_name: str
A string with the process name. This is used to store the
results of inverting the current image
factor: int
The spatial factor: how many times does this observation
fit in with the state spatial resolution?
optimisation_options: dict
A dictionary of optimisation options
verbose: bool
A verbose flag
"""
# Get the state [CENSORED]
self.the_state = StandardStatePROSAIL ( state_config, state_grid, \
optimisation_options=optimisation_options, \
output_name=process_name, verbose=verbose )
self._build_observation_constraint ( state_grid, image, mask,
emulator, band_unc, factor,
band_pass, bw )
# Note that this isn't very satisfactory, but we can always change
# that: the prior just needs a mean and a sparse covariance, and
# the regularisation needs the gamma(s) and parameters to apply it
# to
self.the_prior = prior
self.the_model = regularisation
print "WARNING! No prior involved here!"
print "PRIOR needs defining!"
#self.the_state.add_operator ( "Prior", self.the_prior )
self.the_state.add_operator ( "Regularisation", self.the_model )
self.the_state.add_operator ( "Observations", self.the_observations )
def _build_observation_constraint ( self, state_grid, image, image_mask, \
emulator, band_uncertainty, factor, band_pass, bw, per_band=True):
"""This is just a method to build up the observation operator."""
self.the_observations = ObservationOperatorImageGP ( state_grid, \
self.the_state, image, image_mask, emulator, \
bu=band_uncertainty, factor=factor, \
band_pass=band_pass, bw=bw, per_band=per_band )
def first_guess ( self ):
return self.the_observations.first_guess ( \
self.the_state.state_config )
def solve ( self, x0=None ):
return self.the_state.optimize ( x0=x0, do_unc=True )
class SingleImageProcessor(object):
"""This class sets up and solves a single date/single image using
eoldas_ng"""
def __init__ ( self, state_config, state_grid, image, band_unc, mask, \
bw, band_pass, emulator, prior, regularisation, process_name, \
factor, optimisation_options=None, verbose=False ):
"""The class creator. Takes state configuration and grids (spatial),
band uncertainties, image and associated mask, relevant emulator, a
prior object and a regularisation (spatial) model. We also have a
process name to store the results more or less neatly.
TODO We ought to have private methods to set up the different
objects, as that allows the user to select a different type of
"pre-packaged" state (e.g. not the prosail one).
Parameters
-----------
state_config: dict
A state configuration object
state_grid: array
A state grid (2D)
image: array
A multispectral image (3D)
band_unc: array
Per band uncertainty. Note that I need to check whether you
can have a per pixel uncertainty, or whether that requires
some extra recoding. CHECK
mask: array
A 2D array, indicating pixels which are OK and pixels which
aren't
bw: array?
A BW array. This is needed for somem reason, but i think it's
superfluous CHECK
band_pass: list
A list of band_pass objects. We use this list (and the ``bw``
parameter above) to calculate per band emulators and the
initial inverse emulators.
emulator: list
The emulator for this particular image acquisition geometry.
In principle this is a MultivariateEmulator in a single element
list, but could be per band emulators
prior: Prior
An ``eoldas_ng`` prior object
regularisation: SpatialSmoother
An ``eoldas_ng`` SpatialSmoother object.
process_name: str
A string with the process name. This is used to store the
results of inverting the current image
factor: int
The spatial factor: how many times does this observation
fit in with the state spatial resolution?
optimisation_options: dict
A dictionary of optimisation options
verbose: bool
A verbose flag
"""
# Get the state [CENSORED]
self.the_state = StandardStatePROSAIL ( state_config, state_grid, \
optimisation_options=optimisation_options, \
output_name=process_name, verbose=verbose )
self._build_observation_constraint(state_grid, image, mask, emulator,
band_unc, factor, band_pass, bw)
# Note that this isn't very satisfactory, but we can always change
# that: the prior just needs a mean and a sparse covariance, and
# the regularisation needs the gamma(s) and parameters to apply it
# to
self.the_prior = prior
self.the_model = regularisation
print "WARNING! No prior involved here!"
print "PRIOR needs defining!"
#self.the_state.add_operator ( "Prior", self.the_prior )
self.the_state.add_operator("Regularisation", self.the_model)
self.the_state.add_operator("Observations", self.the_observations)
def _build_observation_constraint ( self, state_grid, image, image_mask, \
emulator, band_uncertainty, factor, band_pass, bw, per_band=True):
"""This is just a method to build up the observation operator."""
self.the_observations = ObservationOperatorImageGP ( state_grid, \
self.the_state, image, image_mask, emulator, \
bu=band_uncertainty, factor=factor, \
band_pass=band_pass, bw=bw, per_band=per_band )
def first_guess(self):
return self.the_observations.first_guess ( \
self.the_state.state_config )
def solve(self, x0=None):
return self.the_state.optimize(x0=x0, do_unc=True)
