"""
Hyperspectral imagery (hsi) class
Created on Tue Apr 12 13:06:30 2016
@author: chc
"""
import numpy as _np
import copy as _copy
from crikit.data.frequency import Frequency as _Frequency
from crikit.data.spectrum import Spectrum as _Spectrum
from crikit.data.replicate import Replicate as _Replicate
__all__ = ['Hsi']
[docs]class Hsi(_Spectrum):
"""
Hyperspectral imagery class
Parameters
----------
data : 3D ndarray [y_pix, x_pix, f_pix]
HSI image
mask : 3D ndarray (int) [y_pix, x_pix, f_pix]
0,1 mask with 1 is a usable pixel and 0 is not
freq : crikit.data.frequency.Frequency instance
Frequency [wavelength, wavenumber] object (i.e., the independent \
variable)
label : str
Image label (i.e., a string describing what the image is)
units : str
Units of image (e.g., intensity)
x_rep : crikit.data.replicate.Replicate instance, Not implemented yet
x-axis spatial object
y_rep : crikit.data.replicate.Replicate instance, Not implemented yet
x-axis spatial object
x : 1D ndarray
x-axis spatial vector
y : 1D ndarray
y-axis spatial vector
meta : dict
Meta-data dictionary
Attributes
----------
shape : tuple, read-only
Shape of data
size : int, read-only
Size of data (i.e., total number of entries)
Methods
-------
mean : 1D ndarray
Mean spectrum. If extent [a,b] is provided, calculate mean over that\
inclusive region.
std : 1D ndarray
Standard deviation of spectrum. If extent [a,b] is provided, calculate standard\
deviation over that inclusive region.
subtract : 3D ndarray or None
Subtract spectrum or object
Notes
-----
* freq object contains some useful parameters such as op_range_* and \
plot_range_*, which define spectral regions-of-interest. (It's debatable \
as to whether those parameters should be in Frequency or Spectrum classes)
"""
# Configurations
config = {}
config['nd_axis'] = -1
def __init__(self, data=None, freq=None, x=None, y=None, x_rep=None,
y_rep=None, label=None, units=None, meta=None):
super().__init__(data, freq, label, units, meta)
self._x_rep = _Replicate()
self._y_rep = _Replicate()
self._mask = None
self._x_rep = _Replicate(data=x)
self._y_rep = _Replicate(data=y)
if x is None and x_rep is not None:
self.x_rep = _copy.deepcopy(x_rep)
if y is None and y_rep is not None:
self.y_rep = _copy.deepcopy(y_rep)
[docs] @staticmethod
def _mean_axes(*args, **kwargs):
""" Inhereted from Spectrum """
raise NotImplementedError('Only applicable to Spectrum class.')
[docs] @staticmethod
def _reshape_axes(shape, spectral_axis):
"""
Parameters
----------
shape : tuple
Input data shape
spectral_axis : int
Spectral axis
Returns
-------
Reshape vector
"""
ndim = len(shape)
if ndim == 1:
out = [1, 1, 1]
out[spectral_axis] = shape[0]
elif ndim == 2: # ! Super-wonky
out = [1, shape[0], shape[1]]
elif ndim == 3:
out = shape
elif ndim > 3:
out = [-1, shape[-2], shape[-1]]
else:
raise ValueError('Shape error')
return tuple(out)
@property
def mask(self):
return self._mask
@property
def x_rep(self):
return self._x_rep
@x_rep.setter
def x_rep(self, value):
if isinstance(value, _Replicate):
self._x_rep = value
elif isinstance(value, _np.ndarray):
self._x_rep.data = value
@property
def y_rep(self):
return self._y_rep
@property
def x(self):
return self._x_rep.data
@x.setter
def x(self, value):
self._x_rep.data = value
@property
def y(self):
return self._y_rep.data
@y.setter
def y(self, value):
self._y_rep.data = value
@y_rep.setter
def y_rep(self, value):
if isinstance(value, _Replicate):
self._y_rep = value
elif isinstance(value, _np.ndarray):
self._y_rep.data = value
@property
def data(self):
return self._data
@data.setter
def data(self, value):
if not isinstance(value, _np.ndarray):
raise TypeError('data must be of type ndarray, not {}'.format(type(value)))
ax_rs = self._reshape_axes(value.shape, self.config['nd_axis'])
# self._mask = _np.ones(tuple([n for n in range(3) if n != self.config['nd_axis']]),
# dtype=_np.int)
if self.freq is None or self.freq.op_list_pix is None:
self._data = value.reshape(ax_rs)
else:
if value.shape[self.config['nd_axis']] == self.freq.op_range_pix.size:
temp = _np.zeros((self._data.shape),dtype=value.dtype)
temp[:,:,self.freq.op_range_pix] = value.reshape(ax_rs)
self._data = 1*temp
del temp
elif value.shape[self.config['nd_axis']] == self._data.shape[self.config['nd_axis']]:
temp = _np.zeros((self._data.shape),dtype=value.dtype)
temp[..., self.freq.op_range_pix] = value.reshape(ax_rs)[..., self.freq.op_range_pix]
self._data = 1*temp
del temp
[docs] def check(self):
"""
Check x, y, and freq to make sure the dimensions agree with data
"""
if self._data is None:
print('Hsi check: data is None, not checking')
else:
if self._x_rep._data is None:
self._x_rep._data = _np.arange(self.shape[1])
self._x_rep._label = 'X'
self._x_rep._units = 'pix'
print('Hsi check: setting x to pixels')
elif self._x_rep._data.size != self._data.shape[1]:
self._x_rep = _Replicate()
self._x_rep._data = _np.arange(self.shape[1])
self._x_rep._label = 'X'
self._x_rep._units = 'pix'
print('Hsi check: setting x to pixels')
if self._y_rep._data is None:
self._y_rep._data = _np.arange(self.shape[0])
self._y_rep._label = 'Y'
self._y_rep._units = 'pix'
print('Hsi check: setting y to pixels')
elif self._y_rep._data.size != self._data.shape[0]:
self._y_rep = _Replicate()
self._y_rep._data = _np.arange(self.shape[0])
self._y_rep._label = 'Y'
self._y_rep._units = 'pix'
print('Hsi check: setting y to pixels')
if self.freq._data is None:
self.freq._data = _np.arange(self.shape[-1])
self.freq._label = 'Frequency'
self.freq._units = 'pix'
print('Hsi check: setting freq to pixels')
elif self.freq._data.size != self._data.shape[-1]:
self.freq = _Frequency()
self.freq._data = _np.arange(self.shape[-1])
print('Hsi check: setting freq to pixels')
return None
[docs] def subtract(self, spectra, overwrite=True):
"""
Subtract spectrum from data
"""
# Order IS important
if isinstance(spectra, Hsi):
if overwrite:
self.data -= spectra.data
return None
else:
return self.data - spectra.data
elif isinstance(spectra, _Spectrum):
if overwrite:
self.data -= spectra.data[None,None,:]
return None
else:
return self.data - spectra.data
elif isinstance(spectra, _np.ndarray):
if spectra.shape == self.data.shape:
if overwrite:
self.data -= spectra
return None
else:
return self.data - spectra
else:
if overwrite:
self.data -= spectra[None,None,:]
return None
else:
return self.data - spectra[None,None,:]
[docs] def get_rand_spectra(self, num, pt_sz=1, quads=False, full=False):
mlen, nlen, freqlen = self.data.shape
if quads:
num_spectra = num + 5
else:
num_spectra = num
if _np.iscomplexobj(self.data):
dtype = _np.complex
else:
dtype = _np.float
temp = _np.zeros((num_spectra, self.data.shape[-1]), dtype=dtype)
quad_mid_row = int(_np.round(mlen/2))
quad_mid_col = int(_np.round(nlen/2))
center_row = (int(_np.round(mlen/3)), int(_np.round(2*mlen/3)))
center_col = (int(_np.round(nlen/3)), int(_np.round(2*nlen/3)))
start_count = 0
if quads:
# QUADS
# Bottom-left
temp[0, :] = _np.mean(self.data[0:quad_mid_row, 0:quad_mid_col, :], axis=(0, 1))
# Upper-left
temp[1, :] = _np.mean(self.data[0:quad_mid_row, quad_mid_col+1::, :], axis=(0, 1))
# Upper-right
temp[2, :] = _np.mean(self.data[quad_mid_row+1::, quad_mid_col+1::, :], axis=(0, 1))
# Bottom-right
temp[3, :] = _np.mean(self.data[quad_mid_row+1::, 0:quad_mid_col, :], axis=(0, 1))
# Center
temp[4, :] = _np.mean(self.data[center_row[0]:center_row[1], center_col[0]:center_col[1], :], axis=(0, 1))
start_count += 5
else:
pass
rand_rows = ((mlen-pt_sz-1)*_np.random.rand(num_spectra)).astype(int)
rand_cols = ((nlen-pt_sz-1)*_np.random.rand(num_spectra)).astype(int)
for count in _np.arange(start_count,num_spectra):
if pt_sz == 1:
temp[count, :] = _np.squeeze(self.data[rand_rows[count-start_count],
rand_cols[count-start_count]])
else:
rows = [rand_rows[count-start_count]-(pt_sz-1),
rand_rows[count-start_count]+pt_sz]
cols = [rand_cols[count-start_count]-(pt_sz-1),
rand_cols[count-start_count]+pt_sz]
if rows[0] < 0:
rows[0] = 0
if rows[1] >= mlen:
rows[1] = mlen-1
if cols[0] < 0:
cols[0] = 0
if cols[1] >= nlen:
cols[1] = nlen-1
if cols[0] == cols[1] or rows[0] == rows[1]:
pass
else:
temp[count,:] = _np.squeeze(_np.mean(self.data[rows[0]:rows[1], cols[0]:cols[1], :], axis=(0, 1)))
if (not full) and (self.freq.data is not None):
temp = temp[..., self.freq.op_range_pix]
return temp
def __sub__(self, spectrum):
return self.subtract(spectrum, overwrite=False)
if __name__ == '__main__': # pragma: no cover
x = _np.linspace(0,100,10)
y = _np.linspace(0,100,10)
freq = _np.arange(20)
data = _np.random.rand(10,10,20)
hs = Hsi(data=data, freq=freq, x=x, y=y)
print(hs.shape)
print(isinstance(hs, _Spectrum))