# 10 consecutive channels must be above the MAD level to be real emission.
num_chans = 7
peak_snr = 4.5
# Cutoff level
min_snr = 3
# Where to cut at the line edges
edge_thresh = 0.5
# Smooth the cube, then create a noise model
spec_kernel = Box1DKernel(smooth_chans)
smooth_cube = cube.spectral_smooth(spec_kernel)
noise = Noise(smooth_cube)
noise.estimate_noise(spectral_flat=True)
noise.get_scale_cube()
snr = noise.snr.copy()
posns = np.where(snr.max(axis=0) >= min_snr)
bad_pos = np.where(snr.max(axis=0) < min_snr)
mask[:, bad_pos[0], bad_pos[1]] = False
# In case single spectra need to be inspected.
verbose = False
for i, j in ProgressBar(zip(*posns)):
spectrum = cube[:, i, j].value
# 10 consecutive channels must be above the MAD level to be real emission.
num_chans = 7
peak_snr = 4.5
# Cutoff level
min_snr = 3
# Where to cut at the line edges
edge_thresh = 0.5
# Smooth the cube, then create a noise model
spec_kernel = Box1DKernel(smooth_chans)
smooth_cube = cube.spectral_smooth(spec_kernel)
noise = Noise(smooth_cube)
noise.estimate_noise(spectral_flat=True)
noise.get_scale_cube()
snr = noise.snr.copy()
posns = np.where(snr.max(axis=0) >= min_snr)
bad_pos = np.where(snr.max(axis=0) < min_snr)
mask[:, bad_pos[0], bad_pos[1]] = False
# In case single spectra need to be inspected.
verbose = False
for i, j in ProgressBar(zip(*posns)):
spectrum = cube[:, i, j].value
def make_signal_mask(cube, smooth_chans=200. / 66., min_chan=7, peak_snr=5.,
min_snr=3.5, edge_thresh=1.5, verbose=False):
'''
Create a robust signal mask by requiring spatial and spectral
connectivity.
'''
import astropy.units as u
from astropy.convolution import Box1DKernel
from signal_id import Noise
from scipy import ndimage as nd
from astropy.wcs.utils import proj_plane_pixel_scales
from astropy.utils.console import ProgressBar
import skimage.morphology as mo
import numpy as np
from radio_beam import Beam
from itertools import groupby, chain
from operator import itemgetter
import matplotlib.pyplot as p
pixscale = proj_plane_pixel_scales(cube.wcs)[0]
# # Want to smooth the mask edges
mask = cube.mask.include().copy()
# Set smoothing parameters and # consecutive channels.
smooth_chans = int(round_up_to_odd(smooth_chans))
# consecutive channels to be real emission.
num_chans = min_chan
# Smooth the cube, then create a noise model
spec_kernel = Box1DKernel(smooth_chans)
smooth_cube = cube.spectral_smooth(spec_kernel)
noise = Noise(smooth_cube)
noise.estimate_noise(spectral_flat=True)
noise.get_scale_cube()
snr = noise.snr.copy()
snr[np.isnan(snr)] = 0.0
posns = np.where(snr.max(axis=0) >= min_snr)
bad_pos = np.where(snr.max(axis=0) < min_snr)
mask[:, bad_pos[0], bad_pos[1]] = False
for i, j in ProgressBar(zip(*posns)):
# Look for all pixels above min_snr
good_posns = np.where(snr[:, i, j] > min_snr)[0]
# Reject if the total is less than connectivity requirement
if good_posns.size < num_chans:
mask[:, i, j] = False
continue
# Find connected pixels
sequences = []
for k, g in groupby(enumerate(good_posns), lambda (i, x): i - x):
sequences.append(map(itemgetter(1), g))
# Check length and peak. Require a minimum of 3 pixels above the noise
# to grow from.
sequences = [seq for seq in sequences if len(seq) >= 3 and
np.nanmax(snr[:, i, j][seq]) >= peak_snr]
# Continue if no good sequences found
if len(sequences) == 0:
mask[:, i, j] = False
continue
# Now take each valid sequence and expand the edges until the smoothed
# spectrum approaches zero.
edges = [[seq[0], seq[-1]] for seq in sequences]
for n, edge in enumerate(edges):
# Lower side
if n == 0:
start_posn = edge[0]
stop_posn = 0
else:
start_posn = edge[0] - edges[n - 1][0]
stop_posn = edges[n - 1][0]
for pt in np.arange(start_posn, stop_posn, -1):
# if smoothed[pt] <= mad * edge_thresh:
if snr[:, i, j][pt] <= edge_thresh:
break
sequences[n].insert(0, pt)
# Upper side
start_posn = edge[1]
if n == len(edges) - 1:
stop_posn = cube.shape[0]
else:
stop_posn = edges[n + 1][0]
for pt in np.arange(start_posn, stop_posn, 1):
# if smoothed[pt] <= mad * edge_thresh:
if snr[:, i, j][pt] <= edge_thresh:
break
sequences[n].insert(0, pt)
# Final check for the min peak level and ensure all meet the
# spectral connectivity requirement
sequences = [seq for seq in sequences if len(seq) >= num_chans and
np.nanmax(snr[:, i, j][seq]) >= peak_snr]
if len(sequences) == 0:
mask[:, i, j] = False
continue
bad_posns = \
list(set(np.arange(cube.shape[0])) - set(list(chain(*sequences))))
mask[:, i, j][bad_posns] = False
if verbose:
p.subplot(121)
p.plot(cube.spectral_axis.value, noise.snr[:, i, j])
min_val = cube.spectral_axis.value[np.where(mask[:, i, j])[0][-1]]
max_val = cube.spectral_axis.value[np.where(mask[:, i, j])[0][0]]
p.vlines(min_val, 0,
np.nanmax(noise.snr[:, i, j]))
p.vlines(max_val, 0,
np.nanmax(noise.snr[:, i, j]))
p.plot(cube.spectral_axis.value,
noise.snr[:, i, j] * mask[:, i, j], 'bD')
p.subplot(122)
p.plot(cube.spectral_axis.value, cube[:, i, j], label='Cube')
p.plot(cube.spectral_axis.value, smooth_cube[:, i, j],
label='Smooth Cube')
p.axvline(min_val)
p.axvline(max_val)
p.plot(cube.spectral_axis.value,
smooth_cube[:, i, j] * mask[:, i, j], 'bD')
p.draw()
raw_input("Next spectrum?")
p.clf()