def get_tsm_pixels(self, channels):
"""Determine pixels affected by the scan motor issue.
Uses channels 1, 2, 4 and 5. Neither 3a, nor 3b.
"""
return get_tsm_idx(channels[:, :, 0], channels[:, :, 1],
channels[:, :, 4], channels[:, :, 5])
def test_get_tsm_idx(self):
"""Test identification of TSM affected pixels."""
cols = np.arange(409)
rows = np.arange(3000)
mcols, mrows = np.meshgrid(cols, rows)
# Create dummy reflectances in [0, 1] and BTs [170, 350]
refl = np.sin(mrows / float(rows.size)) * \
np.cos(mcols / float(cols.size))
bt = np.cos(mrows / float(rows.size)) * \
np.sin(mcols / float(cols.size))
bt = 170 + (350 - 170) * bt
ch1 = refl.copy()
ch2 = refl.copy()
ch4 = bt.copy()
ch5 = bt.copy()
# Add rectangle with noise to channel 1 and 4
noise_cols, noise_rows = np.meshgrid(np.arange(200, 300),
np.arange(1000, 2000))
for ch in [ch1, ch4]:
ch[noise_rows, noise_cols] += 1000 + 1000*np.random.rand(
noise_rows.shape[0], noise_rows.shape[1])
# We expect the filter to also detect the edges of the noisy
# rectangle: At the edges there is a transition from zero to
# nonzero channel diff which leads to an increased 3x3 standard
# deviation.
noise_cols_exp, noise_rows_exp = np.meshgrid(np.arange(199, 301),
np.arange(999, 2001))
idx = tsm.get_tsm_idx(ch1, ch2, ch4, ch5)
numpy.testing.assert_array_equal(idx[0], noise_rows_exp.ravel())
numpy.testing.assert_array_equal(idx[1], noise_cols_exp.ravel())