def test_mf_target_no_bg_eq_one(self):
'''Matched Filter response of target should be one.'''
from spectral.algorithms.detectors import matched_filter
(i, j) = self.target_ij
target = self.data[i, j]
mf = matched_filter(self.data, target)
np.testing.assert_approx_equal(mf[i, j], 1)
def test_mf_target_no_bg_eq_one(self):
'''Matched Filter response of target should be one.'''
from spectral.algorithms.detectors import matched_filter
(i, j) = self.target_ij
target = self.data[i, j]
mf = matched_filter(self.data, target)
np.testing.assert_approx_equal(mf[i, j], 1)
mask=None,
index=None)
print("Calculating stats of matchFilter...")
'''
Matched Filter for processing hyperspectral data
H0 background distribution -> get it from the most of the images so as to
have a good background distribution
H1 target distribution -> get it from the multiple image with the source information.
aplha(x)=\frac{(\mu_t-\mu_b)^T\Sigma^{-1}(x-\mu_b)}{(\mu_t-\mu_b)^T\Sigma^{-1}(\mu_t-\mu_b)}
OR
aplha(x) = a_hat = (transpose(x-u) . inv_cov . (t-u) / (transpose(t-u) . inv_cov . (t-u))
'''
print("b_img_data", b_img_data[:,columns:col_range,:].shape, "target_mean : ", \
target_mean[start_pos:start_pos+bands_in_set].shape)
alpha = np.concatenate((alpha, matched_filter(b_img_data[:,columns:col_range,:], \
target_mean[start_pos:start_pos+bands_in_set], b_mean_cov_obj)), axis=1)
print("Shape of Alpha : ", alpha.shape)
b_img_data = None
del b_img_data
bands_set_end = start_pos + bands_in_set
processed_file = f'{mf_output_folder}/{sname}_{start_pos}_{end_pos}.npy'
print("Saving the alpha output", processed_file)
np.save(processed_file, alpha)
start_pos += bands_in_set
reset_pos += bands_in_set
if (start_pos > channel): break
if ((reset_pos % bands_in_set) == 0):
start_pos = start_pos - overlap
reset_pos = 0