import os
import numpy as np
from polsartools.utils.proc_utils import process_chunks_parallel
from polsartools.utils.utils import conv2d,time_it
from polsartools.utils.convert_matrices import C3_T3_mat
from .fp_infiles import fp_c3t3files
[docs]
@time_it
def tsvm(in_dir, win=1, fmt="tif", cog=False,
ovr = [2, 4, 8, 16], comp=False,
max_workers=None,block_size=(512, 512),
progress_callback=None, # for QGIS plugin
):
"""Perform Touzi Decomposition for full-pol SAR data.
Examples
--------
>>> # Basic usage with default parameters
>>> tsvm("/path/to/fullpol_data")
>>> # Advanced usage with custom parameters
>>> tsvm(
... in_dir="/path/to/fullpol_data",
... win=5,
... fmt="tif",
... cog=True,
... block_size=(1024, 1024)
... )
Parameters
----------
in_dir : str
Path to the input folder containing full-pol T3 or C3 matrix files.
win : int, default=1
Size of the spatial averaging window. Larger windows reduce speckle noise
but decrease spatial resolution.
fmt : {'tif', 'bin'}, default='tif'
Output file format:
- 'tif': GeoTIFF format with georeferencing information
- 'bin': Raw binary format
cog : bool, default=False
If True, creates Cloud Optimized GeoTIFF (COG) outputs with internal tiling
and overviews for efficient web access.
ovr : list[int], default=[2, 4, 8, 16]
Overview levels for COG creation. Each number represents the
decimation factor for that overview level.
comp : bool, default=False
If True, uses LZW compression for GeoTIFF outputs.
max_workers : int | None, default=None
Maximum number of parallel processing workers. If None, uses
CPU count - 1 workers.
block_size : tuple[int, int], default=(512, 512)
Size of processing blocks (rows, cols) for parallel computation.
Larger blocks use more memory but may be more efficient.
Returns
-------
None
Writes the follwing files to disk:
1. TSVM_alpha1
2. TSVM_alpha2
3. TSVM_alpha3
4. TSVM_phi1
5. TSVM_phi2
6. TSVM_phi3
7. TSVM_tau1
8. TSVM_tau2
9. TSVM_tau3
10. TSVM_psi1
11. TSVM_psi2
12. TSVM_psi3
13. TSVM_alphas
14. TSVM_phis
15. TSVM_taus
16. TSVM_psis
"""
write_flag=True
input_filepaths = fp_c3t3files(in_dir)
output_filepaths = []
if fmt == "bin":
output_filepaths.append(os.path.join(in_dir, "TSVM_alpha1.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_alpha2.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_alpha3.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_phi1.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_phi2.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_phi3.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_tau1.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_tau2.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_tau3.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_psi1.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_psi2.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_psi3.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_alphas.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_phis.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_taus.bin"))
output_filepaths.append(os.path.join(in_dir, "TSVM_psis.bin"))
else:
output_filepaths.append(os.path.join(in_dir, "TSVM_alpha1.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_alpha2.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_alpha3.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_phi1.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_phi2.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_phi3.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_tau1.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_tau2.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_tau3.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_psi1.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_psi2.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_psi3.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_alphas.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_phis.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_taus.tif"))
output_filepaths.append(os.path.join(in_dir, "TSVM_psis.tif"))
process_chunks_parallel(input_filepaths, list(output_filepaths),
window_size=win, write_flag=write_flag,
processing_func=process_chunk_tsvm,block_size=block_size,
max_workers=max_workers, num_outputs=len(output_filepaths),
cog=cog, ovr=ovr, comp=comp,
progress_callback=progress_callback
)
def process_chunk_tsvm(chunks, window_size, input_filepaths,*args):
if 'T11' in input_filepaths[0] and 'T22' in input_filepaths[5] and 'T33' in input_filepaths[8]:
t11_T1 = np.array(chunks[0])
t12_T1 = np.array(chunks[1])+1j*np.array(chunks[2])
t13_T1 = np.array(chunks[3])+1j*np.array(chunks[4])
t21_T1 = np.conj(t12_T1)
t22_T1 = np.array(chunks[5])
t23_T1 = np.array(chunks[6])+1j*np.array(chunks[7])
t31_T1 = np.conj(t13_T1)
t32_T1 = np.conj(t23_T1)
t33_T1 = np.array(chunks[8])
T_T1 = np.array([[t11_T1, t12_T1, t13_T1],
[t21_T1, t22_T1, t23_T1],
[t31_T1, t32_T1, t33_T1]])
if 'C11' in input_filepaths[0] and 'C22' in input_filepaths[5] and 'C33' in input_filepaths[8]:
C11 = np.array(chunks[0])
C12 = np.array(chunks[1])+1j*np.array(chunks[2])
C13 = np.array(chunks[3])+1j*np.array(chunks[4])
C21 = np.conj(C12)
C22 = np.array(chunks[5])
C23 = np.array(chunks[6])+1j*np.array(chunks[7])
C31 = np.conj(C13)
C32 = np.conj(C23)
C33 = np.array(chunks[8])
C3 = np.array([[C11, C12, C13],
[C21, C22, C23],
[C31, C32, C33]])
T_T1 = C3_T3_mat(C3)
if window_size>1:
kernel = np.ones((window_size,window_size),np.float32)/(window_size*window_size)
t11f = conv2d(T_T1[0,0,:,:],kernel)
t12f = conv2d(np.real(T_T1[0,1,:,:]),kernel)+1j*conv2d(np.imag(T_T1[0,1,:,:]),kernel)
t13f = conv2d(np.real(T_T1[0,2,:,:]),kernel)+1j*conv2d(np.imag(T_T1[0,2,:,:]),kernel)
t21f = np.conj(t12f)
t22f = conv2d(T_T1[1,1,:,:],kernel)
t23f = conv2d(np.real(T_T1[1,2,:,:]),kernel)+1j*conv2d(np.imag(T_T1[1,2,:,:]),kernel)
t31f = np.conj(t13f)
t32f = np.conj(t23f)
t33f = conv2d(T_T1[2,2,:,:],kernel)
T_T1 = np.array([[t11f, t12f, t13f], [t21f, t22f, t23f], [t31f, t32f, t33f]])
pi = np.pi
nx, ny = T_T1.shape[2], T_T1.shape[3]
N = nx * ny
eps = 1e-8
# Reshape M to (N, 3, 3) for batch processing
M_flat = T_T1.transpose(2, 3, 0, 1).reshape(N, 3, 3)
# Eigen decomposition
lambda_vals, V = np.linalg.eig(M_flat) # V: (N, 3, 3), lambda_vals: (N, 3)
# Sort eigenvalues and corresponding eigenvectors
idx = np.argsort(lambda_vals.real, axis=1)[:, ::-1] # descending order
lambda_sorted = np.take_along_axis(lambda_vals.real, idx, axis=1)
V_sorted = np.array([V[i][:, idx[i]] for i in range(N)]) # shape: (N, 3, 3)
# Clip negative eigenvalues
lambda_sorted = np.clip(lambda_sorted, 0, None)
# Normalize eigenvectors: remove global phase
phase = np.arctan2(V_sorted[:, 0, :].imag, eps + V_sorted[:, 0, :].real)
V_sorted *= np.exp(-1j * phase[:, np.newaxis, :])
# Compute psi
psi = 0.5 * np.arctan2(V_sorted[:, 2, :].real, eps + V_sorted[:, 1, :].real)
# Rotate V[1] and V[2] by 2*psi manually
cos2psi = np.cos(2 * psi)
sin2psi = np.sin(2 * psi)
V1r, V1i = V_sorted[:, 1, :].real, V_sorted[:, 1, :].imag
V2r, V2i = V_sorted[:, 2, :].real, V_sorted[:, 2, :].imag
V1r_new = V1r * cos2psi + V2r * sin2psi
V1i_new = V1i * cos2psi + V2i * sin2psi
V2r_new = -V1r * sin2psi + V2r * cos2psi
V2i_new = -V1i * sin2psi + V2i * cos2psi
V_sorted[:, 1, :] = V1r_new + 1j * V1i_new
V_sorted[:, 2, :] = V2r_new + 1j * V2i_new
# Compute tau
tau = 0.5 * np.arctan2(-V_sorted[:, 2, :].imag, eps + V_sorted[:, 0, :].real)
# Compute phi
phi = np.arctan2(V_sorted[:, 1, :].imag, eps + V_sorted[:, 1, :].real)
# Rotate V[0] and V[2] by 2*tau manually
cos2tau = np.cos(2 * tau)
sin2tau = np.sin(2 * tau)
V0r, V0i = V_sorted[:, 0, :].real, V_sorted[:, 0, :].imag
V2r, V2i = V_sorted[:, 2, :].real, V_sorted[:, 2, :].imag
V0r_new = V0r * cos2tau - V2i * sin2tau
V0i_new = V0i * cos2tau + V2r * sin2tau
V2r_new = -V0i * sin2tau + V2r * cos2tau
V2i_new = V0r * sin2tau + V2i * cos2tau
V_sorted[:, 0, :] = V0r_new + 1j * V0i_new
V_sorted[:, 2, :] = V2r_new + 1j * V2i_new
# Compute alpha
alpha = np.arccos(V_sorted[:, 0, :].real)
# Flip signs if psi out of bounds
flip_mask = (psi < -pi / 4) | (psi > pi / 4)
tau[flip_mask] *= -1
phi[flip_mask] *= -1
# Compute probabilities
total_lambda = eps + np.sum(lambda_sorted, axis=1, keepdims=True)
p = np.clip(lambda_sorted / total_lambda, 0, 1)
# Weighted mean angles
alpha_mean = np.sum(alpha * p, axis=1)
phi_mean = np.sum(phi * p, axis=1)
tau_mean = np.sum(tau * p, axis=1)
psi_mean = np.sum(psi * p, axis=1)
# Reshape outputs back to (3, nx, ny) and (nx, ny)
shape = (nx, ny)
to_deg = 180. / pi
alphas = (alpha.T.reshape((3,) + shape) * to_deg).astype(np.float32)
phis = (phi.T.reshape((3,) + shape) * to_deg).astype(np.float32)
taus = (tau.T.reshape((3,) + shape) * to_deg).astype(np.float32)
psis = (psi.T.reshape((3,) + shape) * to_deg).astype(np.float32)
alpha_mean = (alpha_mean.reshape(shape) * to_deg).astype(np.float32)
phi_mean = (phi_mean.reshape(shape) * to_deg).astype(np.float32)
tau_mean = (tau_mean.reshape(shape) * to_deg).astype(np.float32)
psi_mean = (psi_mean.reshape(shape) * to_deg).astype(np.float32)
return alphas[0], alphas[1], alphas[2], \
phis[0], phis[1], phis[2], \
taus[0], taus[1], taus[2], \
psis[0], psis[1], psis[2], \
alpha_mean, phi_mean, tau_mean, psi_mean
# return {
# "alpha": alpha.T.reshape((3,) + shape) * to_deg,
# "phi": phi.T.reshape((3,) + shape) * to_deg,
# "tau": tau.T.reshape((3,) + shape) * to_deg,
# "psi": psi.T.reshape((3,) + shape) * to_deg,
# "alpha_mean": alpha_mean.reshape(shape) * to_deg,
# "phi_mean": phi_mean.reshape(shape) * to_deg,
# "tau_mean": tau_mean.reshape(shape) * to_deg,
# "psi_mean": psi_mean.reshape(shape) * to_deg,
# "p": p.T.reshape((3,) + shape)
# }