| 1 |
""" |
| 2 |
This code is part of the Python PolSARpro software: |
| 3 |
|
| 4 |
"A re-implementation of selected PolSARPro functions in Python, |
| 5 |
following the scientific recommendations of PolInSAR 2021" |
| 6 |
|
| 7 |
developed within an ESA funded project with SATIM. |
| 8 |
|
| 9 |
Author: Olivier D'Hondt, 2025. |
| 10 |
Scientific advisors: Armando Marino and Eric Pottier. |
| 11 |
|
| 12 |
Licensed under the Apache License, Version 2.0 (the "License"); |
| 13 |
you may not use this file except in compliance with the License. |
| 14 |
You may obtain a copy of the License at |
| 15 |
|
| 16 |
http://www.apache.org/licenses/LICENSE-2.0 |
| 17 |
|
| 18 |
Unless required by applicable law or agreed to in writing, software |
| 19 |
distributed under the License is distributed on an "AS IS" BASIS, |
| 20 |
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 21 |
See the License for the specific language governing permissions and |
| 22 |
limitations under the License. |
| 23 |
|
| 24 |
----- |
| 25 |
|
| 26 |
# Description: module containing various utility functions |
| 27 |
|
| 28 |
""" |
| 29 |
|
| 30 |
import logging |
| 31 |
|
| 32 |
import dask.array as da |
| 33 |
import matplotlib.pyplot as plt |
| 34 |
import numpy as np |
| 35 |
import xarray |
| 36 |
import xarray as xr |
| 37 |
from scipy.ndimage import convolve |
| 38 |
|
| 39 |
from polsarpro.auxil import validate_dataset |
| 40 |
|
| 41 |
log = logging.getLogger(__name__) |
| 42 |
|
| 43 |
|
| 44 |
def S_to_C2(S: xarray.Dataset, p1: str = "hh", p2: str = "hv") -> xarray.Dataset: |
| 45 |
"""Converts the Sinclair scattering matrix S to the lexicographic dual polarization covariance matrix C2. |
| 46 |
|
| 47 |
Args: |
| 48 |
S (xarray.Dataset): input image of scattering matrices |
| 49 |
p1 (str): first polarization. |
| 50 |
p2 (str): second polarization. |
| 51 |
|
| 52 |
Returns: |
| 53 |
xarray.Dataset: C2 covariance matrix |
| 54 |
Note: |
| 55 |
p1 and p2 must be different and belong to {'hh', 'hv', 'vv', 'vh'} |
| 56 |
""" |
| 57 |
|
| 58 |
_ = validate_dataset(S, allowed_poltypes="S") |
| 59 |
|
| 60 |
pols = {"hh", "hv", "vv", "vh"} |
| 61 |
if not {p1, p2}.issubset(pols): |
| 62 |
raise ValueError(f"Polarizations need to be in {pols}") |
| 63 |
|
| 64 |
if p1 == p2: |
| 65 |
raise ValueError(f"Polarizations must be different.") |
| 66 |
|
| 67 |
# scattering vector, enforce type as in C version |
| 68 |
k1 = S[p1].astype("complex64", copy=False) |
| 69 |
k2 = S[p2].astype("complex64", copy=False) |
| 70 |
|
| 71 |
# compute the Hermitian matrix elements |
| 72 |
C2 = {} |
| 73 |
|
| 74 |
# force real diagonal to save space |
| 75 |
C2["m11"] = (k1 * k1.conj()).real |
| 76 |
C2["m22"] = (k2 * k2.conj()).real |
| 77 |
|
| 78 |
# upper diagonal terms |
| 79 |
C2["m12"] = k1 * k2.conj() |
| 80 |
|
| 81 |
attrs = { |
| 82 |
"poltype": "C2", |
| 83 |
"description": f"Covariance matrix (2x2), with polarizations {p1} and {p2}", |
| 84 |
} |
| 85 |
return xr.Dataset(C2, attrs=attrs) |
| 86 |
|
| 87 |
|
| 88 |
def S_to_C3(S: xarray.Dataset) -> xarray.Dataset: |
| 89 |
"""Converts the Sinclair scattering matrix S to the lexicographic covariance matrix C3. |
| 90 |
|
| 91 |
Args: |
| 92 |
S (xarray.Dataset): input image of scattering matrices |
| 93 |
|
| 94 |
Returns: |
| 95 |
xarray.Dataset: C3 covariance matrix |
| 96 |
""" |
| 97 |
_ = validate_dataset(S, allowed_poltypes="S") |
| 98 |
|
| 99 |
# scattering vector, enforce type as in C version |
| 100 |
c = np.sqrt(np.float32(2)) |
| 101 |
k1 = S.hh.astype("complex64", copy=False) |
| 102 |
k2 = ((S.hv + S.vh) / c).astype("complex64", copy=False) |
| 103 |
k3 = S.vv.astype("complex64", copy=False) |
| 104 |
|
| 105 |
# compute the Hermitian matrix elements |
| 106 |
C3 = {} |
| 107 |
|
| 108 |
# force real diagonal to save space |
| 109 |
C3["m11"] = (k1 * k1.conj()).real |
| 110 |
C3["m22"] = (k2 * k2.conj()).real |
| 111 |
C3["m33"] = (k3 * k3.conj()).real |
| 112 |
|
| 113 |
# upper diagonal terms |
| 114 |
C3["m12"] = k1 * k2.conj() |
| 115 |
C3["m13"] = k1 * k3.conj() |
| 116 |
C3["m23"] = k2 * k3.conj() |
| 117 |
|
| 118 |
attrs = {"poltype": "C3", "description": "Covariance matrix (3x3)"} |
| 119 |
return xr.Dataset(C3, attrs=attrs) |
| 120 |
|
| 121 |
|
| 122 |
def S_to_C4(S: xarray.Dataset) -> xarray.Dataset: |
| 123 |
"""Converts the Sinclair scattering matrix S to the lexicographic covariance matrix C4. |
| 124 |
|
| 125 |
Args: |
| 126 |
S (xarray.Dataset): input image of scattering matrices |
| 127 |
|
| 128 |
Returns: |
| 129 |
xarray.Dataset: C4 covariance matrix |
| 130 |
""" |
| 131 |
_ = validate_dataset(S, allowed_poltypes="S") |
| 132 |
|
| 133 |
# scattering vector, enforce type as in C version |
| 134 |
k1 = S.hh.astype("complex64", copy=False) |
| 135 |
k2 = S.hv.astype("complex64", copy=False) |
| 136 |
k3 = S.vh.astype("complex64", copy=False) |
| 137 |
k4 = S.vv.astype("complex64", copy=False) |
| 138 |
|
| 139 |
# compute the Hermitian matrix elements |
| 140 |
C4 = {} |
| 141 |
# force real diagonal to save space |
| 142 |
C4["m11"] = (k1 * k1.conj()).real |
| 143 |
C4["m22"] = (k2 * k2.conj()).real |
| 144 |
C4["m33"] = (k3 * k3.conj()).real |
| 145 |
C4["m44"] = (k4 * k4.conj()).real |
| 146 |
|
| 147 |
# upper diagonal terms |
| 148 |
C4["m12"] = k1 * k2.conj() |
| 149 |
C4["m13"] = k1 * k3.conj() |
| 150 |
C4["m14"] = k1 * k4.conj() |
| 151 |
C4["m23"] = k2 * k3.conj() |
| 152 |
C4["m24"] = k2 * k4.conj() |
| 153 |
C4["m34"] = k3 * k4.conj() |
| 154 |
|
| 155 |
attrs = {"poltype": "C4", "description": "Covariance matrix (4x4)"} |
| 156 |
return xr.Dataset(C4, attrs=attrs) |
| 157 |
|
| 158 |
|
| 159 |
def S_to_T3(S: xarray.Dataset) -> xarray.Dataset: |
| 160 |
"""Converts the Sinclair scattering matrix S to the Pauli coherency matrix T3. |
| 161 |
|
| 162 |
Args: |
| 163 |
S (xarray.Dataset): input image of scattering matrices |
| 164 |
|
| 165 |
Returns: |
| 166 |
xarray.Dataset: T3 covariance matrix |
| 167 |
""" |
| 168 |
_ = validate_dataset(S, allowed_poltypes="S") |
| 169 |
|
| 170 |
# scattering vector |
| 171 |
c = np.sqrt(np.float32(2)) |
| 172 |
k1 = ((S.hh + S.vv) / c).astype("complex64", copy=False) |
| 173 |
k2 = ((S.hh - S.vv) / c).astype("complex64", copy=False) |
| 174 |
k3 = ((S.hv + S.vh) / c).astype("complex64", copy=False) |
| 175 |
|
| 176 |
# compute the Hermitian matrix elements |
| 177 |
T3 = {} |
| 178 |
|
| 179 |
# force real diagonal to save space |
| 180 |
T3["m11"] = (k1 * k1.conj()).real |
| 181 |
T3["m22"] = (k2 * k2.conj()).real |
| 182 |
T3["m33"] = (k3 * k3.conj()).real |
| 183 |
|
| 184 |
# upper diagonal terms |
| 185 |
T3["m12"] = k1 * k2.conj() |
| 186 |
T3["m13"] = k1 * k3.conj() |
| 187 |
T3["m23"] = k2 * k3.conj() |
| 188 |
|
| 189 |
attrs = {"poltype": "T3", "description": "Coherency matrix (3x3)"} |
| 190 |
return xr.Dataset(T3, attrs=attrs) |
| 191 |
|
| 192 |
|
| 193 |
def S_to_T4(S: xarray.Dataset) -> xarray.Dataset: |
| 194 |
"""Converts the Sinclair scattering matrix S to the Pauli coherency matrix T4. |
| 195 |
|
| 196 |
Args: |
| 197 |
S (xarray.Dataset): input image of scattering matrices |
| 198 |
|
| 199 |
Returns: |
| 200 |
xarray.Dataset: T4 coherency matrix |
| 201 |
""" |
| 202 |
_ = validate_dataset(S, allowed_poltypes="S") |
| 203 |
|
| 204 |
# scattering vector |
| 205 |
c = np.sqrt(np.float32(2)) |
| 206 |
k1 = ((S.hh + S.vv) / c).astype("complex64", copy=False) |
| 207 |
k2 = ((S.hh - S.vv) / c).astype("complex64", copy=False) |
| 208 |
k3 = ((S.hv + S.vh) / c).astype("complex64", copy=False) |
| 209 |
# NOTE: it might work only with ALOS-1 (hv-vh is inverted, according to C-PSP) |
| 210 |
k4 = (1j * (S.vh - S.hv) / c).astype("complex64", copy=False) |
| 211 |
# alternate version |
| 212 |
# k4 = (1j * (S.hv - S.vh) / c).astype("complex64", copy=False) |
| 213 |
|
| 214 |
# compute the Hermitian matrix elements |
| 215 |
T4 = {} |
| 216 |
|
| 217 |
# force real diagonal to save space |
| 218 |
T4["m11"] = (k1 * k1.conj()).real |
| 219 |
T4["m22"] = (k2 * k2.conj()).real |
| 220 |
T4["m33"] = (k3 * k3.conj()).real |
| 221 |
T4["m44"] = (k4 * k4.conj()).real |
| 222 |
|
| 223 |
# upper diagonal terms |
| 224 |
T4["m12"] = k1 * k2.conj() |
| 225 |
T4["m13"] = k1 * k3.conj() |
| 226 |
T4["m14"] = k1 * k4.conj() |
| 227 |
T4["m23"] = k2 * k3.conj() |
| 228 |
T4["m24"] = k2 * k4.conj() |
| 229 |
T4["m34"] = k3 * k4.conj() |
| 230 |
|
| 231 |
attrs = {"poltype": "T4", "description": "Coherency matrix (4x4)"} |
| 232 |
return xr.Dataset(T4, attrs=attrs) |
| 233 |
|
| 234 |
|
| 235 |
def C3_to_T3(C3: xarray.Dataset) -> xarray.Dataset: |
| 236 |
"""Converts the lexicographic covariance matrix C3 to the Pauli coherency matrix T3. |
| 237 |
|
| 238 |
Args: |
| 239 |
C3 (xarray.Dataset): input image of covariance matrices |
| 240 |
|
| 241 |
Returns: |
| 242 |
xarray.Dataset: T3 coherency matrix |
| 243 |
""" |
| 244 |
_ = validate_dataset(C3, allowed_poltypes="C3") |
| 245 |
|
| 246 |
T3 = {} |
| 247 |
|
| 248 |
c = 1 / np.sqrt(np.float32(2)) |
| 249 |
|
| 250 |
# force real diagonal to save space |
| 251 |
T3["m11"] = 0.5 * (C3.m11 + C3.m33) + C3.m13.real |
| 252 |
T3["m22"] = 0.5 * (C3.m11 + C3.m33) - C3.m13.real |
| 253 |
T3["m33"] = C3.m22 |
| 254 |
# upper diagonal terms |
| 255 |
T3["m12"] = 0.5 * (C3.m11 - C3.m33) - 1j * C3.m13.imag |
| 256 |
T3["m13"] = c * (C3.m12 + C3.m23.conj()) |
| 257 |
T3["m23"] = c * (C3.m12 - C3.m23.conj()) |
| 258 |
|
| 259 |
attrs = {"poltype": "T3", "description": "Coherency matrix (3x3)"} |
| 260 |
return xr.Dataset(T3, attrs=attrs) |
| 261 |
|
| 262 |
|
| 263 |
def T3_to_C3(T3: xarray.Dataset) -> xarray.Dataset: |
| 264 |
"""Converts the Pauli coherency matrix T3 to the lexicographic covariance matrix C3. |
| 265 |
|
| 266 |
Args: |
| 267 |
T3 (xarray.Dataset): input image of coherency matrices |
| 268 |
|
| 269 |
Returns: |
| 270 |
xarray.Dataset: C3 covariance matrix |
| 271 |
""" |
| 272 |
|
| 273 |
_ = validate_dataset(T3, allowed_poltypes="T3") |
| 274 |
|
| 275 |
C3 = {} |
| 276 |
|
| 277 |
c = 1 / np.sqrt(np.float32(2)) |
| 278 |
|
| 279 |
# force real diagonal to save space |
| 280 |
C3["m11"] = 0.5 * (T3.m11 + T3.m22) + T3.m12.real |
| 281 |
C3["m22"] = T3.m33 |
| 282 |
C3["m33"] = 0.5 * (T3.m11 + T3.m22) - T3.m12.real |
| 283 |
|
| 284 |
# upper diagonal terms |
| 285 |
C3["m12"] = c * (T3.m13 + T3.m23) |
| 286 |
C3["m13"] = 0.5 * (T3.m11.real - T3.m22.real) - 1j * T3.m12.imag |
| 287 |
C3["m23"] = c * (T3.m13.conj() - T3.m23.conj()) |
| 288 |
|
| 289 |
attrs = {"poltype": "C3", "description": "Covariance matrix (3x3)"} |
| 290 |
return xr.Dataset(C3, attrs=attrs) |
| 291 |
|
| 292 |
|
| 293 |
def C4_to_T4(C4: xarray.Dataset) -> xarray.Dataset: |
| 294 |
"""Converts the lexicographic covariance matrix C4 to the Pauli coherency matrix T4. |
| 295 |
|
| 296 |
Args: |
| 297 |
C4 (xarray.Dataset): input image of covariance matrices |
| 298 |
|
| 299 |
Returns: |
| 300 |
xarray.Dataset: T4 coherency matrix |
| 301 |
""" |
| 302 |
_ = validate_dataset(C4, allowed_poltypes="C4") |
| 303 |
|
| 304 |
T4 = {} |
| 305 |
|
| 306 |
# force real diagonal to save space |
| 307 |
# diagonal terms |
| 308 |
T4["m11"] = 0.5 * (C4.m11 + 2 * C4.m14.real + C4.m44) |
| 309 |
T4["m22"] = 0.5 * (C4.m11 - 2 * C4.m14.real + C4.m44) |
| 310 |
T4["m33"] = 0.5 * (C4.m22 + C4.m33 + 2 * C4.m23.real) |
| 311 |
T4["m44"] = 0.5 * (C4.m22 + C4.m33 - 2 * C4.m23.real) |
| 312 |
|
| 313 |
# m12 |
| 314 |
T4["m12"] = 0.5 * (C4.m11 - C4.m44) - 1j * C4.m14.imag |
| 315 |
|
| 316 |
# m13 |
| 317 |
T4["m13"] = 0.5 * (C4.m12 + C4.m13 + C4.m24.conj() + C4.m34.conj()) |
| 318 |
|
| 319 |
# m14 |
| 320 |
T4["m14"] = 0.5 * ( |
| 321 |
C4.m12.imag - C4.m13.imag - C4.m24.imag + C4.m34.imag |
| 322 |
) + 0.5 * 1j * (-C4.m12.real + C4.m13.real - C4.m24.real + C4.m34.real) |
| 323 |
|
| 324 |
# m23 |
| 325 |
T4["m23"] = 0.5 * (C4.m12 + C4.m13 - C4.m24.conj() - C4.m34.conj()) |
| 326 |
|
| 327 |
# m24 |
| 328 |
T4["m24"] = 0.5 * ( |
| 329 |
C4.m12.imag - C4.m13.imag + C4.m24.imag - C4.m34.imag |
| 330 |
) + 0.5 * 1j * (-C4.m12.real + C4.m13.real + C4.m24.real - C4.m34.real) |
| 331 |
|
| 332 |
# m34 |
| 333 |
T4["m34"] = -C4.m23.imag + 0.5j * (-C4.m22 + C4.m33) |
| 334 |
|
| 335 |
attrs = {"poltype": "T4", "description": "Coherency matrix (4x4)"} |
| 336 |
return xr.Dataset(T4, attrs=attrs) |
| 337 |
|
| 338 |
|
| 339 |
def C4_to_C3(C4: xarray.Dataset) -> xarray.Dataset: |
| 340 |
"""Converts the lexicographic covariance matrix C4 to the lexicographic covariance matrix C3. |
| 341 |
|
| 342 |
Args: |
| 343 |
C4 (xarray.Dataset): input image of covariance matrices |
| 344 |
|
| 345 |
Returns: |
| 346 |
xarray.Dataset: C3 covariance matrix |
| 347 |
""" |
| 348 |
_ = validate_dataset(C4, allowed_poltypes="C4") |
| 349 |
|
| 350 |
C3 = {} |
| 351 |
|
| 352 |
c = 1 / np.sqrt(np.float32(2)) |
| 353 |
|
| 354 |
# force real diagonal to save space |
| 355 |
C3["m11"] = C4.m11 |
| 356 |
C3["m22"] = 0.5 * (C4.m22 + C4.m33 + 2 * C4.m23.real) |
| 357 |
C3["m33"] = C4.m44 |
| 358 |
|
| 359 |
# upper diagonal terms |
| 360 |
C3["m12"] = c * (C4.m12 + C4.m13) |
| 361 |
C3["m13"] = C4.m14 |
| 362 |
C3["m23"] = c * (C4.m24 + C4.m34) |
| 363 |
|
| 364 |
attrs = {"poltype": "C3", "description": "Covariance matrix (3x3)"} |
| 365 |
return xr.Dataset(C3, attrs=attrs) |
| 366 |
|
| 367 |
|
| 368 |
def C4_to_T3(C4: xarray.Dataset) -> xarray.Dataset: |
| 369 |
"""Converts the lexicographic covariance matrix C4 to the Pauli coherency matrix T3. |
| 370 |
|
| 371 |
Args: |
| 372 |
C4 (xarray.Dataset): input image of covariance matrices |
| 373 |
|
| 374 |
Returns: |
| 375 |
xarray.Dataset: T3 coherency matrix |
| 376 |
""" |
| 377 |
_ = validate_dataset(C4, allowed_poltypes="C4") |
| 378 |
|
| 379 |
T3 = {} |
| 380 |
|
| 381 |
# force real diagonal to save space |
| 382 |
T3["m11"] = 0.5 * (C4.m11 + C4.m44) + C4.m14.real |
| 383 |
T3["m22"] = 0.5 * (C4.m11 + C4.m44) - C4.m14.real |
| 384 |
T3["m33"] = 0.5 * (C4.m22 + C4.m33 + 2 * C4.m23.real) |
| 385 |
|
| 386 |
# upper diagonal terms |
| 387 |
T3["m12"] = 0.5 * (C4.m11 - C4.m44) - 1j * C4.m14.imag |
| 388 |
T3["m13"] = 0.5 * (C4.m12 + C4.m13 + C4.m24.conj() + C4.m34.conj()) |
| 389 |
T3["m23"] = 0.5 * (C4.m12 + C4.m13 - C4.m24.conj() - C4.m34.conj()) |
| 390 |
|
| 391 |
attrs = {"poltype": "T3", "description": "Coherency matrix (3x3)"} |
| 392 |
return xr.Dataset(T3, attrs=attrs) |
| 393 |
|
| 394 |
|
| 395 |
def T4_to_C4(T4: xarray.Dataset) -> xarray.Dataset: |
| 396 |
"""Converts the Pauli coherency matrix T4 to the lexicographic covariance matrix C4. |
| 397 |
|
| 398 |
Args: |
| 399 |
T4 (xarray.Dataset): input image of coherency matrices |
| 400 |
|
| 401 |
Returns: |
| 402 |
xarray.Dataset: C4 covariance matrix |
| 403 |
|
| 404 |
""" |
| 405 |
_ = validate_dataset(T4, allowed_poltypes="T4") |
| 406 |
|
| 407 |
C4 = {} |
| 408 |
|
| 409 |
# force real diagonal to save space |
| 410 |
C4["m11"] = 0.5 * (T4.m11 + 2 * T4.m12.real + T4.m22) |
| 411 |
C4["m22"] = 0.5 * (T4.m33 - 2 * T4.m34.imag + T4.m44) |
| 412 |
C4["m33"] = 0.5 * (T4.m33 + T4.m44 + 2 * T4.m34.imag) |
| 413 |
C4["m44"] = 0.5 * (T4.m11 - 2 * T4.m12.real + T4.m22) |
| 414 |
|
| 415 |
# upper diagonal terms |
| 416 |
C4["m12"] = 0.5 * ( |
| 417 |
T4.m13.real + T4.m23.real - T4.m14.imag - T4.m24.imag |
| 418 |
) + 0.5 * 1j * (T4.m13.imag + T4.m23.imag + T4.m14.real + T4.m24.real) |
| 419 |
|
| 420 |
C4["m13"] = 0.5 * ( |
| 421 |
T4.m13.real + T4.m23.real + T4.m14.imag + T4.m24.imag |
| 422 |
) + 0.5 * 1j * (T4.m13.imag + T4.m23.imag - T4.m14.real - T4.m24.real) |
| 423 |
|
| 424 |
C4["m14"] = 0.5 * (T4.m11 - T4.m22) - 1j * T4.m12.imag |
| 425 |
|
| 426 |
C4["m23"] = 0.5 * (T4.m33 - T4.m44) - 1j * T4.m34.real |
| 427 |
|
| 428 |
C4["m24"] = 0.5 * ( |
| 429 |
T4.m13.real - T4.m23.real - T4.m14.imag + T4.m24.imag |
| 430 |
) + 0.5 * 1j * (-T4.m13.imag + T4.m23.imag - T4.m14.real + T4.m24.real) |
| 431 |
|
| 432 |
C4["m34"] = 0.5 * ( |
| 433 |
T4.m13.real - T4.m23.real + T4.m14.imag - T4.m24.imag |
| 434 |
) + 0.5 * 1j * (-T4.m13.imag + T4.m23.imag + T4.m14.real - T4.m24.real) |
| 435 |
|
| 436 |
attrs = {"poltype": "C4", "description": "Covariance matrix (4x4)"} |
| 437 |
return xr.Dataset(C4, attrs=attrs) |
| 438 |
|
| 439 |
|
| 440 |
def T4_to_C3(T4: xarray.Dataset) -> xarray.Dataset: |
| 441 |
"""Converts the Pauli coherency matrix T4 to the lexicographic covariance matrix C3. |
| 442 |
|
| 443 |
Args: |
| 444 |
T4 (xarray.Dataset): input image of coherency matrices |
| 445 |
|
| 446 |
Returns: |
| 447 |
xarray.Dataset: C3 covariance matrix |
| 448 |
|
| 449 |
""" |
| 450 |
_ = validate_dataset(T4, allowed_poltypes="T4") |
| 451 |
|
| 452 |
C3 = {} |
| 453 |
|
| 454 |
c = 1 / np.sqrt(np.float32(2)) |
| 455 |
|
| 456 |
# force real diagonal to save space |
| 457 |
C3["m11"] = 0.5 * (T4.m11 + 2 * T4.m12.real + T4.m22) |
| 458 |
C3["m22"] = T4.m33 |
| 459 |
C3["m33"] = 0.5 * (T4.m11 - 2 * T4.m12.real + T4.m22) |
| 460 |
|
| 461 |
# upper diagonal terms |
| 462 |
C3["m12"] = c * (T4.m13 + T4.m23) |
| 463 |
C3["m13"] = 0.5 * (T4.m11 - T4.m22) - 1j * T4.m12.imag |
| 464 |
C3["m23"] = c * (T4.m13.conj() - T4.m23.conj()) |
| 465 |
|
| 466 |
attrs = {"poltype": "C3", "description": "Covariance matrix (3x3)"} |
| 467 |
return xr.Dataset(C3, attrs=attrs) |
| 468 |
|
| 469 |
|
| 470 |
def T4_to_T3(T4: xarray.Dataset) -> xarray.Dataset: |
| 471 |
"""Converts the Pauli coherency matrix T4 to the Pauli coherency matrix T3. |
| 472 |
|
| 473 |
Args: |
| 474 |
T4 (xarray.Dataset): input image of coherency matrices |
| 475 |
|
| 476 |
Returns: |
| 477 |
xarray.Dataset: T3 coherency matrix |
| 478 |
|
| 479 |
""" |
| 480 |
_ = validate_dataset(T4, allowed_poltypes="T4") |
| 481 |
|
| 482 |
T3 = { |
| 483 |
"m11": T4.m11, |
| 484 |
"m22": T4.m22, |
| 485 |
"m33": T4.m33, |
| 486 |
"m12": T4.m12, |
| 487 |
"m13": T4.m13, |
| 488 |
"m23": T4.m23, |
| 489 |
} |
| 490 |
|
| 491 |
attrs = {"poltype": "T3", "description": "Coherency matrix (3x3)"} |
| 492 |
return xr.Dataset(T3, attrs=attrs) |
| 493 |
|
| 494 |
|
| 495 |
# this function is used only in unit tests |
| 496 |
def vec_to_mat(vec: np.ndarray) -> np.ndarray: |
| 497 |
if vec.ndim != 3: |
| 498 |
raise ValueError("Vector valued image is expected (dimension 3)") |
| 499 |
return vec[:, :, None, :] * vec[:, :, :, None].conj() |
| 500 |
|
| 501 |
|
| 502 |
def boxcar(img: xarray.Dataset, dim_az: int, dim_rg: int) -> xarray.Dataset: |
| 503 |
""" |
| 504 |
Apply a boxcar filter to an image. |
| 505 |
|
| 506 |
Args: |
| 507 |
img (xarray.Dataset): Input image with variables of shape (naz, nrg, ...). |
| 508 |
dim_az (int): Size in azimuth of the filter. |
| 509 |
dim_rg (int): Size in range of the filter. |
| 510 |
|
| 511 |
Returns: |
| 512 |
xarray.Dataset: Filtered image, shape (naz, nrg, ...). |
| 513 |
|
| 514 |
Note: |
| 515 |
The filter is always applied along 2 dimensions (azimuth, range). Please ensure to provide a valid image. |
| 516 |
""" |
| 517 |
_ = validate_dataset(img) |
| 518 |
|
| 519 |
if type(dim_az) != int and type(dim_rg) != int: |
| 520 |
raise ValueError("dimaz and dimrg must be integers") |
| 521 |
if (dim_az < 1) or (dim_rg < 1): |
| 522 |
raise ValueError("dimaz and dimrg must be strictly positive") |
| 523 |
|
| 524 |
process_args = dict( |
| 525 |
dim_az=dim_az, |
| 526 |
dim_rg=dim_rg, |
| 527 |
depth=(dim_az, dim_rg), |
| 528 |
) |
| 529 |
data_out = {} |
| 530 |
for var in img.data_vars: |
| 531 |
# if data is not chunked, apply directly |
| 532 |
if isinstance(img[var].data, np.ndarray): |
| 533 |
da_in = _boxcar_core(img[var].data, dim_az=dim_az, dim_rg=dim_rg) |
| 534 |
# for dask chunked arrays, apply blockwise parallel processing |
| 535 |
else: |
| 536 |
da_in = da.map_overlap( |
| 537 |
_boxcar_core, |
| 538 |
img[var].data, |
| 539 |
**process_args, |
| 540 |
dtype=img[var].dtype, |
| 541 |
) |
| 542 |
data_out[var] = (img[var].dims, da_in) |
| 543 |
|
| 544 |
return xr.Dataset(data_out, coords=img.coords, attrs=img.attrs) |
| 545 |
|
| 546 |
|
| 547 |
# this should not be used directly |
| 548 |
def _boxcar_core(img: np.ndarray, dim_az: int, dim_rg: int) -> np.ndarray: |
| 549 |
n_extra_dims = img.ndim - 2 |
| 550 |
|
| 551 |
ker_dtype = img.dtype if not np.iscomplexobj(img) else img.real.dtype |
| 552 |
|
| 553 |
# this convolution mode reduces error between C and python implementations |
| 554 |
mode = "constant" |
| 555 |
if (dim_az > 1) or (dim_rg > 1): |
| 556 |
# avoid nan propagation |
| 557 |
msk = np.isnan(img) |
| 558 |
img_ = img.copy() |
| 559 |
|
| 560 |
img_[msk] = 0 |
| 561 |
ker = np.ones((dim_az, dim_rg), dtype=ker_dtype) / (dim_az * dim_rg) |
| 562 |
ker = np.expand_dims(ker, axis=tuple(range(2, 2 + n_extra_dims))) |
| 563 |
if np.iscomplexobj(img_): |
| 564 |
imgout = convolve(img_.real, ker, mode=mode) + 1j * convolve( |
| 565 |
img_.imag, ker, mode=mode |
| 566 |
) |
| 567 |
imgout[msk] = np.nan + 1j * np.nan |
| 568 |
else: |
| 569 |
imgout = convolve(img_, ker, mode=mode) |
| 570 |
imgout[msk] = np.nan |
| 571 |
return imgout |
| 572 |
else: |
| 573 |
return img |
| 574 |
|
| 575 |
|
| 576 |
# This is an alternative version that uses xarray's rolling mean |
| 577 |
# performance is not clear so I just keep it here for now |
| 578 |
# def boxcar(img: xarray.Dataset, dim_az: int, dim_rg: int) -> xarray.Dataset: |
| 579 |
# """ |
| 580 |
# Apply a boxcar filter to an image. |
| 581 |
|
| 582 |
# Args: |
| 583 |
# img (complex or real array): Input image with arbitrary number of dimensions, shape (naz, nrg, ...). |
| 584 |
# dim_az (int): Size in azimuth (or latitude) of the filter. |
| 585 |
# dim_rg (int): Size in range (or longitude) of the filter. |
| 586 |
|
| 587 |
# Returns: |
| 588 |
# complex or real array: Filtered image, shape (naz, nrg, ...). |
| 589 |
|
| 590 |
# Note: |
| 591 |
# The filter is always applied along 2 dimensions (azimuth, range). |
| 592 |
# If the input is a geocoded image, azimuth and range become latitude and longitude. |
| 593 |
# Please ensure to provide a valid image. |
| 594 |
# """ |
| 595 |
|
| 596 |
# if not isinstance(img, xarray.Dataset): |
| 597 |
# raise TypeError("Input must be a valid PolSARPro Dataset.") |
| 598 |
|
| 599 |
# if not all((isinstance(dim_az, int), isinstance(dim_rg, int))): |
| 600 |
# raise TypeError("Parameters dim_az and dim_rg must be integers.") |
| 601 |
|
| 602 |
# if dim_az <= 0 or dim_rg <= 0: |
| 603 |
# raise ValueError("Parameters dim_az and dim_rg must strictly positive.") |
| 604 |
|
| 605 |
# dims_sar = {"y", "x"} |
| 606 |
# dims_geo = {"lat", "lon"} |
| 607 |
|
| 608 |
# if dims_sar.issubset(img.dims): |
| 609 |
# dict_filter = dict(x=dim_rg, y=dim_az) |
| 610 |
# dict_slice = dict(x=slice(dim_rg, -dim_rg), y=slice(dim_az, -dim_az)) |
| 611 |
# elif dims_geo.issubset(img.dims): |
| 612 |
# dict_filter = dict(lon=dim_rg, lat=dim_az) |
| 613 |
# dict_slice = dict(lon=slice(dim_rg, -dim_rg), lat=slice(dim_az, -dim_az)) |
| 614 |
# else: |
| 615 |
# raise ValueError("Input data must have dimensions ('y','x') or ('lat', 'lon').") |
| 616 |
|
| 617 |
# # pad the data with zeros |
| 618 |
# res = img.pad(dict_filter, mode="constant", constant_values=0) |
| 619 |
|
| 620 |
# # compute rolling mean |
| 621 |
# res = res.rolling(**dict_filter, center=True).mean() |
| 622 |
|
| 623 |
# # trim the padded borders |
| 624 |
# res = res.isel(**dict_slice) |
| 625 |
# return res |
| 626 |
|
| 627 |
|
| 628 |
def multilook(input_data: xr.Dataset, dim_az: int = 2, dim_rg: int = 2) -> xr.Dataset: |
| 629 |
"""Apply multilooking to polarimetric matrices. |
| 630 |
|
| 631 |
Args: |
| 632 |
input_data (xr.Dataset): Input PolSARpro Dataset. |
| 633 |
dim_az (int): Multilook dimension in azimuth. |
| 634 |
dim_rg (int): Multilook dimension in range. |
| 635 |
|
| 636 |
Returns: |
| 637 |
xr.Dataset: Multilooked PolSARpro Dataset. |
| 638 |
|
| 639 |
Note: |
| 640 |
The input dataset must be in the SAR geometry (i.e. have 'y' and 'x' coordinates). |
| 641 |
""" |
| 642 |
|
| 643 |
if type(dim_az) != int and type(dim_rg) != int: |
| 644 |
raise ValueError("dimaz and dimrg must be integers") |
| 645 |
|
| 646 |
if (dim_az < 1) or (dim_rg < 1): |
| 647 |
raise ValueError("dimaz and dimrg must be strictly positive") |
| 648 |
|
| 649 |
if not {"y", "x"}.issubset(set(input_data.coords)): |
| 650 |
raise ValueError( |
| 651 |
"Multilooking requires images in the SAR geometry. 'y' and 'x' must be present in the coordinates. For geocoded data, please use xarray.coarsen." |
| 652 |
) |
| 653 |
|
| 654 |
allowed_poltypes = ("C2", "C3", "C4", "T3", "T4") |
| 655 |
validate_dataset(input_data, allowed_poltypes=allowed_poltypes) |
| 656 |
return input_data.coarsen(y=dim_az, x=dim_rg, boundary="trim").mean() |
| 657 |
|
| 658 |
|
| 659 |
def plot_h_alpha_plane(ds, bins=500, min_pts=5): |
| 660 |
"""Plot H-Alpha 2D histogram. |
| 661 |
|
| 662 |
Args: |
| 663 |
ds (xr.Dataset): Dataset containing 'entropy' and 'alpha' variables. |
| 664 |
bins (int): Number of bins along each dimension. |
| 665 |
min_pts (int): If the number of points in one bin is less than this value, display is omitted. |
| 666 |
""" |
| 667 |
|
| 668 |
if not ds.poltype == "h_a_alpha": |
| 669 |
raise ValueError("Input must be a valid PolSARpro H/A/Alpha result.") |
| 670 |
|
| 671 |
if not {"entropy", "alpha"}.issubset(ds.data_vars): |
| 672 |
raise ValueError("Entropy or Alpha is missing from the input data.") |
| 673 |
|
| 674 |
def curve1(n_points): |
| 675 |
|
| 676 |
m = np.linspace(1e-30, 1, n_points) |
| 677 |
l1 = np.ones(n_points) |
| 678 |
l2 = m |
| 679 |
l3 = m |
| 680 |
l = np.vstack([l1, l2, l3]) |
| 681 |
|
| 682 |
p = l / l.sum(0) |
| 683 |
H = np.sum(-p * np.log(p), 0) / np.float32(np.log(3)) |
| 684 |
alpha = p[1] + p[2] |
| 685 |
|
| 686 |
return H, alpha |
| 687 |
|
| 688 |
def curve2(n_points): |
| 689 |
m = np.linspace(1e-30, 1, n_points) |
| 690 |
l1 = np.zeros_like(m) |
| 691 |
l2 = np.ones_like(m) |
| 692 |
l3 = np.zeros_like(m) |
| 693 |
# l1[m < 0.5] = 0 |
| 694 |
l1[m >= 0.5] = 2 * m[m >= 0.5] - 1 |
| 695 |
|
| 696 |
# l3[m < 0.5] = 2 * m[m<0.5] |
| 697 |
l3[m >= 0.5] = 1 |
| 698 |
|
| 699 |
l = np.vstack([l1, l2, l3]) |
| 700 |
|
| 701 |
p = l / l.sum(0) |
| 702 |
H = np.sum(-p * np.log(p + 1e-30), 0) / np.float32(np.log(3)) |
| 703 |
alpha = p[1] + p[2] |
| 704 |
|
| 705 |
return H[m >= 0.5], alpha[m >= 0.5] |
| 706 |
|
| 707 |
x1, y1 = curve1(bins) |
| 708 |
x2, y2 = curve2(bins) |
| 709 |
|
| 710 |
# Flatten and remove NaNs |
| 711 |
entropy = ds.entropy.values.ravel() |
| 712 |
alpha = ds.alpha.values.ravel() |
| 713 |
mask = ~np.isnan(entropy) & ~np.isnan(alpha) |
| 714 |
entropy = entropy[mask].clip(0, 1) |
| 715 |
alpha = alpha[mask].clip(0, 90) |
| 716 |
|
| 717 |
fig, ax = plt.subplots(figsize=(8, 6)) |
| 718 |
|
| 719 |
# 2D histogram |
| 720 |
h = ax.hist2d( |
| 721 |
entropy, alpha, bins=bins, range=((0, 1), (0, 90)), cmap="jet", cmin=min_pts |
| 722 |
) |
| 723 |
plt.colorbar(h[3], ax=ax, label="Number of points") |
| 724 |
|
| 725 |
# Overlay curves |
| 726 |
ax.plot(x1, 90 * y1, color="black", linewidth=1) |
| 727 |
ax.plot(x2, 90 * y2, color="black", linewidth=1) |
| 728 |
|
| 729 |
# Labels and title |
| 730 |
ax.set_xlabel("Entropy (H)") |
| 731 |
ax.set_ylabel("Alpha (°)") |
| 732 |
ax.set_title("H-Alpha plane") |
| 733 |
|
| 734 |
return ax, fig |
| 735 |
|
| 736 |
|
| 737 |
def pauli_rgb(input_data: xr.Dataset, q: float = 0.98) -> xr.DataArray: |
| 738 |
"""Compute Pauli RGB representation from polarimetric data. |
| 739 |
|
| 740 |
Args: |
| 741 |
input_data (xr.Dataset): Input PolSARpro Dataset. |
| 742 |
q (float): Quantile for dynamic range clipping (between 0 and 1). |
| 743 |
|
| 744 |
Returns: |
| 745 |
xr.DataArray: RGB representation with 'band' dimension. |
| 746 |
""" |
| 747 |
allowed_poltypes = ("S", "C3", "T3") |
| 748 |
validate_dataset(input_data, allowed_poltypes=allowed_poltypes) |
| 749 |
|
| 750 |
def pauli_from_S(S): |
| 751 |
r = abs(S.hh - S.vv) ** 2 |
| 752 |
g = abs(0.5 * (S.hv + S.vh)) ** 2 |
| 753 |
b = abs(S.hh + S.vv) ** 2 |
| 754 |
return r, g, b |
| 755 |
|
| 756 |
def pauli_from_T3(T3): |
| 757 |
r = T3.m22 |
| 758 |
g = T3.m33 |
| 759 |
b = T3.m11 |
| 760 |
return r, g, b |
| 761 |
|
| 762 |
def pauli_from_C3(C3): |
| 763 |
T3 = C3_to_T3(C3) |
| 764 |
return pauli_from_T3(T3) |
| 765 |
|
| 766 |
if input_data.poltype == "S": |
| 767 |
r, g, b = pauli_from_S(input_data) |
| 768 |
elif input_data.poltype == "C3": |
| 769 |
r, g, b = pauli_from_C3(input_data) |
| 770 |
elif input_data.poltype == "T3": |
| 771 |
r, g, b = pauli_from_T3(input_data) |
| 772 |
|
| 773 |
rgb = xr.concat([r, g, b], dim="band").rename("Pauli RGB").chunk("auto") |
| 774 |
|
| 775 |
# resampling may introduce negative values |
| 776 |
rgb = xr.where(rgb < 0, 0, rgb) |
| 777 |
rgb = np.sqrt(rgb) |
| 778 |
|
| 779 |
# compute clipping values to handle the high dynamic range |
| 780 |
# clip_val = rgb.quantile(dim=("x", "y"), q=q).astype("float32").drop_vars("quantile") |
| 781 |
clip_val = ( |
| 782 |
rgb.quantile(dim=input_data.dims, q=q).astype("float32").drop_vars("quantile") |
| 783 |
) |
| 784 |
|
| 785 |
# clip and normalize |
| 786 |
return rgb.clip(max=clip_val) / clip_val |