import os
import numpy as np
from polsartools.utils.proc_utils import process_chunks_parallel
from polsartools.utils.utils import conv2d,time_it,eig22
from .dxp_infiles import dxpc2files, S_norm, stokes_global_stats
[docs]
@time_it
def dprsi(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
):
""" This codes computes the Dual-pol Radar Surface Index (DpRSI) from C2 matrix
Examples
--------
>>> # Basic usage with default parameters
>>> dprsi("/path/to/c2_data")
>>> # Advanced usage with custom parameters
>>> dprsi(
... in_dir="/path/to/c2_data",
... win=3,
... fmt="tif",
... cog=True,
... block_size=(1024, 1024)
... )
Parameters
----------
in_dir : str
Path to the input folder containing C2 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 a Cloud Optimized GeoTIFF (COG) 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, applies LZW compression to the output GeoTIFF files.
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
Results are written to disk as either 'DpRSI.tif' or 'DpRSI.bin'
in the input folder.
"""
write_flag=True
input_filepaths = dxpc2files(in_dir)
output_filepaths = []
if fmt == "bin":
output_filepaths.append(os.path.join(in_dir, "DpRSI.bin"))
else:
output_filepaths.append(os.path.join(in_dir, "DpRSI.tif"))
print("Computing global statistics for normalization...")
c11File = next(f for f in input_filepaths if "C11" in f)
c22File = next(f for f in input_filepaths if "C22" in f)
c12realFile = next(f for f in input_filepaths if "C12_real" in f)
c12imagFile = next(f for f in input_filepaths if "C12_imag" in f)
S0_2, S0_98, S0_max, S1_2, S1_98, S1_max, S2_2, S2_98, S2_max, S3_2, S3_98, S3_max = stokes_global_stats(c11File, c22File, c12realFile, c12imagFile)
process_chunks_parallel(input_filepaths, list(output_filepaths), win, write_flag,
process_chunk_dprsi,
*[S0_2, S0_98, S0_max, S1_2, S1_98, S1_max, S2_2, S2_98, S2_max, S3_2, S3_98, S3_max],
block_size=block_size, max_workers=max_workers, num_outputs=1,
cog=cog,ovr=ovr, comp=comp,
progress_callback=progress_callback
)
def process_chunk_dprsi(chunks, window_size,*args):
S0_2 = float(args[-12])
S0_98 = float(args[-11])
S0_max = float(args[-10])
S1_2 = float(args[-9])
S1_98 = float(args[-8])
S1_max = float(args[-7])
S2_2 = float(args[-6])
S2_98 = float(args[-5])
S2_max = float(args[-4])
S3_2 = float(args[-3])
S3_98 = float(args[-2])
S3_max = float(args[-1])
kernel = np.ones((window_size,window_size),np.float32)/(window_size*window_size)
c11_T1 = np.array(chunks[0])
c12_T1 = np.array(chunks[1])+1j*np.array(chunks[2])
# c21_T1 = np.conj(c12_T1)
c22_T1 = np.array(chunks[3])
if window_size>1:
c11s = conv2d(np.real(c11_T1),kernel)+1j*conv2d(np.imag(c11_T1),kernel)
c12s = conv2d(np.real(c12_T1),kernel)+1j*conv2d(np.imag(c12_T1),kernel)
# c21s = conv2d(np.real(c21_T1),kernel)+1j*conv2d(np.imag(c21_T1),kernel)
c22s = conv2d(np.real(c22_T1),kernel)+1j*conv2d(np.imag(c22_T1),kernel)
else:
c11s = c11_T1
c12s = c12_T1
c22s = c22_T1
C11_av_db = 10*np.log10(c11s)
s0 = c11s + c22s
s1 = c11s - c22s
s2 = 2*c12s.real
s3 = 2*c12s.imag
##### Calculate Entropy
## Here eigen values are calculated using Stokes vector elements
tpp = np.sqrt(np.square(s1) + np.square(s2) + np.square(s3))
lmbd1 = (s0 + tpp)/2
lmbd2 = (s0 - tpp)/2
prob1 = lmbd1/(lmbd1 + lmbd2)
prob2 = lmbd2/(lmbd1 + lmbd2)
ent = -prob1*np.log2(prob1) - prob2*np.log2(prob2)
##### Taking abs of Stokes vector elements
s0 = np.abs(s0)
s1 = np.abs(s1)
# s1_s_norm = S_norm(s1) #This is S1 normalzied for DpRSI, does not include slope mask
s1_s_norm = np.clip(s1, S1_2, S1_98) / S1_max
##### Dual-pol Radar Surface Index
dprsi_con1 = (1 - ent)*np.sqrt(1 - np.square(s1_s_norm)); # For Valid pixels
dprsi_con2 = np.sqrt(1 - np.square(s1_s_norm)); # For Noise pixels
NESZ = -16 ## For Sentinel-1
dprsi = np.where(C11_av_db > NESZ, dprsi_con1, dprsi_con2)
return dprsi.astype(np.float32)
