pqm: Pixel quality metrics

moraine.api.pqm（源码）。每个名字也是 moraine.<name>（import moraine as mr）。

temp_coh

temp_coh(intf: ndarray, rslc: ndarray, image_pairs: ndarray = None, block_size: int = 128)

Estimation of temporal coherence.

Parameters:

Name	Type	Description	Default
intf	ndarray	complex interferograms/coherence metrix, dtype cp/np.complex64, shape 2D(pc) or 3D(ras)	required
rslc	ndarray	complex rslc/phase history, dtype cp/np.complex64, shape 2D(pc) or 3D(ras)	required
image_pairs	ndarray	image pairs	None
block_size	int	the CUDA block size, a multiple of 32, only applied for cuda	128

示例

构建本手册时运行；输出紧随每段代码。

import numpy as np
import moraine as mr

rng = np.random.default_rng(0)
n_points, nimages = 4, 6
phi = rng.uniform(-np.pi, np.pi, (n_points, nimages))
ph = np.exp(1j * phi).astype(np.complex64)                       # phase history of the points, (n_points, nimages)
pairs = mr.TempNet.from_bandwidth(nimages, bandwidth=1).image_pairs
clean = (ph[:, pairs[:, 0]] * np.conj(ph[:, pairs[:, 1]])).astype(np.complex64)   # the sequential interferograms
noise = np.exp(1j * 0.6 * rng.standard_normal((n_points, nimages))).astype(np.complex64)
noise[0] = 1                                                     # the first point is clean

t_coh = mr.temp_coh(clean, ph * noise, pairs)                   # agreement of the (filtered) interferograms with the raw phases
print(t_coh.shape, t_coh.dtype)
print(t_coh.round(3))

(4,) float32
[1.    0.968 0.566 0.856]

In the pipelines the interferograms are the deep learning filtered ones and rslc the raw stack or phase history: points whose filtered phase agrees with the raw phase are kept.