zhanghuilai/BRDF
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Initial commit

Browse Source
This commit is contained in:
zhanghuilai
2026-04-10 16:46:45 +08:00
commit 4fd1b0a203
165 changed files with 25698 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

363
Flexbrdf/hytools/brdf/flex.py Normal file
View File

@ -0,0 +1,363 @@
# -*- coding: utf-8 -*-
"""
HyTools: Hyperspectral image processing library
Copyright (C) 2021 University of Wisconsin
Authors: Adam Chlus, Zhiwei Ye, Philip Townsend.
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, version 3 of the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <https://www.gnu.org/licenses/>.
本模块包含应用经验性 BRDF 校正的函数，如下论文所述：
方程和常数可在以下论文中找到：
"""
import numpy as np
import ray
from scipy.interpolate import interp1d
from .kernels import calc_volume_kernel,calc_geom_kernel
from ..masks import mask_create
from ..misc import progbar, pairwise
from ..misc import update_brdf
from ..plotting import flex_diagno_plot
def flex_brdf(actors,config_dict):
brdf_dict= config_dict['brdf']
if brdf_dict['grouped']:
calc_flex_group(actors,brdf_dict)
else:
_ = ray.get([a.do.remote(calc_flex_single,brdf_dict) for a in actors])
if "diagnostic_plots" in brdf_dict:
if brdf_dict['diagnostic_plots']:
print('Exporting diagnostic plots.')
_ = ray.get([a.do.remote(flex_diagno_plot,config_dict) for a in actors])
def ndvi_stratify(hy_obj):
'''创建 NDVI 分箱分层掩膜
'''
ndvi = hy_obj.ndi()
class_mask = np.zeros((hy_obj.lines, hy_obj.columns))
for bin_num in hy_obj.brdf['bins']:
start,end = hy_obj.brdf['bins'][bin_num]
class_mask[(ndvi > start) & (ndvi <= end)] = bin_num
class_mask[~hy_obj.mask['calc_brdf']] = 0
#Subsample data
idx = np.array(np.where(class_mask!=0)).T
idxRand= idx[np.random.choice(range(len(idx)),int(len(idx)*(1-hy_obj.brdf['sample_perc'])), replace = False)].T
class_mask[idxRand[0],idxRand[1]] = 0
class_mask = class_mask.astype(np.int8)
hy_obj.ancillary['ndvi_classes'] = class_mask
def ndvi_2nd_split(ndvi_bins_dynamic, all_ndvi_array, ndvi_bin_range_thres=0.15):
''' 执行第二次 NDVI 分割
'''
ndvi_bin_range_thres = -0.015625 * (len(ndvi_bins_dynamic)-1) + 0.43125
ndvi_bin_range = np.array(ndvi_bins_dynamic[1:]) - np.array(ndvi_bins_dynamic[:-1])
bin_for_split = np.argwhere(ndvi_bin_range>=ndvi_bin_range_thres).ravel()
new_break = []
if bin_for_split.shape[0]>0:
for bin_id in bin_for_split:
# Use median of the bin as the new break point
new_break += [np.median(all_ndvi_array[(all_ndvi_array > ndvi_bins_dynamic[bin_id]) & (all_ndvi_array < ndvi_bins_dynamic[bin_id+1])]).astype(np.float64)]
# New list of bin break points
ndvi_bins_dynamic = sorted(ndvi_bins_dynamic + new_break)
return ndvi_bins_dynamic
def ndvi_bins(ndvi,brdf_dict):
'''计算 NDVI 分箱范围
'''
perc_range = brdf_dict['ndvi_perc_max'] - brdf_dict['ndvi_perc_min'] + 1
ndvi_break_dyn_bin = np.percentile(ndvi[ndvi > 0],
np.arange(brdf_dict['ndvi_perc_min'],
brdf_dict['ndvi_perc_max'] + 1,
perc_range / (brdf_dict['num_bins'] - 1)))
ndvi_thres = [brdf_dict['ndvi_bin_min']]
ndvi_thres += ndvi_break_dyn_bin.tolist()
ndvi_thres += [brdf_dict['ndvi_bin_max']]
ndvi_thres = sorted(list(set(ndvi_thres)))
# 对 NDVI 分箱进行第二次分割
ndvi_thres = ndvi_2nd_split(ndvi_thres, ndvi)
bins = [[x,y] for x,y in pairwise(ndvi_thres)]
return bins
def get_kernel_samples(hy_obj):
'''计算并采样 BRDF 核函数
'''
geom_kernel = hy_obj.geom_kernel(hy_obj.brdf['geometric'],
b_r=hy_obj.brdf["b/r"] ,
h_b =hy_obj.brdf["h/b"])
geom_kernel = geom_kernel[hy_obj.ancillary['ndvi_classes'] !=0]
vol_kernel = hy_obj.volume_kernel(hy_obj.brdf['volume'])
vol_kernel = vol_kernel[hy_obj.ancillary['ndvi_classes'] !=0]
classes = hy_obj.ancillary['ndvi_classes'][hy_obj.ancillary['ndvi_classes'] !=0]
X = np.vstack([vol_kernel,geom_kernel,
np.ones(vol_kernel.shape),classes]).T
return X
def get_band_samples(hy_obj,args):
band = hy_obj.get_band(args['band_num'],
corrections = hy_obj.corrections)
return band[hy_obj.ancillary['ndvi_classes'] !=0]
def calc_flex_single(hy_obj,brdf_dict):
''' 计算单个图像的 BRDF 系数
'''
hy_obj.brdf['coeffs'] ={}
# 确定分箱维度并创建类别掩膜
if hy_obj.brdf['bin_type'] == 'dynamic':
bins = ndvi_bins(hy_obj.ndi()[hy_obj.mask['no_data']],brdf_dict)
# 更新分箱数量
hy_obj.brdf['num_bins'] = len(bins)
else:
bins = brdf_dict['bins']
hy_obj.brdf['bins'] = {k:v for (k,v) in enumerate(bins,start=1)}
ndvi_stratify(hy_obj)
kernel_samples= get_kernel_samples(hy_obj)
# 计算每个波段和类别的系数
for band_num,band in enumerate(hy_obj.bad_bands):
if ~band:
hy_obj.brdf['coeffs'][band_num] = {}
band_samples = hy_obj.do(get_band_samples, {'band_num':band_num})
coeffs= []
for bin_num in hy_obj.brdf['bins']:
bin_mask = (kernel_samples[:,3] == bin_num)
X = kernel_samples[:,:3][bin_mask]
y = band_samples[bin_mask]
coeffs.append(np.linalg.lstsq(X, y,rcond=-1)[0].flatten().tolist())
hy_obj.brdf['coeffs'][band_num] = coeffs
def calc_flex_group(actors,brdf_dict):
''' 计算一组图像的 BRDF 系数
'''
# 从图像聚合 NDVI 值
ndvi = ray.get([a.ndi.remote(mask = 'no_data') for a in actors])
ndvi = np.concatenate([n.flatten() for n in ndvi])
# 确定分箱维度
if brdf_dict['bin_type'] == 'dynamic':
bins = ndvi_bins(ndvi,brdf_dict)
# 更新分箱数量
_ = ray.get([a.do.remote(update_brdf,{'key':'num_bins',
'value': len(bins)}) for a in actors])
else:
bins = brdf_dict['bins']
bins = {k:v for (k,v) in enumerate(bins,start=1)}
# 更新 BRDF 系数
_ = ray.get([a.do.remote(update_brdf,{'key':'bins',
'value': bins}) for a in actors])
# 创建 NDVI 类别掩膜并采样核函数
_ = ray.get([a.do.remote(ndvi_stratify) for a in actors])
kernel_samples = ray.get([a.do.remote(get_kernel_samples) for a in actors])
kernel_samples = np.concatenate(kernel_samples)
bad_bands = ray.get(actors[0].do.remote(lambda x: x.bad_bands))
coeffs = {}
for band_num,band in enumerate(bad_bands):
if ~band:
coeffs[band_num] = {}
band_samples = ray.get([a.do.remote(get_band_samples,
{'band_num':band_num}) for a in actors])
band_samples = np.concatenate(band_samples)
band_coeffs= []
for bin_num in bins:
bin_mask = (kernel_samples[:,3] == bin_num)
X = kernel_samples[:,:3][bin_mask]
y = band_samples[bin_mask]
band_coeffs.append(np.linalg.lstsq(X, y,rcond=-1)[0].flatten().tolist())
coeffs[band_num] = band_coeffs
progbar(np.sum(~bad_bands[:band_num+1]),np.sum(~bad_bands))
print('\n')
# 更新 BRDF 系数
_ = ray.get([a.do.remote(update_brdf,{'key':'coeffs',
'value': coeffs}) for a in actors])
def apply_flex(hy_obj,data,dimension,index):
''' 对数据切片应用 flex BRDF 校正
参数:
hy_obj : Hytools 类对象。
data (np.ndarray): 数据切片。
index (int,list): 数据索引。
返回:
data (np.ndarray): BRDF 校正后的数据切片。
'''
if 'k_vol' not in hy_obj.ancillary:
hy_obj.ancillary['k_vol'] = hy_obj.volume_kernel(hy_obj.brdf['volume'])
if 'k_geom' not in hy_obj.ancillary:
hy_obj.ancillary['k_geom'] = hy_obj.geom_kernel(hy_obj.brdf['geometric'],
b_r=hy_obj.brdf["b/r"],
h_b =hy_obj.brdf["h/b"])
if ('k_vol_nadir' not in hy_obj.ancillary) or ('k_geom_nadir' not in hy_obj.ancillary):
solar_zn = hy_obj.brdf['solar_zn_norm_radians'] * np.ones((hy_obj.lines,hy_obj.columns))
hy_obj.ancillary['k_vol_nadir'] = calc_volume_kernel(0,solar_zn,
0,0,hy_obj.brdf['volume'])
hy_obj.ancillary['k_geom_nadir'] = calc_geom_kernel(0,solar_zn,
0,0,hy_obj.brdf['geometric'],
b_r=hy_obj.brdf["b/r"],
h_b =hy_obj.brdf["h/b"])
if 'apply_brdf' not in hy_obj.mask:
hy_obj.gen_mask(mask_create,'apply_brdf',hy_obj.brdf['apply_mask'])
if 'ndvi' not in hy_obj.ancillary:
hy_obj.ancillary['ndvi'] = hy_obj.ndi()
if 'interpolators' not in hy_obj.ancillary:
bin_centers = np.mean(list(hy_obj.brdf['bins'].values()),axis=1)
hy_obj.ancillary['interpolators'] ={}
# 生成插值器
for i in hy_obj.brdf['coeffs']:
coeffs= np.array(hy_obj.brdf['coeffs'][i])
interpolator = interp1d(bin_centers, coeffs, kind = hy_obj.brdf['interp_kind'],
axis=0,fill_value="extrapolate")
hy_obj.ancillary['interpolators'][int(i)] = interpolator
# 转换为浮点数
data = data.astype(np.float32)
brdf_bands = [int(x) for x in hy_obj.ancillary['interpolators']]
if dimension == 'line':
# index= 3000
# data = hy_obj.get_line(3000)
interpolated_f = [hy_obj.ancillary['interpolators'][band](hy_obj.ancillary['ndvi'][index,:]) for band in brdf_bands]
interpolated_f = np.array(interpolated_f)
fvol, fgeo, fiso = interpolated_f[:,:,0], interpolated_f[:,:,1], interpolated_f[:,:,2]
brdf = fvol*hy_obj.ancillary['k_vol'][index,:]
brdf+= fgeo*hy_obj.ancillary['k_geom'][index,:]
brdf+= fiso
brdf_nadir = fvol*hy_obj.ancillary['k_vol_nadir'][index,:]
brdf_nadir+= fgeo*hy_obj.ancillary['k_geom_nadir'][index,:]
brdf_nadir+= fiso
correction_factor = brdf_nadir/brdf
correction_factor[:,~hy_obj.mask['apply_brdf'][index]] = 1
data[:,brdf_bands] = data[:,brdf_bands]*correction_factor.T
elif dimension == 'column':
#index= 300
#data = hy_obj.get_column(index)
interpolated_f = [hy_obj.ancillary['interpolators'][band](hy_obj.ancillary['ndvi'][:,index]) for band in brdf_bands]
interpolated_f = np.array(interpolated_f)
fvol, fgeo, fiso = interpolated_f[:,:,0], interpolated_f[:,:,1], interpolated_f[:,:,2]
brdf = fvol*hy_obj.ancillary['k_vol'][:,index]
brdf+= fgeo*hy_obj.ancillary['k_geom'][:,index]
brdf+= fiso
brdf_nadir = fvol*hy_obj.ancillary['k_vol_nadir'][:,index]
brdf_nadir+= fgeo*hy_obj.ancillary['k_geom_nadir'][:,index]
brdf_nadir+= fiso
correction_factor = brdf_nadir/brdf
correction_factor = np.moveaxis(correction_factor,0,1)
correction_factor[:,~hy_obj.mask['apply_brdf'][index]] = 1
data[:,brdf_bands] = data[:,brdf_bands]*correction_factor.T
elif (dimension == 'band') & (index in brdf_bands):
# index= 8
# data = hy_obj.get_band(index)
interpolated_f = hy_obj.ancillary['interpolators'][index](hy_obj.ancillary['ndvi'])
fvol, fgeo, fiso = interpolated_f[:,:,0], interpolated_f[:,:,1], interpolated_f[:,:,2]
brdf = fvol*hy_obj.ancillary['k_vol']
brdf += fgeo*hy_obj.ancillary['k_geom']
brdf += fiso
brdf_nadir = fvol*hy_obj.ancillary['k_vol_nadir']
brdf_nadir += fgeo*hy_obj.ancillary['k_geom_nadir']
brdf_nadir += fiso
correction_factor = brdf_nadir/brdf
correction_factor[~hy_obj.mask['apply_brdf']] = 1
data= data* correction_factor
elif dimension == 'chunk':
# index = 200,501,3000,3501
x1,x2,y1,y2 = index
# data = hy_obj.get_chunk(x1,x2,y1,y2)
interpolated_f = [hy_obj.ancillary['interpolators'][band](hy_obj.ancillary['ndvi'][y1:y2,x1:x2]) for band in brdf_bands]
interpolated_f = np.array(interpolated_f)
interpolated_f = np.swapaxes(interpolated_f,0,-1)
fvol, fgeo, fiso = interpolated_f[0,:,:,:], interpolated_f[1,:,:,:], interpolated_f[2,:,:,:]
brdf = fvol*hy_obj.ancillary['k_vol'][y1:y2,x1:x2,np.newaxis]
brdf+= fgeo*hy_obj.ancillary['k_geom'][y1:y2,x1:x2,np.newaxis]
brdf+= fiso
brdf_nadir = fvol*hy_obj.ancillary['k_vol_nadir'][y1:y2,x1:x2,np.newaxis]
brdf_nadir+= fgeo*hy_obj.ancillary['k_geom_nadir'][y1:y2,x1:x2,np.newaxis]
brdf_nadir+= fiso
correction_factor = brdf_nadir/brdf
correction_factor[~hy_obj.mask['apply_brdf'][y1:y2,x1:x2]] = 1
data[:,:,brdf_bands] = data[:,:,brdf_bands]*correction_factor
elif dimension == 'pixels':
# index = [[2000,2001],[200,501]]
y,x = index
# data = hy_obj.get_pixels(y,x)
interpolated_f = [hy_obj.ancillary['interpolators'][band](hy_obj.ancillary['ndvi'][y,x]) for band in brdf_bands]
interpolated_f = np.array(interpolated_f)
interpolated_f = np.swapaxes(interpolated_f,0,1)
fvol, fgeo, fiso = interpolated_f[:,:,0], interpolated_f[:,:,1], interpolated_f[:,:,2]
brdf = fvol*hy_obj.ancillary['k_vol'][y,x,np.newaxis]
brdf+= fgeo*hy_obj.ancillary['k_geom'][y,x,np.newaxis]
brdf+= fiso
brdf_nadir = fvol*hy_obj.ancillary['k_vol_nadir'][y,x,np.newaxis]
brdf_nadir+= fgeo*hy_obj.ancillary['k_geom_nadir'][y,x,np.newaxis]
brdf_nadir+= fiso
correction_factor = brdf_nadir/brdf
correction_factor[~hy_obj.mask['apply_brdf'][y,x]] = 1
data[:,brdf_bands] = data[:,brdf_bands]*correction_factor
return data