Question

我得到以下代码来从多波段图像中导出索引。因为我的图像很大，所以我不得不按块处理它。现在我想将它转换为并行for循环来更快地完成它。我已经看到有这样的库（例如ParallelProcessing）可以做到这一点，但我很难将它翻译成我的情况......任何想法？数据可以在http://www.gis.usu.edu/~chrisg/python/2009/lectures/ospy_data5.zip

找到

import os, sys, numpy, gdal, utils2
from gdalconst import *

os.chdir('c:/temp/py5')

# register all of the GDAL drivers
gdal.AllRegister()

# open the image
inDs = gdal.Open('aster.img', GA_ReadOnly)
if inDs is None:
  print 'Could not open aster.img'
  sys.exit(1)

# get image size
rows = inDs.RasterYSize
cols = inDs.RasterXSize
bands = inDs.RasterCount

# get the bands and block sizes
inBand2 = inDs.GetRasterBand(2)
inBand3 = inDs.GetRasterBand(3)
blockSizes = utils2.GetBlockSize(inBand2)
xBlockSize = blockSizes[0]
yBlockSize = blockSizes[1]
print yBlockSize, xBlockSize

# create the output image
driver = inDs.GetDriver()
outDs = driver.Create('ndvi.img', cols, rows, 1, GDT_Float32)
if outDs is None:
  print 'Could not create ndvi.img'
  sys.exit(1)
outBand = outDs.GetRasterBand(1)

# loop through the rows
for i in range(0, rows, yBlockSize):
  if i + yBlockSize < rows:
    numRows = yBlockSize
  else:
    numRows = rows - i

  # loop through the columns
  for j in range(0, cols, xBlockSize):
    if j + xBlockSize < cols:
      numCols = xBlockSize
    else:
      numCols = cols - j

    # read the data in
    data2 = inBand2.ReadAsArray(j, i, numCols, numRows).astype(numpy.Float16)
    data3 = inBand3.ReadAsArray(j, i, numCols, numRows).astype(numpy.Float16)

    # do the calculations
    mask = numpy.greater(data2 + data3, 0)
    ndvi = numpy.choose(mask, (-99, (data3 - data2) / (data3 + data2)))

    # write the data
    outBand.WriteArray(ndvi, j, i)

# flush data to disk, set the NoData value and calculate stats
outBand.FlushCache()
outBand.SetNoDataValue(-99)
stats = outBand.GetStatistics(0, 1)

# georeference the image and set the projection
outDs.SetGeoTransform(inDs.GetGeoTransform())
outDs.SetProjection(inDs.GetProjection())

# build pyramids
gdal.SetConfigOption('HFA_USE_RRD', 'YES')
outDs.BuildOverviews(overviewlist=[2,4,8,16,32,64,128])

inDs = None
outDs = None

通过块并行访问图像块

0 个答案: