ArcPy和NumPy：处理Landsat 8图像的内存错误

Question

我试图使用ArcPy（有经验的）和NumPy（新手）比较NDVI calculations，但在下面的脚本所示的行中遇到了MemoryError。在

由^{}创建的数组的形状和数据类型是(8191, 8101)和{}。这是一个标准的陆地卫星8号图像的大小。在

这是一个不合理的尺寸期望处理？division提升数据类型以容纳小数（diff和{}成功完成）的问题吗？如果是这样的话，我可以强制一个整数输出吗？在

import cProfile, numpy
def arcpyNDVI():
    NIRras = arcpy.Raster("LC80460222013184LGN00_B5.TIF")
    REDras = arcpy.Raster("LC80460222013184LGN00_B4.TIF")
    NDVIap = (NIRras - arcpy.sa.Float(REDras))/ (NIRras + REDras)
    NDVIap.save(r'C:/junk/ap_ras.tif')
def numpyNDVI():
    NIRras = arcpy.Raster("LC80460222013184LGN00_B5.TIF")
    NIRll = arcpy.Point(NIRras.extent.XMin,NIRras.extent.YMin)
    NIRcs = NIRras.meanCellWidth
    REDras = arcpy.Raster("LC80460222013184LGN00_B4.TIF") 
    arcpy.env.outputCoordinateSystem = NIRras.spatialReference
    NIRnp = arcpy.RasterToNumPyArray(NIRras)
    REDnp = arcpy.RasterToNumPyArray(REDras)
    diff = NIRnp - REDnp
    sum = NIRnp + REDnp
    NDVInp = diff / sum # MEMORY ERROR HERE
    NDVInpRas = arcpy.NumPyArrayToRaster(NDVInp,NIRll,NIRcs,NIRcs)
    NDVInpRas.save(r'C:/junk/np_ras.tif')
cProfile.runctx('arcpyNDVI()',None,locals())
cProfile.runctx('numpyNDVI()',None,locals())

更新1：上面的脚本有时运行到完成（有时给出MemoryError），但它输出uint16，而我需要小数。换行后： NDVInp = diff / sum到：NDVInp = numpy.true_divide(diff,sum)，我一直得到：

^{pr2}$

Answer 1

答案似乎是使用del语句释放内存，尽早删除变量。如果有人想评论这是否是numpy脚本（或Python，一般来说）的常见/最佳实践，我会洗耳恭听。以下是我的工作脚本：

import numpy
NIRras = arcpy.Raster("LC80460222013184LGN00_B5.TIF")
NIRll = arcpy.Point(NIRras.extent.XMin,NIRras.extent.YMin)
NIRcs = NIRras.meanCellWidth
REDras = arcpy.Raster("LC80460222013184LGN00_B4.TIF") 
arcpy.env.outputCoordinateSystem = NIRras.spatialReference
NIRnp = arcpy.RasterToNumPyArray(NIRras)
del NIRras
REDnp = arcpy.RasterToNumPyArray(REDras)
del REDras
diff = NIRnp - REDnp
sum = NIRnp + REDnp
del NIRnp, REDnp
NDVInp = numpy.true_divide(diff,sum) # MEMORY ERROR HERE
NDVInpRas = arcpy.NumPyArrayToRaster(NDVInp,NIRll,NIRcs,NIRcs)
del NDVInp, NIRll, NIRcs
NDVInpRas.save(r'C:/junk/np_ras.tif')
del NDVInpRas