python中的数组/图像插值

Question

我一年中的每一天都有365个2d numpy阵列，显示的图像如下：

​

​

我把它们都堆成了一个3Dnumpy数组。我想要搜索前7天，或未来7天(前7层，下7层)以找到云以外的值，然后用该像素的其他可能值(相应像素在其他天/层中经历的值)替换云值。

我是python的新手，有点迷失方向。

有什么想法吗？

谢谢

Answer 1

我通过以下方式解决了这个问题：

interpdata = []
j = 0
for i in stack:
    try:
        temp = np.where( stack[j] == 50, stack[j-1], modis[j] )
        temp = np.where( temp == 50, stack[j+1], temp )
        temp = np.where( temp == 50, stack[j-2], temp )
        temp = np.where( temp == 50, stack[j+2], temp )
        temp = np.where( temp == 50, stack[j-3], temp )
        temp = np.where( temp == 50, stack[j+3], temp ) 
        temp = np.where( temp == 50, stack[j-4], temp )
        temp = np.where( temp == 50, stack[j+4], temp )
    except IndexError:
        print 'IndexError Passed'       
        pass
    else:
        pass
    interpdata [j, :, :] = temp
    j = j + 1   

Answer 2

您实际上是在尝试为您的数组编写一个过滤器。

在你的例子中，该函数将接受一维数组，并返回最接近中间索引的元素，而不是cloud：

import numpy as np
from scipy.ndimage.filters import generic_filter

_cloud = -1

def findNearestNonCloud(elements):
    middleIndex = len(elements) / 2
    if elements[middleIndex] != _cloud:
        return elements[middleIndex] # middle value is not cloud

    nonCloudIndices, = np.where(elements != _cloud)
    if len(nonCloudIndices) == 0:
        return elements[middleIndex] # all values were cloud

    prevNonCloudIndex = np.where(nonCloudIndices < middleIndex, 
            nonCloudIndices, -1).max()
    nextNonCloudIndex = -np.where(nonCloudIndices > middleIndex, 
            -nonCloudIndices, 1).min()
    # -1 means no non-cloud index

    # pick index closest to middle index    
    if (abs(prevNonCloudIndex - middleIndex) 
            <= abs(nextNonCloudIndex - middleIndex)):
        return elements[prevNonCloudIndex]
    else:
        return elements[nextNonCloudIndex]

现在，您需要将此函数应用于您感兴趣的元素。要做到这一点，您需要一个掩码来表示您对特定元素感兴趣的其他元素。

from scipy.ndimage.filters import generic_filter

# creates 5 days worth of a 3x3 plot of land
input = np.ones((5, 3, 3)) * _cloud
input[0,:,:] = 10 # set first "image" to all be 10s
input[4,0,0] = 12 # uppper left corner of fourth image is set to 12
print "input data\n", input, "\n"

mask = (5, 1, 1)
# mask represents looking at the present day, 2 days in the future and 2 days in 
# the past for 5 days in total.

print "result\n", generic_filter(input, findNearestNonCloud, size=mask)
# second and third images should mirror first image, 
# except upper left corner of third image should be 12

Answer 3

我认为你可以这样做：

data = somehow_get_your_3d_data() #indexed as [day_of_year,y,x]
for i,dat in enumerate(data):
    weeks2 = data[max(i-7,i):min(i+7,len(data)), ... ]
    new_value = get_new_value(weeks2) #get value from weeks2 here somehow
    dat[dat == cloud_value] = new_value