How to change code to classify the whole picture？

[forestbat]

I want to classify the whole remote sensing picture not reference pixel，so I changed code：

    raster = rasterio.open(img_path, 'r')
    arr = raster.read()
    ref_count = 1
    ref_cord = [[round(arr.shape[1]), round(arr.shape[2])]]
    ……
    # ref_vect is reference pixel
    ref_vect = [[0 for x in range(bands)] for y in range(ref_count)]
    for i in range(ref_count):
        for j in range(bands):
            ref_vect[i][j] = arr[j, int(ref_cord[i][0])-1, int(ref_cord[i][1])-1]

But when I run the code，ref_vect is [[0, 0, 0, 0]] and r = ref_vect[j][nb] is all 0，so classified picture is all black.
How to specify ref_count，ref_cord and other variables to make the program work normally？

[cjaca]

Hi,
The algorithm itself is analyzing all 2d layers of the multilayer picture, not only one pixel.
Could you please elaborate more on what remote sensing picture do you have, or upload it here?

[forestbat]

GF_6.zip
Thank you for your reply.
This picture come from GaoFen-1 satellite，and it has 4 bands.

[forestbat]

3 days passed，do you have answer of the problem？

[cjaca]

Hi, I understand the problem now. I need some time to make it work, since it's not my daily business topic. I will notify you once It will be ready solution to analyze without reference pixels for less amount of layers

[lazyn1997]

Your image has a black edge, all the channel values are 0, ref_vect is correct, and then the following formula will give the nan value, which does not satisfy the if condition, so it is classified to the background value, and the whole image is black.

You can get the following classification by arr = numpy.where(arr <= 0, 1e-6, arr) setting the arr ≤0 to a minimal value.