visualize.FeatureSpace doesn't work well with MNF-transformed data

[arthur-e]

Though visualize.FeatureSpace can accept MNF-transformed data (by setting transform=False to indicate the data is already transformed), the resulting plots of the mixing space don't look right... I think that something is going wrong with the indexing (e.g., indexing on m,n axes instead of the p-axis).

[vijai9111995]

how to get the end members form the mnf transformed data?
the manual: -- https://github.com/arthur-e/unmixing/blob/master/docs/Example_Spatially_Adaptive_Spectral_Mixture_Analysis_SASMA.ipynb gives the data which had already got the endmembers. It will be so much useful if you tell how did you got the end members from the image file in python (spyder environment)

[arthur-e]

Hi @vijai9111995, choosing endmembers is a user-driven, subjective process. There is no single, correct answer, but there are two broad approaches:

1. Use expert knowledge or independent data to identify the endmembers. This might take the form of a spectral library, perhaps based on samples from a field spectroradiometer (though these are not as reliable as so-called "image endmembers," the spectra taken from the imagery directly). Or, information on large, spectrally "pure" endmember locations might be known in advance.
2. Use the mixing space, with or without other spectral information, to infer the "image endmembers." These are usually pixels with "extreme" spectral characteristics, e.g., the corners of a simplex fit to the mixing space.

The unmixing library offers tools to help with both approaches. For incorporating expert knowledge, there are tools to convert geographic coordinates to image (row-column) coordinates and visualize the spectra. But the approach best supported here is to infer the endmembers from the mixing space itself.

You should be looking at this demo, on ordinary LSMA, for hints about that. In the section "Visualization of the Mixing Space" you can see plots of the mixing space and a demonstration of the use of Graham's scan algorithm to identify the corners of a convex hull, which are likely to be the corners of the best-fit simplex.

[vijai9111995]

25.tiff <https://drive.google.com/file/d/1fYOmz4lYdPRxApx4E9YeyM5_-jZI_f8K/view?usp=drive_web> 40.tiff <https://drive.google.com/file/d/1YfhQodwNxEfI9EgLvj3YtySZfQiYirdx/view?usp=drive_web> 50.tiff <https://drive.google.com/file/d/1SibTw9omY0U9wXZZBRlLL2Mfvy4NasXn/view?usp=drive_web> I'm using this type of data with 425 bands. The data which i have sent is the sample of it. The original data is in .hdr format. I couldn't perform the mnf with that data so I converted it into a tiff file. After that I couldn't stack this file and do mnf and lsma. So I averaged it and performed mnf but when I plot the result it's not coming correctly. How to perform lsma and get the end members from it. Can you please see that using this data how can i get the endmembers

…

On Wed, Oct 7, 2020 at 8:21 PM K. Arthur Endsley ***@***.***> wrote: Hi @vijai9111995 <https://github.com/vijai9111995>, choosing endmembers is a user-driven, subjective process. There is no single, correct answer, but there are two broad approaches: 1. Use expert knowledge or independent data to identify the endmembers. This might take the form of a spectral library, perhaps based on samples from a field spectroradiometer (though these are not as reliable as so-called "image endmembers," the spectra taken from the imagery directly). Or, information on large, spectrally "pure" endmember locations might be known in advance. 2. Use the mixing space, with or without other spectral information, to infer the "image endmembers." These are usually pixels with "extreme" spectral characteristics, e.g., the corners of a simplex fit to the mixing space. The unmixing library offers tools to help with both approaches. For incorporating expert knowledge, there are tools to convert geographic coordinates to image (row-column) coordinates and visualize the spectra. But the approach best supported here is to infer the endmembers from the mixing space itself. You should be looking at this demo, on ordinary LSMA, <https://github.com/arthur-e/unmixing/blob/master/docs/Overview_LSMA_in_Python.ipynb> for hints about that. In the section "Visualization of the Mixing Space" you can see plots of the mixing space and a demonstration of the use of Graham's scan algorithm to identify the corners of a convex hull, which are likely to be the corners of the best-fit simplex. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AN4E3IOUVCXSIFINSDER3ETSJR57XANCNFSM4DEKF2PQ> .

[arthur-e]

Hi @vijai9111995, please file a new Issue and include code snippets that show the specific steps you are taking. This Issue (#2), about a specific improvement for the code base.

[vijai9111995]

yeah i have filed a new issue -- #9 but i didn't get any response

…

On Thu, Oct 8, 2020 at 7:30 PM K. Arthur Endsley ***@***.***> wrote: Hi @vijai9111995 <https://github.com/vijai9111995>, please file a new Issue and include code snippets that show the specific steps you are taking. This Issue (#2 <#2>), about a specific improvement for the code base. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AN4E3IODUWECPAFREQB5U6DSJXAYRANCNFSM4DEKF2PQ> .