Ludovic Legros New Member
Posts:29  
08 Dec 2017 09:36 AM |
|
Hello, I tried a MTMF method to map alteration minerals (alunite, kaolinite), unfortunately, it doesn't match at all, the results I had with the SAM method or PCA (the pixels with high MF values and low unfeasibility are quite the opposite actually). I probably did something wrong. Here is how I preceeded: I performed a MNF on my image. I used the forward MNF to spectra tool to convert the spectra from usgs library to MNF space (I saved the resulted plot in ASCII). I used the MTMF method and imported the ascii file created previously. But when I try a scatter plot (X= MF ; Y=Unfeasibility), The areas detected by SAM and PCA methods (supposed to contain high concentrations of alteration minerals) have high infeasibility values and negative MF values. Do you see where I could have done wrong ? Thanks.
|
|
|
|
MariM Veteran Member
Posts:2396  
08 Dec 2017 10:12 AM |
|
Since the spectral library is lab collected reflectance data, did you also convert your image to reflectance by performing a Radiometric and Atmospheric correction prior to running MNF? If you use something like FLAASH or Quac from the Atmospheric correction module, the result is scaled reflectance by 10000. I would then use Band Math to divide the data by 10000 to get it into percent reflectance between 0-1.0. This way you are working from similar data ranges and values.
|
|
|
|
Ludovic Legros New Member
Posts:29  
08 Dec 2017 01:56 PM |
|
My image is already in relfectance with values between 0-1.0.
|
|
|
|
MariM Veteran Member
Posts:2396  
08 Dec 2017 03:19 PM |
|
Try using the Spectral Hourglass wizard and you can extract endmembers from the image itself and also use a library. The output products can be SAM and MTMF. Compare the results from following the Wizard with what you get when using only the spectral library. Sometimes the libraries aren't suitable for a particular image's materials. Sometimes the image is too different to compare to the optimal conditions of a spectral library created in a lab.
|
|
|
|
Ludovic Legros New Member
Posts:29  
13 Dec 2017 09:21 AM |
|
Hi ! Finally, I found where was the main problem. I forgot to apply a resampling of the spectral library before using the tool "forward MNF to spectra". Now, it works. But I have one question: after I select a class from the scatter plot (MF vs infeasibility), what do I need to do to produce a classified map according to my selected class ? Thank you !
|
|
|
|
MariM Veteran Member
Posts:2396  
13 Dec 2017 09:45 AM |
|
Typically you will export the selections to ROIs. ROIs can be used as training data in any of the classifiers such as SAM, Maximum likelihood, MTMF, etc. I would use the Endmember Selection tool to import the ROIs, then select the preferred classification/mapping method under 'Algorithm'.
|
|
|
|
Ludovic Legros New Member
Posts:29  
13 Dec 2017 02:11 PM |
|
I have on more question. I tried for the first time the new version of ENVI, the 5.4.1. It's a trial version. In the scatter plot (MF vs infeasibility), when I create one class with all the data (all the point in the scatter plot), some pixels are not selected. If I use the cursor value tool, I remark that the pixels not selected have values within the area I selected with the scatter plot. Is there a problem with the scatter plot of the new version, or I use it wrong ? I tried with my old 5.2 version and when I select all the data in the scatter plot, all the pixels are selected. With the same dataset.
|
|
|
|
Ludovic Legros New Member
Posts:29  
15 Dec 2017 10:03 AM |
|
I tried again with the ENVI classic 5.4 and it works. Do you have an idea why it doesn't work correctly with the ENVI 5.4 version with the "new" interface ?
|
|
|
|