How to turn an ordinary digital camera into a hyperspectral camera using deep learning?

Поділитися
Вставка
  • Опубліковано 1 гру 2024

КОМЕНТАРІ • 33

  • @mojtabamaali1127
    @mojtabamaali1127 2 роки тому

    Thank you for making this video!

  • @markstephen6677
    @markstephen6677 2 роки тому

    Thanks for making this video!!

  • @YomiOlaoye
    @YomiOlaoye Рік тому

    @morteza thanks for this video..I would like to ask if this model would be able to turn an RGB image(e.g. a picture of a tree in my backyard) that is not in the RGB and HSI files folder you have into a hyperspectral image? Also, how can I get the code files too, I would really appreciate it. Thank you.

  • @AmanSharma-vf7vs
    @AmanSharma-vf7vs 3 місяці тому

    This model need dataset that contains both RGB and hyperspectral images ???
    because i do not have hyperspectral images

    • @Morteza_Sc
      @Morteza_Sc  3 місяці тому

      @@AmanSharma-vf7vs yes, every training models need two sets of data, training and testing. There are datasets out there where you can find the training and testing sets of your need. Email me and I can send the link to you…

  • @infinity3865
    @infinity3865 11 місяців тому

    Is the same images which names u saved in untilled 2. Same input images in png form u r giving as input for matlab code

  • @ДСеребряков
    @ДСеребряков Рік тому

    Thank you for making this video! It seems like everything is clear, the only thing that writes the error "ValueError: Unknown mat file type, version 179, 29" at the output (7.52-11.23)

  • @1bhogan
    @1bhogan 10 місяців тому

    Thanks for the great research content, Morteza! I am confused as to the size of the training output, that is the hyperspectral cubes. Given the dataset you have, there are 32 sample images of 512 x 512 dimension in a hypercube with 31 spectral bands, but each band is represented by an image with still 3 channels (RBG), correct? It makes sense that the RGB input is of size (32, 512, 512, 3), but what is the size of the training output?

    • @1bhogan
      @1bhogan 10 місяців тому

      Ah, do you have to convert the hyperspectral images to grayscale when you make the hypercube?

    • @Morteza_Sc
      @Morteza_Sc  10 місяців тому +1

      @@1bhogan So, the size of the input is (512,512,3) and the size of the output is (512,512,31). The number of images don't matter, as in the training the training images would go through the network one by one, according to the number of filters, which is three for the first layer, and 31 for the last layer.

  • @rusticagenerica
    @rusticagenerica Рік тому

    How does this method compare to a real physical H.S. camera, like the Specim IQ Portable Hyperspectral Imaging Camera ?

    • @Morteza_Sc
      @Morteza_Sc  Рік тому

      Using this method, you could only estimate the spectral reflectance in the visible part of the spectrum but with a HS system you could get the spectral reflectance in areas like IR, UV as well. You should be careful what dataset you use as training using this method too! If you have material made out of plastic in your testing, you should make sure you have material made out of plastic in your training as well. Hope this helps!

  • @ДСеребряков
    @ДСеребряков Рік тому

    @Morteza
    Sorry, I understand what the error is, the only thing is not clear how the file with the hyperspectral image turned out

    • @Morteza_Sc
      @Morteza_Sc  Рік тому

      The file that contains the hyperspectral image is a mat file, if that what you mean. Please let me know if that answers your question. Also, sorry for a late response, I was off for Christmas.

    • @ДСеребряков
      @ДСеребряков Рік тому

      @@Morteza_Sc
      Thanks for the answer, yes exactly. Merry Christmas!

  • @sohamparab2258
    @sohamparab2258 Рік тому

    Hii morteza, can you please provide a link to the code for reference.

    • @Morteza_Sc
      @Morteza_Sc  Рік тому +1

      Send me an email and I’ll send you the code!

    • @sohamparab2258
      @sohamparab2258 Рік тому

      Can you drop your email here please

  • @infinity3865
    @infinity3865 11 місяців тому

    Hey what input images u are giving in matlab code?

    • @infinity3865
      @infinity3865 11 місяців тому

      Is the same images which names u saved in untilled 2. Same input images in png form u r giving as input for matlab code

    • @Morteza_Sc
      @Morteza_Sc  11 місяців тому

      @@infinity3865 I’m not sure what you are asking exactly. But the inputs are png images in RGB FORMAT, but the output is hyperspectral images.

    • @infinity3865
      @infinity3865 11 місяців тому

      I am asking what input have you given to accessories

    • @infinity3865
      @infinity3865 11 місяців тому

      Accessories.m what files are there

    • @Morteza_Sc
      @Morteza_Sc  11 місяців тому

      @@infinity3865 I’m away on vacation now but when I come back I’ll have a look at the code and let you know! Thank you!!

  • @EnriqueAPOLOAPOLO
    @EnriqueAPOLOAPOLO Рік тому

    This is a big error. You are not obtaining actual data from this type of transformation. The idea when you are using a hyperspectral camera is to see those things that your eyes can not see. I mean, seeing in other regions of the the spectra.

    • @Morteza_Sc
      @Morteza_Sc  Рік тому

      It depends on the training and testing data actually. You can estimate the spectral reflectance at a very high accuracy depending on whether or not your training data and test data are of the same type ( material wise). There are papers out there that have shown the accuracy is actually very much acceptable :)

    • @EnriqueAPOLOAPOLO
      @EnriqueAPOLOAPOLO Рік тому

      @@Morteza_Sc, It could be. But at the end of your video you are showing just two RGB images. I did not see the spectral signature.

    • @Morteza_Sc
      @Morteza_Sc  Рік тому

      @@EnriqueAPOLOAPOLO yeah, that’s the sRGB image of the hyperspectral image. I plan to make a new video of this topic and make
      It contain some more useful things like the one you are saying.

    • @EnriqueAPOLOAPOLO
      @EnriqueAPOLOAPOLO Рік тому

      @@Morteza_Sc if you are able to show me the spectral signature from an actual hyperspectral image and then the spectral signature for an image generated by your method I will believe in your method ;).

  • @infinity3865
    @infinity3865 Рік тому

    can you please provide your email ID for sending an email to get the code?