How to import three dimensional hyperspectral data (.mat) into Deep learning toolbox?

I am using Deep Network Designer (DND). I have three folder, filled with .mat files. These .mat files are cropped hyperspectral images of dimension (X,Y, 214). Here X and Y are image dimension which vary with every image and 214 is the number of channels or bands and is constant in all the images.

I was having no problem in importing these data using the function imagedatastore. But, I am facing these two problems.

1. While importing the images is successful. I cannot view them on the DND dialog window.

2. When I try to train the model. I get an error that the input images need to be 3D.

If I open these individual matfiles using hyperspectralViewer. There is no problem.

Attached is the code that I have used to create the datastore and the network image.

Trainimds = imageDatastore("location of directory","FileExtensions",".mat","LabelSource","foldernames","IncludeSubfolders",true);

the network image

Here is a paper that does the same thing: https://www.researchgate.net/publication/347958844_Deep_learning_classifiers_for_near_infrared_spectral_imaging_A_tutorial

[parsa]

Did you solve it?

[Billy Ram]

No, not yet.

Answer 1

[Parth Parikh]

Hi Billy,

For the first issue, you can try using fileDatastore. As mentioned that you have 3 folders, contain all .mat files.

fds = fileDatastore(<path>, "ReadFcn", @load, "FileExtensions", ".mat"); 

For the second error,

It looks like you are passing 4D input to "image3dInputLayer".

image3dInputLayer([X Y 214 1], 'Name', 'Input');

If you wish to learn more about hyperspectral imaging, please have a look at the below links:

Hyperspectral Image Processing - MATLAB & Simulink (mathworks.com)

Image Processing Toolbox Hyperspectral Imaging Library - File Exchange - MATLAB Central (mathworks.com)