Hi,

In this case, you can applied the 3-D Convolutional Neural Netwok (3-D CNN) in order to preserve the spatial and spectral dimensions of HSI (pixels neighbors and number of bands). Therefore, you can use the mdCNN MATLAB toolbox.

https://fr.mathworks.com/matlabcentral/fileexchange/58447-hagaygarty-mdcnn

You can use this code based on this toolbox:

%%%%%%%%%%%%%%%%%%%%% Input Setting%%%%%%%%%%

net.hyperParam.sizeFmInput = [5 5 128]; % size of input feature map

net.hyperParam.numFmInput = 1; % number of feature map of the input

%%%%%%%%%%%%%%%%%%%%% Layers specification %%%%%%%%%%

net.layers{end+1}.properties = struct('type',2,'numFm',7 , 'Activation',@Relu, 'dActivation',@dRelu,'kernel',5,'pad',2, 'stride', [2 2 4], 'pooling', [1 1 1]);

net.layers{end+1}.properties = struct('type',2,'numFm',17 , 'kernel',[5 5 3] ,'pad',[1 1 0], 'pooling', [1 1 1] , 'dropOut' ,0.8);

net.layers{end+1}.properties = struct('type',1,'numFm',128);

net.layers{end+1}.properties = struct('type',1,'numFm',10);

First create a network using a config file

net = CreateNet('HSI_classify.conf');

Then, call ‘Train’ function with the dataset containing the train/test samples:

net = Train(HSI,net, 3000);

Here , HSI is the dataset , this will train for 3000 samples from the test set .

GOOD LUCK

This publications may help

Chapter Spectral Reflectance Images and Applications

Article Spectral imaging method for material classification and insp...