How to transform non image data so that it will be able to train with Matlab CNN ?

For example if I am having 10 non image data point and its 10 corresponding label. Each datapoint is a 4x12 matrix where the matrix element is some small non negative number (for example 1.32E-05 -2.74E-06 -6.65E-06 ).

What would be the appropriate way to transform these input so that I could work with Matlab CNN ?

Also, what kind of input layer should be use for this task ?

Should I turn them into gray scale image or not ?

Answer 1

[Uday Pradhan]

Hi Tuong,

You can use the "imageInputLayer" to input your data points into your neural network. You can take:

inputSize = [4 12 1]; %for a single data - point, you may use more than 1 channels to stack input data
layer = imageInputLayer(inputSize,'Normalization','rescale-zero-one');

This will normalize the elements of each data - point to be in the interval [0,1]. Try this normalization and see the results, if not then you can try other normalization options as well. Please refer to this link to know more.

Hope this helps!