Converting 1d signal to 2d image

Hello,

I would like to ask how can I convert 1d signal to timeseries image, which will be used as CNN input?

Atatched schema is presenting this operation.

Answer 1

[Abderrahim. B]

Hello!

If i understood your question properly, you want to convert time series (signal is type of time series) to like-image to train CNN. See below:

% Generate signal
Fs = 2500 ;
F = [40 70 400]

F = 1×3

40 70 400

t        = 0:1/Fs:2-1/Fs ; % 2 seconds
noisySig = 2.5*sum(cos(2*pi*F.'*t)) + randn(1, length(t)) ; % noisy signal

% convert to spectrogram
spectrogram(noisySig, Fs)

hope this helps

[Image Analyst]

The how do you get the image from the spectrogram axes? Use getimage or getframe?

Answer 2

[Image Analyst]

How about using reshape to make it into a 2-D matrix of the proper aspect ratio? Then use imresize to size it to the size your network requires. Like to get 227 rows from your signal:

reshapedSignal = reshape(originalSignal, 227, []);
reshapedSignal = imresize(reshapedSignal, [227, 227]); % 227 for Alexnet

[PRASANTA KUMAR MOHANTY]

Dear Image Analyst,

Will the reshaping of the original signal retain the all the information of the original one?

[Image Analyst]

Yes. No information is lost.