How to get Spectrogram of Signal for Googlenet Deep Learning

Hello,

I have some signal dataset and i have created a CNN. if i'm not wrong, the input of my Googlenet CNN should be Spectrograms of signals. My dataset includes 5 different device, 400+ signal datas per each. I'd like to classify them but my cnn results seems bad and reason might be the input data. Are there any suggestion? Here is the code:

spectrogram(y,256,250,256,FS);

saveas(gcf,fileName); %Use to display save as image

RGB = imread(fileName);

imresize(RGB, [224, 224]);

imwrite(RGB,fileName);

thanks for help

Answer 1

[Meet]

Hi,

There are several improvements you can consider to enhance your classification results:

1. Normalize your signals before generating spectrograms to ensure consistent input data.
2. You can use "imresize" with an interpolation method like "bilinear" or "bicubic" for better quality.
3. Use data augmentation techniques to increase the diversity of your training data, some of those techniques include Time Stretching, Pitch Shifting, Noise Addition etc, you can use "imageDataAugmenter" for augmenting the data.
4. Fine-tune the GoogleNet model by training only the last few layers initially, then progressively unfreeze more layers.

Hope this helps!