How would you train a CNN to accurately predict the output of say the swt2 or fft2 function on an image?

Does anyone have any documentation, theories, code, methods or achitecture to train a CNN to accurately predict the output of say the swt2 or fft2 or any wavelet transform for a given image?

I'd be looking to train the CNN with a dataset of images as the X and the pre-processed images (swt2(ImageX)) as the Y value to predict an unknown image's wavelet transform output with high accuracy,

I'm just looking to gain an understanding of neural network's for less cookie-cutter problems but I'm struggling to find much documentation,

any responses are appreciated,

thanks.

Answer 1

[yanqi liu]

yes，sir，may be make the input as vector，and use convolution2dLayer to get one dimension data，then define cnn model，such as

imageInputLayer([vector_length 1 1])  
convolution2dLayer([100 1],3,'Stride',1)

if possible, may be upload your data and preprocess method.