How to remove clouds from an image using a neural network?

Question

Neural networks

How to remove clouds from an image using a neural network?
We were given the task of making an application that removes clouds from images, for example, satellite images and aerial photographs, I can figure out the code, but it doesn’t occur to me how to solve the problem. Many questions immediately come to mind, for example:

• How to edit a photo
  
• How to recognize clouds
  
• How to replace clouds with terrain below them
  
• What if the clouds are too dense
  

Explain the principle of operation please

PS For training, I can find 10k different photos of the same area

Answer 1

Do you have ten thousand PAIRS of pictures of the same area with and without clouds, in order to train your neural network on them?
If there is - then everything is trivial - a neural network of the autoencoder-decoder type. At the input a picture with a cloud, at the output we compare it with the same area without clouds. The network is trained on your tens of thousands of pairs.
We feed new pictures to the grid trained in this way and hope that it will be able to cope with them.
If there is no such set, then we start by reading books and understanding how neural networks work in general.
PS No need for 10 thousand photos of "the same area". We need 10 thousand PAIRS, each pair represents its own area, but in a pair - one photo with clouds, and the other without. Still, for a better understanding, I advise you to start by studying the principles of the neural network.
But if there is such a dataset of pairs, the autoencoder-autodecoder network should help you.
PPS Thinking - one more addition. Better, of course, not a couple of photos, but for EACH area - one photo without clouds and several photos with different clouds.
And one more thing - photos of the same area taken with a large time interval (two or three months) cannot be used. Seasons, changes in buildings, etc., will make the result very noisy.