You could certainly do that, but you would need to create the training dataset. It would require many thousands of photos of real and counterfeit bags up close.

You could certainly do that, but you would need to create the training dataset. It would require many thousands of photos of real and counterfeit bags up close. Image recognition has been mostly solved by ML techniques, so if you just assemble the data and train the network you can distinguish any set of things that are significantly different visually.