Hi @qimin, welcome to the Forum!
This will probably give very poor results because quantum computers and simulators at the moment don’t work well with large amounts of data. A greyscale image has one numerical value per pixel while RGB images have three numbers per pixel and thus require many more resources. For hyperspectral images it would be even worse.
If you really want to try it my recommendation would be to first use an existing model, ideally on a small greyscale dataset and once you can run it then try to adapt it.
If you’re interested in learning quantum computing in general I’d recommend starting with the PennyLane Codebook.
Let us know what you think of the Codebook!