I have a hyperspectral image on which I have performed PCA and now intend on using the output PCA components as an input into a classification.

This is the first time I have done PCA, am i correct in thinking that I need to use the output of the Inverse PCA as input into the classification?

The Inverse PCA effectively recreates the original image from the PCA components and therefore has spectral properties, unlike the first outputs from the original PCA.

I presume people have done something similar - am I correct in using the Inverse PCA outputs for classification?

1 Answer 1


The whole point of doing a PCA on a hyperspectral image is to reduce the number of input layers to your classification in order to avoid overfitting your model. As such, you either want to reduce the dimensionality, or work on your PCA-elements.
The only reason to apply the inverse PCA is if you have a need to work to physical values, if you for example were using real world end-members in your classification.

All in all, what you want is a reduced number of input bands to your classification. That can be done by only working on the PCA-elements with the highest eigen values, or by applying the PCA where you create fewer elements than you have bands, and then inverse that PCA.

  • Thanks, I misunderstood slightly I think. I'm using endmembers from the image, so I'm going to use the first 5 PCA's (99.54% of variance) as input into the classification.
    – sm29
    Commented Jul 20, 2017 at 21:17
  • @sm29 - if my answer solved your problem, please mark it as 'Accepted' by clicking on the tick-mark under the up- and down-vote buttons. Commented Jul 21, 2017 at 8:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.