you can start with simple PCA first and see how that performs.

what is your end goal? classification?

Thanks ,,,, my end goal is development of regression model.

Well PCA is an option but you could also start with stepwise Linear Regression. The relevant(significant) spectral bands or predictive variables are identified and used for regression. The significant spectral data is chosen based on p-values (null hypothesis)

Did you try the Random Forest? Their feature importance method is also useful.

did you try cross entropy? but training may not be easy, if arrays are dense and compact.

will try both the methods..thanks

Parrallel LSTm, or SVM ?