How to describe/compare the feature values extracted from different types (classes) of image samples

Shirui · October 24, 2019, 5:40pm

There are two classes A, B, all samples of N = N_A + N_B.
And extracted 500 feature values for each sample, as Feature of each of them. As below:

For the Feature of each of N = N_A + N_B, what kind of mathematical description can describe/discuss it well?

At present, the method I’m thinking of is, after extracting all the image sample feature of class A and class B, add the feature values by class and then average them, that is,
$feature_{A, i} = \frac {1}{N_A} \sum _{a=1} ^{N_A} feature_{a, i}, feature_{B, i} = \frac {1}{N_B} \sum _{b=1} ^{N_B} feature_{b, i}$
Then for the feature_A and feature_B, each PCA (and similar value) obtained is a feature that represents the whole class. From the perspective of feature engineering, can such a method work?