SVM classification of multispectral data

[AdamL]

Hello everyone, 

 I am using Support Vector Machine (SVM) classifier on ASTER and Sentinel-2 datasets to map geological units. I am familiar with the basic statistical concept of SVM classifier, but in more practical terms, how does it use the spectral data from the satellite imagery to classify different classes? How does it determine the most accurate fit from the image-based training dataset? 

Thank you to all of you,
Cheers!