Deep learning techniques have shown the capability to identify the degree of energy loss of high-energy jets traversing hot QCD medium on a jet-by-jet basis. The average amount of quenching of quark and gluon jets in hot QCD medium actually have different characteristics, such as their dependence on the in-medium traversed length and the early-developed jet substructures in the evolution. These observations motivate us to consider these two types of jets separately and classify them from jet images with deep learning techniques. We find that the classification performance gradually decreases with increasing degree of jet modification. In addition, we discuss the predictive power of different jet observables, such as the jet shape, jet fragmentation function, jet substructures as well as their combinations, in order to address the interpretability of the classification task.