In my understanding, each hidden layer in neural network extract the implicit information of the features. So, can we know what kind of information does hidden layers process and how to understand them?
Yes (especially vision problems) and people have done that. We also had good coverage in here.
See @Chaconne's answer for this question
I also had an another there: in other domain, and when we have a large number of parameters, many people stop focusing on what are these implicit features, and the importance of the features.