They catch them with statistics! Although some people say they don't interact with matter, the correct statement is that they interact very weakly. Put another way, the probability of interacting is very low. To overcome this, experiments utilize enormous detectors and enormous numbers of neutrinos. Speaking very loosely, this means that even though the probability of an interaction per neutrino, per little bit of detector, per second is very small, when you multiply by a huge number of neutrinos, a huge amount of detector, you can see interactions if you wait long enough. One of the challenges is that, when you have a huge, sensitive detector and wait a long time, you can also see many other things, like radioactive background, cosmic rays, or miscellaneous other stuff that might arise from the neutrino beam production. So, while you are busy waiting for neutrino events to accumulate, you study all these other things to prove you understand what these so-called background events are, prove you understand what they look like and how often they occur, so that you can subtract or separate them statistically from the neutrino events.