Dealing with observing conditions
Milestone ID: 2056
Previous studies [22] showed that γ-PhysNet is sensitive to the changing of observation conditions such as Night Sky Background (NSB), stars in the field of view, mis-calibrations, atmospheric conditions and other differences that could arise between MC and real data. Domain adaptation techniques [23] are currently being developed by Micha ̈el Dell’aiera as part of his PhD to overcome these issues. Very promising preliminary results have been presented internally to the LST collabo- ration and will very likely lead to a proof-of-concept solution. This research work will then need to be turned into a viable product that includes the constraints of the real-time, which is the objective of this task. The domain adaptation techniques currently developed by M. Dell’aiera involve an offline training using real data corresponding to observation conditions. However, in the case of a real-time application, the training cannot be conducted simultaneously with the observation under the real-time constraints (due to the high data flow from LST-1 and the risks of mistraining without validation), and an alternative strategy must be applied.