부산대학교 도서관

The rising popularity of small aircraft such as multi-rotor drones, and the security concerns linked to small aircraft create an increasing demand for systems that are able to detect and recognize such targets. An effective target recognition system should be able to recognize targets in a real-world setting, under conditions where the system receives noisy inputs. In addition to being able to handle noisy inputs, a target recognition system should be able to handle both known targets (targets that it has been trained on) and unknown targets (targets that do not form part of the training set). Open-set classification such as this is non-trivial, as it can be a challenging task to train a machine learning model to recognize inputs that do not belong to any class observed in training.