A Threshold Insensitive Open-Set Recognition Scheme for AAV Targets Based on HRRP

In the field of radar target recognition, open-set recognition can be used to solve noncooperative target recognition. The main difficulty of open-set recognition is finding a closed classification boundary to distinguish the known and unknown targets simultaneously. This article proposes an open-set recognition method that trains a neural network through a distance-based loss function and combines the OpenMax classifier, which solves the open-set recognition problem of finding the closed boundary. With this method, the known and unknown classes can be effectively in various sample sets identified without relying on a prior threshold to assist in searching boundaries. In addition, simulation results show that the rejection accuracy exceeds 95% for eight types of autonomuos aerial vehicle (AAV) targets based on high-resolution range profile, which indicates excellent performance for open-set recognition.