An interpretable hierarchical framework for rapid UAV type recognition based on multi-sensor data and knowledge

UAV type recognition is the key technology in many applications, such as aviation management, infrastructure safety, and military defense. With the increasing deployment of large-scale heterogeneous sensors, how to recognize UAV type in the accurate, rapid, and even interpretable style has become an important issue. Inspired by bionic cognition, an interpretable hierarchical framework is established based on domain knowledge, experience, and rule constraints. Initially, based on the logic of ‘gradually understanding’, a ‘rough to fine’ collaborative data processing approach is designed to achieve fast and stable reasoning with large-scale multi-sensor data. Futhermore, inspired by the human cognitive habit of ‘deepening knowledge’, an adaptive weight adjustment method is proposed for recursive evidence fusion. In typical simulation scenarios, the method reduces computational time while achieving an average classification accuracy exceeding 90%.