Unknown Object Detection has garnered increasing attention due to its adaptability to open scenarios in the real world. However, previous methods have often struggled with differentiating between unknown objects and non-objects and made unreasonable selections for unknown predictions, resulting in inaccurate unknown detection. In light of this, drawing inspiration from known object detection, we propose an innovative method for unknown object detection called Unknown Object Aware RCNN (UOA-RCNN), which aims to tackle these aforementioned issues. Firstly, to address the challenge of distinguishing between objects and non-objects, we introduce the Unknown Object Aware Module. This module learns a Universal Objectness Score (UOS) using known objects, enabling it to generalize to unknown objects, significantly improving the discriminability between objects and non-objects. Subsequently, we incorporate the notion of the Known Object Probability to refine the identification of unknown objects, effectively suppressing potential non-objects. Finally, we design an innovative unknown object mining scheme based on the UOS. This scheme allows for the accurate localization of both known and unknown objects while removing redundant results during prediction. Through extensive experimentation, our method delivers state-of-the-art performance on the unknown object detection benchmark, outperforming other existing methods.