A Refined Hybrid Network for Object Detection in Aerial Images

Aerial object detection is a challenging task that needs to detect objects with large variations in scale and orientation. Previous dense object detectors rely on heuristic nonmaximum suppression (NMS) to filter out redundant detections. This may reduce the recall rate for objects with arbitrary orientations and large aspect ratios. Recently proposed sparse object detectors treat object detection as a set prediction task, effectively eliminating the need for hand-crafted components. However, applying this paradigm directly to aerial images achieves inferior performance. In this article, we develop an effective refined hybrid network (RHNet) for object detection in aerial images. Our method combines the advantages of both dense and sparse detectors, achieving outstanding performance for aerial objects with large variations. Specifically, considering the highly diverse orientations of objects, we first apply a dynamic query generation (DQG) module to produce high-quality oriented queries. These queries can effectively locate the foreground objects in an image, ensuring a high recall rate. Then, the object queries are sent to a query decoder for further refinement. This refinement stage adopts one-to-one matching to eliminate the negative impact caused by NMS. Moreover, an adaptive feature fusion (AFF) module is designed to learn stronger modeling capabilities for rotated objects at different scales. In addition, we propose a practical mixed query sampling (MQS) strategy that uses many-to-one assignment as an auxiliary scheme to help detector training. Extensive experiments conducted on several aerial datasets demonstrate the superior performance of the proposed method in comparison to other state-of-the-art approaches.