Rotated YOLOv4 with Attention-wise Object Detectors in Aerial Images

Abstract

∗There exist plenty of object detection applications in the field of remote sensing. However, some challenges appear especially for small and dense objects when the image is overlarge or the image with complex background. Thus, we propose a rotated attention wise network with an accuracy-speed balanced real-time objector. Learnable attentionwisemodules are adopted in the networkwith rotated bounding boxes for searching, locating right semantic and features simultaneously. In the process of training, various backbones and data augmentation strategies are employed to achieve higher accuracy and more types of objects. The results that conducted on extensive comparable experiments demonstrate the effectiveness of the proposed framework, achieving state-of-the-art performance in real-time object detection methods.