R-YOLOV4: Enhancing Unmanned Aerial Vehicle Target Detection Through Rotational Object Detection and Feature Alignment

In response to the challenges of target detection from the perspective of unmanned aerial vehicles (UAVs), we propose the R-YOLOV4 Rotational Object Detection Algorithm. Specifically, R-YOLOV4 replaces horizontal detection boxes with rotational detection boxes to improve target shape conformity and detection performance. Furthermore, R-YOLOV4 replaces the backbone network to enhance the capability of detecting small targets. It also incorporates feature alignment structures and orientation detection structures into the detection head to improve the alignment and classification-regression tasks for rotational objects and anchor boxes. Experiments conducted on the DOTAv1.0 dataset demonstrate that the R-YOLOV4 method exhibits excellent performance in terms of both detection efficiency and accuracy.