Lightweight YOLO-based framework for oriented vehicle detection in UAV imagery

Unmanned aerial vehicles (UAVs) provide flexible, low-cost, and wide-coverage sensing for modern intelligent transportation systems (ITS). However, detecting vehicles from UAV imagery remains challenging due to small object scales, dense distributions, complex backgrounds, and diverse orientations. To address these issues, this study proposes R-YOLO-DL, a lightweight and rotation-aware detection framework based on YOLOv11n-OBB. The model integrates a Lightweight Adaptive Extraction (LAE) module to enhance local feature representation with fewer parameters, and a Dynamic Multi-scale Oriented Detection Head (DynamicHeadOBB) to improve rotation modeling through scale-, spatial-, and task-aware attention. Experiments on the DroneVehicle dataset show that R-YOLO-DL achieves 1.6% higher mAP@50 and 5% lower GFLOPs with 16% fewer parameters than the baseline. Compared with advanced detectors such as Oriented R-CNN and ROITransformer, it offers a better balance between accuracy and efficiency. The proposed method demonstrates strong potential for real-time vehicle detection on resource-constrained UAV platforms.