Vehicle Object Detection Algorithm for UAV Aerial Imagery Based on Improved YOLOv11-OBB

To address the problems of numerous parameters and high computational load in vehicle detection algorithms under drone aerial photography scenarios, this paper proposes a lightweight vehicle detection algorithm based on an improved YOLOv11-OBB. First, the GhostConv module is used to replace standard convolutions in the original network; then, the LSKAM attention mechanism is added in the Neck; finally, the Focaler-IoU loss function replaces CIoU. With these improvements, the model's parameter size and computational complexity are significantly reduced, while its adaptability to vehicles of different orientations and aspect ratios is enhanced, along with improved bounding box regression accuracy and model convergence speed. Experiments on the Drone-Vehicle dataset show that the improved model reduces computational load by 29.54% while increasing mAP50-95 to 88.6% and achieving an FPS of 425.4, realizing a better balance between detection accuracy and inference efficiency for vehicle detection from drone perspectives.