A High-Precision Object Detection Algorithm for UAV Aerial Images: WI-YOLO

To tackle the challenges of low small-target detection accuracy, intense back-ground interference, and high model computational complexity in unmanned aerial vehicle (UAV) aerial images, this study presents an enhanced lightweight object detection algorithm, namely WI-YOLO. Built on the YOLOv11n architec-ture, the algorithm integrates two key improvements: first, an improved effi-cient multi-scale attention mechanism (iEMA) is embedded into the backbone network, which strengthens the discriminability of low-contrast targets by fus-ing standard deviation and local contrast features; second, a wavelet transform convolution module (WTConv) is designed to expand the receptive field and preserve multi-scale structural information via a multi-level frequency-domain decomposition and reconstruction mechanism. Experimental evaluations on the DOTA dataset show that WI-YOLO achieves 61.5% in mAP@50 and 35.1% in mAP@50:95—marking significant gains over the baseline model. Concurrently, the model’s parameter count is reduced to 5.9 MB, and its inference speed is increased to 30.5 FPS. These results demon-strate the coordinated optimization of detection accuracy and computational efficiency, making WI-YOLO well-suited for real-time detection tasks on UAV embedded platforms.