基于多尺度注意力聚合的改进YOLOv11n小目标检测算法OA

Aimed at the problem that YOLOv11n is vulnerable to the loss of small target feature informa-tion subjected to the background interference in UAV aerial photography scenarios,this paper proposes an improved YOLOv11n small target detection algorithm.Firstly,a detection head is added to the high-resolu-tion feature layer of the YOLOv11n backbone network.The feature information of small-sized targets is preserved on the high-resolution feature map to alleviate the loss of feature information caused by down-sampling,and improve the small target detection accuracy.And then,a multi-scale attention aggregation module(MSAA)is introduced to enhance the multi-scale fusion effect of space and channel,and at the same time,reduce the background interference.Furthermore,InnerIoU Loss is adopted to replace the tradi-tional IoU Loss to significantly improve the model detection accuracy.The experimental results show that the improved algorithm achieves 38.302％mAP@0.5 and 22.345％mAP@0.5:0.95 on the VisDrone2019 dataset,and increases by 5.735％and 3.581％respectively compared with the baseline model.