计算机工程与应用2025，Vol.61Issue(14)：135-147,13.DOI:10.3778/j.issn.1002-8331.2412-0332
跨尺度特征融合的无人机小目标检测算法
Small Target Detection Algorithm for UAV Based on Cross-Scale Feature Fusion
摘要
Abstract
Aiming at the existing problems of multi-scale,small target,complex background interference and easy occlu-sion in UAV small target detection tasks,an improved YOLOv11n model based on composite feature fusion and cross-scale optimization is proposed.In the backbone network,a multiscale perceptual cascade attention(MPCA)mechanism is proposed to improve the convolutional module,which addresses the lack of traditional convolutional feature expression ability,and significantly improves the feature extraction ability of the network at a lower computing cost.A new efficient multi-scale FPN(EMSFPN)structure is proposed to improve the neck network,enabling mutual integration of features from different levels.On the basis of improving the neck network model,a feature layer with rich semantic information of small targets is added.The selective boundary aggregation(SBA)module is used for interactive fusion of multi-resolution features to improve the multi-scale processing capability of the model.The Inner-IoU loss function is introduced to enhance the Wise-IoU function by replacing the original loss function with Inner-WIoU,improving the positioning accuracy of small targets,and optimizing the calculation of loss value.The improved YOLOv11n algorithm has a 9.8%reduction in the number of parameters compared with the original model on the VisDrone2019 data set,and a significant 9.1 percentage points improvement in mAP50.The performance exceeds that of YOLOv11s,and the performance is greatly improved while the model is lightweight.关键词
YOLOv11n/无人机(UAV)/小目标检测/多尺度特征融合/Inner-WIoUKey words
YOLOv11n/unmanned aerial vehicle(UAV)/small target detection/multi-scale feature fusion/Inner-WIoU分类
信息技术与安全科学引用本文复制引用
罗显志,汪航..跨尺度特征融合的无人机小目标检测算法[J].计算机工程与应用,2025,61(14):135-147,13.基金项目
湖北省自然科学基金面上项目(2024AFB933). （2024AFB933）
