现代电子技术2025，Vol.48Issue(11)：1-7,7.DOI:10.16652/j.issn.1004-373x.2025.11.001
基于深度学习的拒止环境下无人机自主定位方法
Deep learning based autonomous localization method for unmanned aerial vehicles in denial environments
摘要
Abstract
UAVs(unmanned aerial vehicles)rely on satellite systems for positioning.When satellite signals are covered or interfered with,the positioning of UAVs may be affected,which in turn leads to the inability to fly normally.Vision-based techniques can realize UAV localization by image matching methods,but the heterogeneous source images are so different that the existing feature matching methods fail to meet the requirements in terms of robustness and real-time performance.Therefore,a method named FCN-FPI(FocalNet-find point image),which integrates geographical positioning,is proposed.In the method,the Transformer-based FocalNet network is taken as the backbone network for multi-scale feature extraction first,and then a CLMF(cross-level multi-feature)fusion module is designed for fusing spatial semantic information of feature maps with different scales.Finally,the loss function is improved and a Gaussian window loss is proposed,so that different regions of the detection point are given corresponding weights according to their importance,which guides the model to pay more attention to the center region and improves the geolocation performance.The off-line experiments on the dense dataset UL14 show that the positioning performance on the meter-level index MA@20 is improved from 67.28%to 69.58%,and the relative distance score(RDS)is improved from 65.33%to 69.74%,so the proposed method has a certain engineering value.关键词
无人机/拒止环境/图像匹配/Transformer/特征融合/地理定位Key words
UAV/denial environment/image matching/Transformer/feature fusion/geographical positioning分类
电子信息工程引用本文复制引用
姚雨雯,郑恩辉,孙澄,杨健行,高翔..基于深度学习的拒止环境下无人机自主定位方法[J].现代电子技术,2025,48(11):1-7,7.基金项目
浙江省教育厅科研项目(Y202353673)资助 （Y202353673）
