北京测绘2025，Vol.39Issue(8)：1123-1128,6.DOI:10.19580/j.cnki.1007-3000.2025.08.006
一种改进U-Net模型的无人机影像道路提取方法
An improved U-Net model for road extraction from UAV imagery
摘要
Abstract
To address the issues of low automation,information loss,and discontinuity at complex intersections in unmanned aerial vehicle(UAV)imagery road extraction,this paper proposed an improved U-Net model for road extraction,which can effectively enhance the accuracy of road recognition.First,the improved U-Net model integrated the Res2Net structure,replacing the traditional U-Net convolution layers,which enhanced feature extraction precision and improved the sampling depth.Second,the model added the convolution block attention module(CBAM)attention mechanism,recalibrat-ing both spatial and channel layers,and further improved model performance through fine-tuning of parameters.Finally,the model incorporated the improved dense atrous spatial pyramid pooling(DenseASPP)module,which achieved the stitching of lower-level details,thereby strengthening the ability to capture contextual information of road areas.Experimental valida-tion demonstrated the effectiveness of the proposed model.The results show that the accuracy,recall rate,F1 score,and intersection over union(IoU)of the road extraction from UAV imagery are 90.18%,85.85%,87.96%,and 73.56%,respectively,outperforming the comparison models and reflecting the superiority of the proposed model.关键词
无人机影像/道路提取/深度学习/注意力机制Key words
unmanned aerial vehicle(UAV)imagery/road extraction/deep learning/attention mechanism分类
天文与地球科学引用本文复制引用
尧燕,张恒僖..一种改进U-Net模型的无人机影像道路提取方法[J].北京测绘,2025,39(8):1123-1128,6.基金项目
江西省自然科学基金(20232ACB204032) （20232ACB204032）
