摘要
针对多模态遥感影像建筑提取任务,本文提出了一种三分支多层级特征融合网络(TMFFNet)。基于光学影像与合成孔径雷达(SAR)影像在建筑物表征方面具有的互补特性,该网络设计了自适应模态融合模块(AMFM)和跨层级特征融合模块(CFFM),实现了跨模态特征的有效交互与空间特征优化。试验结果表明,本文算法在云覆盖与非云覆盖场景下的建筑物提取精度均优于现有主流模型。通过系统性试验,进一步验证了AMFM模块在跨模态特征选择方面及CFFM模块在跨层级特征融合方面的有效性。
This study proposes a triple-branch multi-level feature fusion network(TMFFNet)to address building extraction tasks in multimodal remote sensing imagery.By leveraging the complementary characteristics of optical images and synthetic aperture radar(SAR)images in building representation,the network introduces an adaptive modality fusion module(AMFM)and a cross-hierarchical feature fusion module(CFFM)to achieve effective cross-modal feature interaction and spatial feature optimization.The AMFM adaptively selects and integrates discriminative features from optical and SAR modalities,while the CFFM enhances multi-scale spatial information through hierarchical feature aggregation.Experimental results demonstrate that the proposed TMFFNet outperforms existing state-of-the-art methods in building extraction accuracy under both cloud-covered and cloud-free scenarios.Systematic experimentation further confirm the effectiveness of the AMFM in cross-modal feature fusion selection and the CFFM in optimizing multi-level spatial feature fusion representation.
作者
秦永志
朱艳
霍爱梅
庞燕
李小芳
QIN Yongzhi;ZHU Yan;HUO Aimei;PANG Yan;LI Xiaofang(Troops 61206,Beijing 100042,China;Information Engineering University,Zhengzhou 450001,China)
出处
《测绘通报》
北大核心
2025年第S1期174-178,共5页
Bulletin of Surveying and Mapping
关键词
建筑提取
多模态遥感
自适应模态融合
跨层级特征融合
building extraction
multimodal remote sensing
adaptive modality fusion
cross-hierarchical feature fusion
作者简介
秦永志(1975-),男,高级工程师,主要研究方向为遥感测绘及地理信息系统。E-mail:cyxhn@126.com
