摘要
为了提高遥感影像水体提取的边界信息和细节信息,提出了一种基于边界增强的多尺度特征融合的高分辨率遥感影像水体提取模型(PANFormer)。首先,采用路径聚合网络中的自底向上的多尺度特征融合路径部分,结合瓶颈注意力模块,通过通道和空间注意力对多尺度特征图融合和增强,引导模型聚焦水体关键区域。其次,在瓶颈注意力模块中引入以Sobel滤波器为核心的边界感知注意力,加权边缘特征图显著提高模型对细粒度边界的感知能力。最后,在预处理后的LoveDA数据集上进行对比和消融实验,结果表明该研究方法在提取复杂背景的水体边界上展现了更强的鲁棒性和准确性。
In order to improve the boundary information and detail information of remote sensing image water body extraction,a high-resolution remote sensing image water body extraction model(PANFormer)based on boundary enhancement and multi-scale feature fusion is proposed.Firstly,the bottom-up multi-scale feature fusion path part of the path aggregation network is used,combined with the bottleneck attention module,to fuse and enhance the multi-scale feature map through channel and spatial attention,and guide the model to focus on key areas of water body.Secondly,the boundary-aware attention with Sobel filter as the core is introduced into the bottleneck attention module,and the weighted edge feature map significantly improves the model′s perception ability of fine-grained boundaries.Finally,comparison and ablation experiments are carried out on the preprocessed LoveDA dataset.The results show that the research method shows stronger robustness and accuracy in extracting water boundaries with complex backgrounds.
作者
熊庭珏
XIONG Tingjue(Jiangxi University of Science and Technology,Ganzhou 341000,China)
出处
《长江信息通信》
2025年第9期21-24,共4页
Changjiang Information & Communications
关键词
水体提取
边界感知
高分辨率遥感影像
多尺度特征融合
Water body extraction
Boundary sensing
High-resolution remote sensing images
Multi-scale feature fusion
