高分辨率遥感图像双解耦语义分割网络模型  

Dual decoupling semantic segmentation model for high-resolution remote sensing images

作　　者：刘帅[1,2] 李笑迎 于梦 邢光龙 LIU Shuai;LI Xiaoying;YU Meng;XING Guanglong(School of Information Science and Engineering,Yanshan University,Qinhuangdao 066004,China;Hebei Key Laboratory of Information Transmission and Signal Processing,Qinhuangdao 066004,China)

机构地区：[1]燕山大学信息科学与工程学院,河北秦皇岛066004 [2]河北省信息传输与信号处理重点实验室,河北秦皇岛066004

出　　处：《测绘学报》2023年第4期638-647,共10页Acta Geodaetica et Cartographica Sinica

基　　金：国家自然科学基金(61671401);河北省自然科学基金(F2020203099)。

摘　　要：语义分割是高空间分辨率遥感图像分析和理解的核心内容之一。现有基于深度学习的语义分割网络会导致遥感图像高频信息损失,边界分割不准确。针对此问题,本文提出一种双解耦语义分割网络模型,将提取的两级特征图解耦为具有高频特性的边界特征和具有低频特性的主体特征,并将解耦后的边界和主体特征图进行融合,从而改善高分辨率遥感图像语义分割性能。进一步提出了一种顾及边界和主体的损失函数,对地物要素及其边界和主体部分进行优化学习。在ISPRS Vaihingen和Potsdam 2D高空间分辨率遥感图像数据集上进行试验,与已有的遥感图像语义分割网络模型结果比较,双解耦语义分割网络模型能有效提高地物要素分割精度。Semantic segmentation is one of the core contents of high spatial resolution remote sensing images analysis and understanding.The existing semantic segmentation network based on deep learning will lead to the loss of high-frequency information and inaccurate edge segmentation of remote sensing images.Aiming at this problem,this study designs a dual decoupling semantic segmentation network model to improve the semantic segmentation performance of high-resolution remote sensing images.The extracted two-level feature maps are decoupled into edge features with high-frequency characteristics and body features with low-frequency characteristics,and the decoupled edge and body feature maps are fused.Furthermore,a loss function considering edge and body is proposed to optimize the ground feature elements.Experiments on ISPRS Vaihingen and ISPRS Potsdam 2D high spatial resolution remote sensing image datasets.Compared with the results of the existing remote sensing images semantic segmentation network model,the dual decoupling semantic segmentation network model can effectively improve the segmentation accuracy of ground feature elements.

关 键 词：高分辨率遥感图像 语义分割 双解耦网络 深度学习 特征融合 

分 类 号：P237[天文地球—摄影测量与遥感]