摘要
干涉相位滤波是干涉合成孔径雷达(InSAR, Interferometric Synthetic Aperture Radar)数据处理的关键,直接影响相位解缠与测量精度。近年来,深度学习方法在该领域表现出色,但现有模型在空间特征学习和边缘细节保留方面仍存在不足。为此,本文提出一种基于空间注意力机制(SAM, Spatial attention mechanism)的降噪卷积神经网络(DnCNN, Denoising Convolutional Neural Network)干涉相位滤波方法。该方法在DnCNN网络中引入空间注意力模块,以增强对图像空间特征的学习能力,从而提升干涉相位边缘细节的保留效果。实验结果表明,该方法在模拟与实测数据上的滤波效果优于现有方法,并有效改善了边缘细节的保持能力。
Interferometric phase filtering is a crucial step in Interferometric Synthetic Aperture Radar(InSAR)data processing,directly impacting phase unwrapping and measurement accuracy.In recent years,deep learning-based methods have demonstrated promising performance in this field.However,existing models still exhibit limitations in learning spatial features and preserving edge details of interferometric phases.To address these issues,this paper proposes a Denoising Convolutional Neural Network(DnCNN)-based interferometric phase filtering method enhanced with a Spatial Attention Mechanism(SAM).By incorporating the SAM module into the DnCNN architecture,the proposed method improves the network's ability to learn spatial features,thereby enhancing the preservation of interferometric phase edge details.Experimental results on both simulated and real datasets demonstrate that the proposed method outperforms existing filtering techniques and effectively improves edge detail retention.
作者
安世铭
李芳芳
An Shiming;Li Fangfang(Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing,China;Key Laboratory of Technology in Geo-spatial Information Processing and Application System Institute of Electronics,Chinese Academy of Sciences,Beijing,China;University of Chinese Academy of Sciences School of Electronic,Electrical and Communication Engineering,Beijing,China)
出处
《科学技术创新》
2025年第18期81-87,共7页
Scientific and Technological Innovation
基金
国家自然科学基金(62171435)资助课题。
作者简介
安世铭(1999-),男,硕士研究生在读,研究方向:干涉合成孔径雷达技术及图像处理。
