王杰, 柳稼航, 凌信鹏, 段则先. 基于深度学习的InSAR影像局部与非局部联合相位滤波[J]. 武汉大学学报 ( 信息科学版). DOI: 10.13203/j.whugis20240052
引用本文: 王杰, 柳稼航, 凌信鹏, 段则先. 基于深度学习的InSAR影像局部与非局部联合相位滤波[J]. 武汉大学学报 ( 信息科学版). DOI: 10.13203/j.whugis20240052
WANG Jie, LIU Jiahang, LING Xinpeng, DUAN Zexian. Deep Learning-Based Joint Local and Non-local InSAR Image Phase Filtering Method[J]. Geomatics and Information Science of Wuhan University. DOI: 10.13203/j.whugis20240052
Citation: WANG Jie, LIU Jiahang, LING Xinpeng, DUAN Zexian. Deep Learning-Based Joint Local and Non-local InSAR Image Phase Filtering Method[J]. Geomatics and Information Science of Wuhan University. DOI: 10.13203/j.whugis20240052

基于深度学习的InSAR影像局部与非局部联合相位滤波

Deep Learning-Based Joint Local and Non-local InSAR Image Phase Filtering Method

  • 摘要: 相位滤波是干涉合成孔径雷达(interferometric synthetic aperture radar,InSAR)数据处理的关键技术之一,干涉图质量显著影响后续处理的精度。传统方法滤波效果差且效率低下,深度学习方法潜力大但目前精度受限。为有效提升滤波效果,本文结合卷积神经网络(convolutional neural network,CNN)与Transformer网络的优势,提出了一种局部与非局部联合的InSAR影像相位滤波方法。首先,基于Transformer全局特征提取能力强的特点,提出了面向相位滤波的PFCT网络结构。该网络可以同时进行局部滤波与非局部滤波,避免了现有方法因忽视干涉条纹非局部特征而导致的精度受限问题。其次,提出了一种新的复合损失函数用于同时训练网络的去噪能力和条纹结构保持能力。在仿真实验中,本文方法的均方误差(mean square error,MSE)比次优算法降低了15.5%,结构相似性(structural similarity,SSIM)指标提高了5.3%。在实测实验中,本文方法的残点去除性能比次优方法提高了1.8%。结果表明,本文方法的滤波性能优于目前的主要算法,能获得更好的效果。

     

    Abstract: Objectives: Phase filtering is one of the key technologies in interferometric synthetic aperture radar (InSAR) data processing, and the quality of interferograms significantly affects the accuracy of subsequent processing steps. The traditional phase filtering methods with manual parameter adjustment not only have low filtering accuracy but also low computational efficiency. Deep learning-based phase filtering methods have great potential, but current convolutional neural networks (CNN) based methods overlook the non-local features causing limiting accuracy. Transformer-based networks have strong global modeling capabilities and can extract non-local features from interference fringes, but they struggle with handling local features of interference fringes. Methods: To effectively improve the filtering effect, this paper proposes a joint local and non-local phase filtering method for InSAR images combined with the advantages of convolutional neural network (CNN) and Transformer network. First, based on the strong global feature extraction capability of the robust Transformer network, a PFCT network structure is proposed for phase filtering. The network performs both local and nonlocal filtering at the same time, avoiding the accuracy limitation problem caused by the neglect of nonlocal features of interference fringes in the existing methods. Then, a new complex loss function is proposed to guide the network to maintain the integrity of interference fringes while improving filtering performance. Results: In the simulation data experiment, the mean square error index was 15.5% lower than the suboptimal algorithm, and the structural similarity index was 5.3% higher. In the true data experiment, the residue removal index was 1.8% higher than the suboptimal method. The experiments show that the PFCT network model only slightly reduces computational efficiency, but achieves better filtering results than other methods, effectively maintaining the stripe structure while maintaining the filtering effect. Conclusions: The proposed method has a great filtering effect and powerful generalization ability on InSAR phase filtering.

     

/

返回文章
返回