基于ControlVAE的约束嵌入稠密时变阵列构建  

Construction of Dense Time-Varying Array with Constraint Embedded Based on ControlVAE

作　　者：孙文力 原达[1] 姜新波 SUN Wen-li;YUAN Da;JIANG Xin-bo(Key Laboratory of Intelligent Information Processing,Shandong Technology and Business University,Yantai,Shandong 264005,China;School of Qilu Transportation,Shandong University,Jinan,Shandong 250002,China)

机构地区：[1]山东工商学院山东高校智能信息处理重点实验室,山东烟台264005 [2]山东大学齐鲁交通学院,山东济南250002

出　　处：《电子学报》2024年第1期354-363,共10页Acta Electronica Sinica

基　　金：国家自然科学基金(No.62072285);山东省重点研发计划(No.2019GGX101040)。

摘　　要：在对探地雷达(Ground Penetrating Radar,GPR)数据进行三维建模与可视化分析过程中,其所依赖的三维时变阵列通常是由一维数据道或二维阵列间接生成的.由于采集到的数据往往比较稀疏,且存在不规则偏移,需要进行稠密处理,以获得高密度三维时变阵列.本文提出了一种基于可控变分自编码(Controllable Variational AutoEncoder,ControlVAE)的约束嵌入稠密时变阵列构建方法 .使用基于ControlVAE的时变数据重构网络,通过隐空间采样插值和深浅层特征信息的融合,生成接近真实分布的伪道数据来增加阵列密度.构建了基于尺度不变特征变换的数据配准模块,提取道间梯度特征与结构对称性特征,可以在时空域内完成数据配准.同时将浅层特征作为约束信息嵌入数据重构网络,以消除数据道偏移对阵列生成的影响.实验结果表明,本文方法仅使用单测线二维B-scan数据集即可重建稠密时变阵列,可以有效降低重构模型累积误差,提高面对复杂真实数据时的鲁棒性.In the 3D modeling and visualization analysis of ground penetrating radar(GPR)data,the 3D time-varying arrays they rely on are usually generated indirectly from 1D data channels or 2D arrays.Since these data are often sparse and irregularly shifted,dense processing is required to obtain high-density 3D time-varying arrays.This paper proposes a constrained embedding dense time-varying array construction method based on controllable variational autoencoder(Con⁃trolVAE).A time-varying data reconstruction network based on the ControlVAE is used to increase the array density by generating pseudo-channel data close to the natural distribution through the fusion of hidden space sampling interpolation and deep and shallow feature information.A data alignment module based on scale-invariant feature transformation is con⁃structed to extract the inter-channel gradient features and structural symmetry features to complete the data alignment in the spatial domain.The shallow features are also embedded as constraint information in the data generation network to elimi⁃nate the influence of data channel offset on array generation.The experimental results show that the proposed method can efficiently reconstruct a time-varying array with a single scan direction using a two-dimensional B-scan data set.This meth⁃od can effectively reduce the cumulative error of the reconstructed model and improve the robustness in the face of complex actual data.

关 键 词：时变阵列 数据生成 探地雷达 约束嵌入 可控变分自编码模型 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]