Data Reconstruction Based on a Multi-branch Deep Neural Network

DOI

10.2991/978-94-6463-728-1_72How to use a DOI?

Keywords

Structural health monitoring; compressive sensing; transformer; CNN; data reconstruction

Abstract

Structural health monitoring systems measure structural deformations, vibrations, stress, etc., in real-time through sensor networks, comprehensively assessing the health of structures. The quality of monitoring data directly influences the accuracy of structural assessments and maintenance decisions. However, data loss is inevitable during the long-term monitoring of large structures. To address this issue, this paper proposes a Compressive Sensing-Convolutional Transformer Networks (C-CTNet) based multi-channel data reconstruction method. The neural network model comprises sampling, feature embedding and transformation, and reconstruction modules. The sampling module utilizes a mask matrix to sample the response matrix. The sampled matrix is then converted into a high-dimensional feature matrix in the feature embedding, and transformation module and segmented into submatrices. During the process, information is extracted, and dimensionality is reduced by learned convolution kernels. In the reconstruction phase, features are projected into a linear initialization module, CNN stem, and transformer stem, respectively. The initialization module mimics traditional compressive sensing reconstruction but generates initial reconstructions in a learnable and efficient manner. The CNN and transformer stem compute and effectively integrate local and long-range features. Additionally, a progressive strategy and window-based transformer model blocks reduce parameter count and computational complexity. Experimental results demonstrate that the proposed neural network model can effectively reconstruct data under random loss conditions. Comparisons with ablation models confirm the critical role of the three branches in the task.

Copyright

© 2025 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Hongyun Tian
AU  - Xiaomin Huang
PY  - 2025
DA  - 2025/05/19
TI  - Data Reconstruction Based on a Multi-branch Deep Neural Network
BT  - Proceedings of the 3rd International Conference on Green Building, Civil Engineering and Smart City (GBCESC 2024)
PB  - Atlantis Press
SP  - 771
EP  - 783
SN  - 2352-5401
UR  - https://doi.org/10.2991/978-94-6463-728-1_72
DO  - 10.2991/978-94-6463-728-1_72
ID  - Tian2025
ER  -