A Dual Transformer Super-Resolution Network for Improving the Definition of Vibration Image

Abstract

Visual measurement methods are gaining more and more attention in the field of structural body health monitoring due to the advantages of long-range, noncontact, and multipoint monitoring. However, the imaging system is usually affected by many factors, such as distortion, blurring, and noise, which lead to displacement measurement errors after the degradation of the acquired image quality. Therefore, in this article, we propose a structural body image super-resolution network based on a dual transformer architecture to improve the clarity of the collected structural body vibration displacement image to better capture the vibration displacement information of the target. Meanwhile, we design a dual transformer block based on an encoder–decoder architecture for the characteristics of vision-based structural body vibration displacement measurement tasks to better extract structural body image details and edge feature information. In this module, we introduce two different transformers. In addition, modules based on the encoder–decoder architecture focus more on the input and output image information and often ignore the feature information in different layers. Therefore, we introduce an attention mechanism in the network and interact with the feature information in different layers of the encoder–decoder architecture to obtain a better structural body image super-resolution effect. After comparison tests with the rest of the latest and most classical networks as well as the current optimal networks, it is shown that our network obtains excellent image reconstruction results under different structural body vibration image datasets (SETs), which also provides a strong guarantee for the task of accurate vision-based structural body vibration displacement measurement.