Construction site image dedusting via pyramid architecture based on the Taylor series
Document Type
Article
Publication Date
7-25-2025
Abstract
The construction industry is increasingly reliant on advanced imaging technologies to enhance operational efficiency and safety. In addition, construction site images often suffer from visibility degradation due to airborne dust, making it challenging for vision-based applications such as monitoring, safety assessment, and autonomous navigation. This research presents DedustNet, a novel approach to image dedusting that combines Taylor's theorem with a Laplacian pyramid architecture when leveraging an attention-sharing mechanism to effectively remove dust while preserving essential structural details. Our method utilizes a Tucker reconstruction technique to reduce noise and preserve high-level and frequency features while employing an attention-sharing mechanism to efficiently process higher-order approximations. The proposed model demonstrates superior performance in both image quality and processing speed compared to state-of-the-art methods, achieving a PSNR of 28.63, SSIM of 0.93, and an inference time of just 12 ms. These results highlight the potential of our approach for enhancing the reliability of vision-based construction site analysis.
Publication Source (Journal or Book title)
Expert Systems with Applications
Recommended Citation
Bonyani, M., Soleymani, M., & Wang, C. (2025). Construction site image dedusting via pyramid architecture based on the Taylor series. Expert Systems with Applications, 284 https://doi.org/10.1016/j.eswa.2025.127837