Affiliation:
1. China Electronic Technology Group Corporation, Beijing 100846, China
2. School of Electronic and Optical Engineering, Nanjing University of Science and Technology, Nanjing 210094, China
Abstract
Recently, the low-rank representation (LRR) model has been widely used in the field of remote sensing image denoising due to its excellent noise suppression capability. However, those low-rank-based methods always discard important edge details as residuals, leading to a common issue of blurred edges in denoised results. To address this problem, we take a new look at low-rank residuals and try to extract edge information from them. Therefore, a hierarchical denoising framework was combined with a low-rank model to extract edge information from low-rank residuals within the edge subspace. A prior knowledge matrix was designed to enable the model to learn necessary structural information rather than noise. Also, such traditional model-driven approaches require multiple iterations, and the solutions may be very complex and computationally intensive. To further enhance the noise suppression performance and computing efficiency, a hierarchical low-rank denoising model based on deep unrolling (HLR-DUR) was proposed, integrating deep neural networks into the hierarchical low-rank denoising framework to expand the information capture and representation capabilities of the proposed shallow model. Sufficient experiments on optical images, hyperspectral images (HSI), and synthetic aperture radar (SAR) images showed that HLR-DUR achieved state-of-the-art (SOTA) denoising results.
Reference41 articles.
1. Sabins, F.F., and Ellis, J.M. (2020). Remote Sensing: Principles, Interpretation, and Applications, Waveland Press.
2. Image restoration for remote sensing: Overview and toolbox;Rasti;IEEE Geosci. Remote Sens. Mag.,2021
3. Low-Frequency Magnetotelluric Data Denoising Using Improved Denoising Convolutional Neural Network and Gated Recurrent Unit;Li;IEEE Trans. Geosci. Remote Sens.,2024
4. IncepTCN: A new deep temporal convolutional network combined with dictionary learning for strong cultural noise elimination of controlled-source electromagnetic data;Li;Geophysics,2023
5. Han, L., Zhao, Y., Lv, H., Zhang, Y., Liu, H., and Bi, G. (2022). Remote sensing image denoising based on deep and shallow feature fusion and attention mechanism. Remote Sens., 14.