A Hierarchical Low-Rank Denoising Model for Remote Sensing Images Based on Deep Unfolding.

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Low-rank representation (LRR) models are effective for remote sensing image denoising but often blur edges.
• Existing methods struggle to retain crucial edge details present in the residuals.

Purpose of the Study:

• To develop an advanced denoising technique that preserves edge information.
• To enhance the noise suppression capability and computational efficiency of low-rank denoising models.

Main Methods:

• A hierarchical denoising framework was integrated with a low-rank model to extract edge information from residuals.
• A prior knowledge matrix was incorporated to guide the model in learning structural information.
• A deep unrolling approach was employed to create the Hierarchical Low-Rank Denoising using Unrolling (HLR-DUR) model.

Main Results:

• The HLR-DUR model successfully extracted edge information from low-rank residuals.
• Experiments demonstrated that HLR-DUR significantly improved denoising performance across optical, hyperspectral (HSI), and synthetic aperture radar (SAR) images.
• The proposed model achieved state-of-the-art (SOTA) denoising results.

Conclusions:

• HLR-DUR effectively addresses the edge blurring issue in low-rank denoising.
• The integration of deep neural networks enhances information capture and representation for superior denoising.
• The proposed method offers a computationally efficient and high-performing solution for remote sensing image denoising.