Hyperspectral remote sensing image destriping via spectral-spatial factorization.

Area of Science:

• Remote Sensing
• Image Processing
• Data Science

Background:

• Hyperspectral images (HSIs) offer rich spectral information but are prone to stripe noise.
• Stripe noise significantly degrades HSI quality, hindering applications.
• Existing destriping methods often struggle to balance noise removal with detail preservation.

Purpose of the Study:

• To propose a novel iterative method for effective hyperspectral image (HSI) destriping.
• To address the challenge of stripe noise removal while preserving spectral and spatial information.
• To develop a method robust to various noise types in HSIs.

Main Methods:

• A spectral-spatial factorization approach is introduced, rearranging HSI data into a 2D matrix.
• The HSI is decomposed into spectral and spatial information matrices.
• Leveraging sparsity of stripe noise, group sparsity of spatial information, and smoothness of spectral information for noise removal.

Main Results:

• The proposed iterative method demonstrates significant stripe noise removal.
• Achieved an average Peak Signal-to-Noise Ratio (PSNR) improvement exceeding 4dB on simulated datasets.
• Yielded superior Structural Similarity Index Measure (SSIM) results compared to existing methods.
• Showcased effective performance on real-world datasets with mixed noise.

Conclusions:

• The spectral-spatial factorization method is highly effective for HSI destriping.
• The technique successfully removes stripe noise while preserving crucial spectral and spatial details.
• This method offers a robust solution for enhancing HSI quality in diverse applications.