Eigenimage2Eigenimage (E2E): A Self-Supervised Deep Learning Network for Hyperspectral Image Denoising

Area of Science:

• Remote Sensing
• Image Processing
• Deep Learning

Background:

• Deep learning denoisers require extensive paired noisy-clean data, which is scarce for hyperspectral images (HSIs).
• Existing methods struggle with the high dimensionality and spectral redundancy of HSIs.

Purpose of the Study:

• To develop a self-supervised learning framework for hyperspectral image denoising.
• To enable effective HSI denoising without requiring paired noisy-clean training data.

Main Methods:

• Proposed the Eigenimage2Eigenimage (E2E) framework, transforming HSI denoising into eigenimage denoising.
• Developed a self-supervised learning strategy to generate noisy-noisy paired training data from single noisy HSIs.
• Applied the E2E framework to denoise HSIs without constraints on the number of spectral bands.

Main Results:

• The E2E framework successfully trained a denoiser using only noisy data.
• Experimental results demonstrated superior performance compared to existing deep learning-based HSI denoising methods.
• The method effectively denoises HSIs across varying spectral band counts.

Conclusions:

• Self-supervised learning offers a viable solution for HSI denoising when clean data is unavailable.
• The E2E framework provides a robust and flexible approach for hyperspectral image denoising.
• The proposed method advances deep learning applications in remote sensing image analysis.