Abstract:Hyperspectral images (HSIs) are inevitably corrupted with various types of noise, which seriously degrades the data quality and usability. Denoising is an essential preprocessing task of HSI processing. Recently, benefiting from the great learning ability of deep learning, convolutional neural network (CNN) denoisers have obtained state-of-the-art performances for Gaussian noise removal. However, one central problem remains largely unsolved: how to deal with the complicated noise in the real-world HSIs, especially when a paired training data set is unavailable. In this article, a self-supervised hyperspectral image denoising network (SHDN) is proposed, which consists of a noise estimator and a CNN denoiser. Rather than defining a complex noise model to generate training pairs on the clean HSIs, a self-supervised training scheme is first proposed by considering the noisy HSI itself as the training data. Through the noise estimator, the realistic noise samples can be extracted and combined with the clean bands to make up the training pairs. In addition, to jointly restore the target noisy band and to maintain the spectral consistency, a flexible multi-to-single band convolutional network is designed, where the noisy band and the neighboring bands are jointly aggregated via multiscale contextualized dilated blocks and the spectral–spatial convolutional unit. Experiments on HSIs from spaceborne, airborne, unmanned aerial vehicle (UAV)-borne, and ground-based data sets demonstrate the applicability and the generalization of SHDN in the real scenarios. Additionally, the usability of the noisy bands and the suitability of the SHDN framework in the subsequent applications are verified in the land-cover mapping experiments.

Hyperspectral image restoration via CNN denoiser prior regularized low-rank tensor recovery

Hyperspectral Image Denoising via Low-Rank Representation and CNN Denoiser

A Self-Supervised Denoising Network for Satellite-Airborne-Ground Hyperspectral Imagery

HSI-DeNet: Hyperspectral Image Restoration via Convolutional Neural Network

A hyperspectral image denoising method based on land cover spectral autocorrelation

Fast non-iterative blind restoration of hyperspectral images with PSFs

Hyperspectral Image Restoration Combining Intrinsic Image Characterization With Robust Noise Modeling

Hyperspectral Image Restoration with Self-Supervised Learning: A Two-Stage Training Approach

A Self-Supervised Hyperspectral Image Restoration Method Based on Convolutional Sparse Coding and Superpixel Segmentation

Hyperspectral Restoration Via $L_0$ Gradient Regularized Low-Rank Tensor Factorization

Hyperspectral Image Denoising via Tensor Low-Rank Prior and Unsupervised Deep Spatial–Spectral Prior

Hyperspectral Image Denoising Via Nonnegative Matrix Factorization and Convolutional Neural Networks.

Nonlocal Spatial–Spectral Neural Network for Hyperspectral Image Denoising

Hyperspectral Image Restoration via Total Variation Regularized Low-rank Tensor Decomposition

Multitask Sparse Representation Model-Inspired Network for Hyperspectral Image Denoising

Hyperspectral Imagery Denoising Via Reweighed Sparse Low-Rank Nonnegative Tensor Factorization

Hyperspectral Image Denoising Via Matrix Factorization and Deep Prior Regularization

Hyperspectral Image Denoising with Partially Orthogonal Matrix Vector Tensor Factorization

Hyperspectral Image Restoration by Tensor Fibered Rank Constrained Optimization and Plug-and-Play Regularization

Hybrid Spatial-spectral Neural Network for Hyperspectral Image Denoising

HIR-Diff: Unsupervised Hyperspectral Image Restoration Via Improved Diffusion Models