Hyperspectral Image Mixed Noise Removal Using Subspace Representation and Deep CNN Image Prior

Zhuang, Lina; Ng, Michael K.; Fu, Xiouhua

doi:10.3390/rs13204098

Cited by 17 publications

(4 citation statements)

References 50 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, the runtime (seconds) is also considered as the index to compute complexity. Especially, five state-of-the-art methods are employed as the baselines to conduct performance comparison, including deep-learningbased HySuDeep (Subspace representation and deep CNN Image prior) [15], tensor-decomposition-based LRTDGS (Weighted group sparsity-regularized low-rank tensor decomposition) [24], matrix-factorization-based E-3DTV (Enhanced 3-D total variation) [36], SNLRSF (Subspace-based nonlocal low-Rank and sparse factorization method) [33] and F-LRNMF (Framelet-regularized low-rank nonnegative matrix factorization method) [34].…”

Section: Experimental Simulations and Performance Analysismentioning

confidence: 99%

“…In the aspect of supervised methods, Xie et al proposed a convolution neural network with trainable nonlinear functions for hyperspectral image restoration [14]. Zhuang et al [15] adopted a fast and flexible denoising convolutional neural network to capture high local correlation in spatial subspace. However, deep learning based methods often involve a great deal of data training, as well as hyperparameters, which will lead to an increase in computational burden [16], [17].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Piecewise Weighted Smoothing Regularization in Tight Framelet Domain for Hyperspectral Image Restoration

Liu

Yang

et al. 2023

IEEE Access

View full text Add to dashboard Cite

Hyperspectral images captured by remote-sensing satellites are easily corrupted by various types of noise. Generally, hyperspectral signatures appear to be scattered in spatial-spectral domain, as well as noise. In transform domain, however, the principal components of a image are often centralized in the low-frequency band, while noise and some details are mainly contained in high-frequency components. The traditional transformation domain based smoothing methods take no account of the respective information distribution carried by different frequency bands. Hyperspectral image restoration aims to remove mixed noise for clean data, which usually amounts to an ill-posed inverse problem. Moreover, the matrixdecomposition-based model needs to reshape hyperspectral datacube into matrix form, which will lead to certain loss of spatial structure information. In order to address these issues, this paper incorporates piecewise weighted smoothing regularization in tight framelet domain to reformulate a novel convex model for hyperspectal image restoration, which maintains image signature to a great extent. Based on the noise-perturbed degradation model, the smoothing regularizer in tight framelet domain is imposed on abundance signature, in which the transformation matrix, as well as the weighted coefficient, is well designed for different frequency bands. As a surrogate of sparsity, the l q -norm is employed to denoise for the enhancement of hyperspectral signatures. To the end, an efficient solver is carefully designed to derive the closed-form solutions by proximal alternating optimization. The experimental results on several synthetic and real datasets, not only demonstrate that the performance of proposed method is better than that of the current state-of-the-art approaches, but also verify the validation of regularization terms for hyperspectral image restoration.INDEX TERMS Hyperspectral image restoration, tight framelet domain, piecewise weighted smoothing;degradation model, mixed noise

show abstract

Section: Experimental Simulations and Performance Analysismentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Piecewise Weighted Smoothing Regularization in Tight Framelet Domain for Hyperspectral Image Restoration

Liu

Yang

et al. 2023

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Almost all existing, either USL-or SSL-based, detection methods measure the abnormal level of each pixel using the reconstruction errors of the AE model. However, the reconstruction-based methods optimise the model by simply equivalent mapping between input and output data, which has the following problems: In the original spectral domain, the spectral information of HSI is affected by noise and spatial resolution to a certain extent [44,45], resulting in the insignificant separability of backgrounds and anomalies. Furthermore, the spectral dimension of HSI is higher.…”

Section: Introductionmentioning

confidence: 99%

Frequency‐to‐spectrum mapping GAN for semisupervised hyperspectral anomaly detection

Wang

Gao

et al. 2023

CAAI Trans on Intel Tech

View full text Add to dashboard Cite

Most unsupervised or semisupervised hyperspectral anomaly detection (HAD) methods train background reconstruction models in the original spectral domain. However, due to the noise and spatial resolution limitations, there may be a lack of discrimination between backgrounds and anomalies. This makes it easy for the autoencoder to capture the low‐level features shared between the two, thereby increasing the difficulty of separating anomalies from the backgrounds, which runs counter to the purpose of HAD. To this end, the authors map the original spectrums to the fractional Fourier domain (FrFD) and reformulate it as a mapping task in which restoration errors are employed to distinguish background and anomaly. This study proposes a novel frequency‐to‐spectrum mapping generative adversarial network for HAD. Specifically, the depth separable features of backgrounds and anomalies are enhanced in the FrFD. Due to the semisupervised approach, FTSGAN needs to learn the embedded features of the backgrounds, thus mapping and restoring them from the FrFD to the original spectral domain. This strategy effectively prevents the model from focussing on the numerical equivalence of input and output, and restricts the ability of FTSGAN to restore anomalies. The comparison and analysis of the experiments verify that the proposed method is competitive.

show abstract

“…To better preserve the spatial and spectral details of hyperspectral data, researchers have combined image and depth priors to give full play to their respective advantages, consequently achieving satisfactory denoising results. For example, by using the low-rank and local self-similarity priors of hyperspectral data in the spectral domain and combining them with the spatial depth prior extracted using a CNN, the mixed noise has been removed, thus providing an improved denoising performance [29][30][31][32]. Given the great success of attention mechanisms in the fields of image recognition, target detection, and other computer vision applications, researchers have proposed the use of a deep learning denoising framework based on an attention mechanism to further utilize the global dependence and correlation between spatial and spectral information in hyperspectral data.…”

Section: Introductionmentioning

confidence: 99%

Self-Supervised Denoising for Real Satellite Hyperspectral Imagery

2022

View full text Add to dashboard Cite

Satellite hyperspectral remote sensing has gradually become an important means of Earth observation, but the existence of various types of noise seriously limits the application value of satellite hyperspectral images. With the continuous development of deep learning technology, breakthroughs have been made in improving hyperspectral image denoising algorithms based on supervised learning; however, these methods usually require a large number of clean/noisy training pairs, a target that is difficult to meet for real satellite hyperspectral imagery. In this paper, we propose a self-supervised learning-based algorithm, 3S-HSID, for denoising real satellite hyperspectral images without requiring external data support. The 3S-HSID framework can perform robust denoising of a single satellite hyperspectral image in all bands simultaneously. It first conducts a Bernoulli sampling of the input data, then uses the Bernoulli sampling results to construct the training pairs. Furthermore, the global spectral consistency and minimum local variance are used in the loss function to train the network. We use the training model to predict different Bernoulli sampling results, and the average of multiple predicted values is used as the denoising result. To prevent overfitting, we adopt a dropout strategy during training and testing. The results of denoising experiments on the simulated hyperspectral data show that the denoising performance of 3S-HSID is better than most state-of-the-art algorithms, especially in terms of maintaining the spectral characteristics of hyperspectral images. The denoising results for different types of real satellite hyperspectral data also demonstrate the reliability of the proposed method. The 3S-HSID framework provides a new technical means for real satellite hyperspectral image preprocessing.

show abstract

Hyperspectral Image Mixed Noise Removal Using Subspace Representation and Deep CNN Image Prior

Cited by 17 publications

References 50 publications

Piecewise Weighted Smoothing Regularization in Tight Framelet Domain for Hyperspectral Image Restoration

Piecewise Weighted Smoothing Regularization in Tight Framelet Domain for Hyperspectral Image Restoration

Frequency‐to‐spectrum mapping GAN for semisupervised hyperspectral anomaly detection

Self-Supervised Denoising for Real Satellite Hyperspectral Imagery

Contact Info

Product

Resources

About