A Single Model CNN for Hyperspectral Image Denoising

Abstract

Denoising is a common pre-processing step prior to the analysis and interpretation of hyperspectral images (HSIs). However, the vast majority of methods typically adopted for HSI denoising exploit architectures originally developed for grayscale or RGB images, exhibiting limitations when processing highdimensional HSI data cubes. In particular, traditional methods do not take into account the high spectral correlation between adjacent bands in HSIs, which leads to unsatisfactory denoising performance as the rich spectral information present in HSIs is not fully exploited. To overcome this limitation, this paper considers deep learning models -such as convolutional neural networks (CNNs)-to perform spectral-spatial HSI denoising. The proposed model, called HSI single denosising CNN (HSI-SDeCNN), efficiently takes into consideration both the spatial and spectral information contained in HSIs. Experimental results on both synthetic and real data demonstrate that the proposed HSI-SDeCNN outperforms other state-of-the-art HSI denoising methods. Source code: https://github.com/mhaut/HSI-SDeCNN

SLRL4D: Joint Restoration of Subspace Low-Rank Learning and Non-Local 4-D Transform Filtering for Hyperspectral Image

“…In recent years, deep learning methods have earned a prestigious position that cannot be ignored in the field of computer vision [62][63][64][65][66]. Aiming to further verify the capacity of the proposed SLRL4D to remove heavy Gaussian noise, two state-of-the-art deep learning HSI denoising methods HSID-CNN [15] and HSI-SDeCNN [16] are employed for comparison. Similar to most deep learning methods, HSID-CNN and HSI-SDeCNN are mainly designed to remove Gaussian noise.…”

“…Similar to most deep learning methods, HSID-CNN and HSI-SDeCNN are mainly designed to remove Gaussian noise. Therefore, for a fair comparison, in the following experiments, we used the exact same WDC dataset and simulation manner of heavy Gaussian noise as in the article corresponding to HSI-SDeCNN [16], that is, the noise level σ n was set to [50,75,100]. The detailed quantitative evaluation results are shown in Table 3.…”

“…The disadvantage of these methods is distinct; that is, the spatial or spectral correlations of the 3-D HSI are not maintained. Although methods based on wavelet shrinkage [7][8][9], non-local means [10], sparse representation [11][12][13][14] or deep learning [15][16][17][18][19] achieve the combination of spatial-spectral information, the sphere of their application is still limited. The prior knowledge of the original image and mixed noises are insufficiently explored and usually remove specific noise only; thus, their performance in real scenarios is not guaranteed.…”

Bayesian Approach in a Learning-Based Hyperspectral Image Denoising Framework

“…In recent works, it is observed that learning-based techniques are utilized for noise reduction in HSI data. Some notable contributions include [25], [26], [27], [28], [29] and [30]. Authors in [25] employ a cubic noisy-clean image learning approach where non-linearity functions are also trainable along with convolutional weights and biases.…”

Hyperspectral Image Denoising Using 3-D Geometrical Kernel With Local Similarity Prior

“…These methods hierarchically extracted image features from local regions, which are also inspired by local prior. For example, the local structures in HS images are extracted in convolutional neural networks (CNNs) level by level [35].…”