Feature extraction is of significance for hyperspectral image (HSI) classification. Compared with conventional hand-crafted feature extraction, deep learning can automatically learn features discriminative information. However, two issues exist in applying to HSIs. One issue how jointly extract spectral and spatial features, the other one train model when training samples are scarce. In this paper, a convolutional neural network two-branch architecture proposed joint spectral-spatial from...

Hyperspectral image (HSI) denoising is an essential preprocess step to improve the performance of subsequent applications. For HSI, there much global and local redundancy correlation (RAC) in spatial/spectral dimensions. In addition, can be improved greatly if RAC utilized efficiently process. this paper, HSI method proposed by jointly utilizing domains. First, sparse coding exploited model spatial domain spectral domain. Noise removed approximated data with learned dictionary. At stage,...

Recently, the graph convolutional network (GCN) has drawn increasing attention in hyperspectral image (HSI) classification. Compared with neural (CNN) fixed square kernels, GCN can explicitly utilize correlation between adjacent land covers and conduct flexible convolution on arbitrarily irregular regions; hence, HSI spatial contextual structure be better modeled. However, to reduce computational complexity promote semantic learning of covers, usually works superpixel-based nodes rather than...

Enhancing the spatial resolution of hyperspectral image (HSI) is significance for applications. Fusing HSI with a high (HR) multispectral (MSI) an important technology enhancement. Inspired by success deep learning in enhancement, this paper, we propose HSI-MSI fusion method designing convolutional neural network (CNN) two branches which are devoted to features and MSI. In order exploit spectral correlation fuse MSI, extract from spectrum each pixel low HSI, its corresponding neighborhood...

Performance of hyperspectral image classification depends on feature extraction. Compared with conventional hand-crafted extraction, deep learning can learn more discriminative information. In this paper, a two-channel convolutional neural network (Two-CNN) is proposed to jointly spectral-spatial from image. The model composed two channels CNN, each which learns spectral domain and spatial respectively. learned are then concatenated fed fully connected layer extract joint for classification....

Assessing the quality of a reconstructed hyperspectral image (HSI) is significance for restoration and super-resolution. Current assessment methods such as peak signal-noise-ratio require availability pristine reference image, which often not available in reality. In this paper, we propose no-reference method based on quality-sensitive features extraction. Difference statistical properties between distorted HSIs analyzed both spectral spatial domains, then multiple statistics that are...

Limited by the shape-fixed kernels, convolutional neural networks (CNNs) are usually difficult to model difform land covers in hyperspectral images (HSIs), leading inadequate use. Recently, benefiting from ability conduct shape-adaptive convolutions and complex patterns graph-structured data, graph (GCNs) have been applied HSI classification. However, due massive computation GCNs, is pretreated into a based on specific superpixel segmentation, which limits modeling of spatial topologies same...

Hyperspectral image (HSI) denoising is significant for correct interpretation. In this paper, a sparse representation framework that unifies and spectral unmixing in closed-loop manner proposed. While conventional approaches treat separately, the proposed scheme utilizes information from as feedback to distortion. Both act constraints others are solved iteratively. Noise suppressed via coding, fractional abundance estimated using sparsity prior of endmembers library. The used regularizer...

Hyperspectral (HS) super-resolution reconstruction is an ill-posed inversion problem, for which the solution from constraint not unique. To address this, HS image method proposed to first utilize joint regulation of spatial and spectral nonlocal similarities. We then fused panchromatic images with sparse regulation. With these two terms, edge sharpness spectrum consistency are preserved noises suppressed. The tested Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) Hyperion evaluated...

Super-resolution (SR) is significant for hyperspectral image (HSI) applications. In single-frame HSI SR, how to reconstruct detailed structures in high resolution (HR) challenging since there no auxiliary (e.g., HR multispectral image) providing structural information. Wavelet could capture different orientations, and emphasis on predicting high-frequency wavelet sub-bands helpful recovering the SR. this study, we propose a multi-scale 3D convolutional neural network (MW-3D-CNN) which...

Hyperspectral–multispectral image (HSI-MSI) fusion relies on a robust degradation model and data prior, where the former describes degeneration of HSI in spectral spatial domains, latter reveals latent statistics expected high-resolution (HR) HSI. In practice, is often unknown, prior usually too complicated to be expressed analytically. this study, we propose variational network for HSI-MSI (VaFuNet), which are implicitly represented by deep learning jointly learned from training data. A...

Convolutional neural networks (CNNs) are of great interest and have demonstrated remarkable performance in hyperspectral images (HSIs) classification. However, due to the current configuration convolution layers with a fixed kernel shape, regular CNNs inherently limited modeling diverse land-cover structures, particularly cross-classes edge regions, where irregular class boundaries would lead high classification errors. To address this issue, we propose content-guided CNN (CGCNN) for HSI...

Fusing low-resolution (LR) hyperspectral images (HSIs) with high-resolution (HR) multispectral (MSIs) is a significant technology to enhance the resolution of HSIs. Despite encouraging results from deep learning (DL) in HSI–MSI fusion, there are still some issues. First, HSI multidimensional signal, and representability current DL networks for features has not been thoroughly investigated. Second, most fusion need HR ground truth training, but it often unavailable reality. In this study, we...

The current advanced hyperspectral super-resolution methods utilize Convolutional Neural Networks (CNNs) that are either deeper or wider. These networks designed to acquire end-to-end mapping capability, facilitating the transformation from Low-Resolution Hyperspectral Images (LR-HSI) and High-Resolution Multispectral (HR-MSI) (HR-HSI). existing lack capability capture details structures in image effectively, while multi-input multi-output can address this issue efficiently. Therefore, paper...

The fusion of high-resolution multispectral (HrMSI) and low-resolution hyperspectral images (LrHSI) has been acknowledged as a promising method for generating image (HrHSI), which is also termed to be an essential part precise recognition cataloguing the underlying materials. In order improve LrHSI HrMSI performance, in this article, we propose novel Nonnegative Matrix Factorization Inspired Deep Unrolling Networks, dubbed NMF-DuNet, fusing HrMSI. For aim, initially, variational model...

A deep convolutional neural network (CNN) has shown its great potential in hyperspectral image (HSI) super-resolution (SR). Integrating CNN with attention mechanism is expected to boost the SR performance. However, how learn along spectral, spatial, and channel dimensions of HSI still an open issue, current not efficient capturing long-range interdependency HSI. In this letter, we first design a local 3-D module spectral-spatial-channel by exploiting contextual information Then, propose...

In recent years, deep learning-based models have produced encouraging results for hyperspectral image (HSI) classification. Specifically, Convolutional Long Short-Term Memory (ConvLSTM) has shown good performance learning valuable features and modeling long-term dependencies in spectral data. However, it is less effective spatial features, which an integral part of images. Alternatively, convolutional neural networks (CNNs) can learn but they possess limitations handling due to the local...

Pansharpening is an image fusion procedure, which aims to produce a high spatial resolution multispectral by combining low and panchromatic image. The most popular successful paradigm for pansharpening the framework known as detail injection, while it cannot fully exploit complex non-linear complementary features of both images. In this paper, we propose injection model inspired deep network (DIM-FuNet). Firstly, treating complicated details learning problem, establish unified optimizing...

Due to the physical boundaries, fusing low spatial resolution hyperspectral (LrHSI) with high multispectral (HrMSI) images is a hot and promising area for obtaining that have spatial-spectral (HrHSI). Effectively formulating fundamental features of (HSI), such as global spectral correlation, nonlocal well complex in HSI-MSI fusion. Moreover, fusion process highly affected by degradation systems, where these systems are not known real scenarios. To this end, article, we proposed model-guided...

This paper presents an interactive feedback scheme of spatial resolution enhancement and spectral unmixing in hyperspectral imaging. Traditionally operations have been carried out separately, often series. In such sequential processing, spatially enhanced images (HSIs) may introduce distortion fidelity making results unreliable, or vice versa. Since both high- low-resolution HSIs the same endmembers, deviation between targets estimated high-resolution can be used as to control enhancement....

Reconstructing the 3D hyperspectral image (HSI) from 2D snapshot measurements is a key task in spectral compressive imaging (SCI). Traditional model-based HSI reconstruction methods rely on hand-crafted priors. Recently, deep unfolding networks (DUNs) learn priors using convolutional neural (CNNs) and have achieved satisfactory results. Most of DUNs assume degradations SCI are known. However, due to phase aberration distortion problems real process, there certain gap between ideal...

The fusion-based super-resolution of hyperspectral images (HSIs) draws more and attention in order to surpass the hardware constraints intrinsic imaging systems terms spatial resolution. Low-resolution (LR)-HSI is combined with a high-resolution multispectral image (HR-MSI) achieve HR-HSI. In this article, we propose multiresolution details enhanced attentive dual-UNet improve resolution HSI. entire network contains two branches. first branch wavelet detail extraction module, which performs...