In recent years, deep learning-based methods have been extensively utilized in remote sensing image scene classification and achieved remarkable performance. The wide geographical coverage resolution differences of images result significant within-class diversity between-class similarity, hindering the further improvement accuracy. Attention-based automatically estimate importance local regions by learning weight assignments, which effectively enhance feature extraction capability network....

This article focuses on hyperspectral image (HSI) super-resolution that aims to fuse a low-spatial-resolution HSI and high-spatial-resolution multispectral form (HR-HSI). Existing deep learning-based approaches are mostly supervised rely large number of labeled training samples, which is unrealistic. The commonly used model-based unsupervised flexible but handcrafted priors. Inspired by the specific properties model, we make first attempt design model-inspired network for in an manner....

Hyperspectral image super-resolution addresses the problem of fusing a low-resolution hyperspectral (LR-HSI) and high-resolution multispectral (HR-MSI) to produce (HR-HSI). In this paper, we propose novel fusion approach for by exploiting specific properties matrix decomposition, which consists four main steps. First, an endmember extraction algorithm is used extract initial spectral from LR-HSI. Then, with matrix, estimate spatial i.e., spatial-contextual information, degraded observations...

Hyperspectral image (HSI) fusion refers to the reconstruction of a high-resolution HSI by fusing low-resolution (LR-HSI) and multispectral (HR-MSI) over same scene. Recently, researchers have proposed many approaches handle this issue. However, most them assume that both spatial spectral degradation functions are known, which often limited or unavailable in reality. This article presents novel model-driven deep network based on matrix decomposition, considers correlations reasonably embeds...

Significant progress has been achieved in remote sensing image scene classification (RSISC) with the development of convolutional neural networks (CNNs) and vision transformers (ViT). However, high intra-class diversity inter-class similarity are still enormous challenges for RSISC. Metric learning can effectively improve discriminative ability deep representations by constraining distance between features. Previous metric methods only optimize feature space representation through function,...

Hyperspectral target detection (HTD) is an important issue in earth observation, with applications both military and civilian domains. However, conventional representation-based detectors are hindered by the reliance on unknown background dictionary, limited ability to capture nonlinear representations using linear mixing model (LMM), insufficient background-target recognition based handcrafted priors. To address these problems, this paper proposes interpretable representation network that...

Deep learning-based methods based on convolutional neural networks (CNNs) have demonstrated remarkable performance in hyperspectral image (HSI) classification. Most of these approaches are only 2-D CNN or 3-D CNN. It is dramatic from the literature that using just may result missing channel relationship information, and make model very complex. Moreover, existing network models do not pay enough attention to extracting spectral-spatial correlation information. To address issues, we propose a...

Hyperspectral and multispectral image fusion aims to fuse a low-spatial-resolution hyperspectral (HSI) high-spatial-resolution form HSI. Motivated by the success of model- deep learning-based approaches, we propose novel patch-aware approach for HSI unfolding subspace-based optimization model, where moderate-sized patches are used in both training test phases. The goal this is make full use information patch under subspace representation, restrict scale enhance interpretability network,...

In this article, we propose a novel tensorial approach, namely, generalized tensor regression, for hyperspectral image classification. First, simple and effective classifier, i.e., the ridge regression multivariate labels, is extended to its version by taking advantages of representation. Then, discrimination information different modes exploited further strengthen capacity model. Moreover, model can be simplified solved easily. Different from traditional methods, proposed utilized capture...

Fusion-based hyperspectral image (HSI) superresolution aims to produce a high-spatial-resolution HSI by fusing low-spatial-resolution and multispectral image.Such super-resolution process can be modeled as an inverse problem, where the prior knowledge is essential for obtaining desired solution.Motivated success of diffusion models, we propose novel spectral fusion-based super-resolution.Specifically, first investigate spectrum generation problem design model data distribution.Then, in...

Convolutional Neural Networks (CNNs) are widely used in various fields, and have shown good performance hyperspectral image (HSI) classification. Recently, utilizing deep networks to learn spatial-spectral features has become of great interest. However, excessively increasing the depth network may result overfitting. Moreover, HSI classification, existing models ignore strong complementary yet correlated information among different hierarchical layers. In order address these two problems, a...

From the non-local self-similarity (NSS)-based image denoising to convolutional-network (ConvNet)-based denoising, performance has been greatly improved. However, it is still not clear how utilize similar web images guide using ConvNet. This paper proposes a novel ConvNet for explore both internal (NSS) and external correlations when are available. Since may be taken with different viewpoints, focal lengths, contain objects, difficult directly at level Therefore, we propose an patch matching...

Fusing a low-resolution hyperspectral (LRHS) image and high-resolution multispectral (HRMS) to generate (HRHS) has grown significant attractive application in remote sensing fields. Recently, the popularization of deep learning injected more possibilities into fusion work. However, there still exists difficulty that is how make best acquired LRHS HRMS images. In this article, we present twice optimizing net with matrix decomposition fulfill task, which can be roughly divided three stages:...

This letter presents a postprocessing algorithm for kernel sparse representation (KSR)-based hyperspectral image classifier, which is based on the integration of spatial and spectral information. A pixelwise KSR first used to find coefficient vectors image. Then, sparsity concentration index (SCI) rule-guided semilocal graph regularization (SSG), called SSG+SCI, proposed determine refined that promote continuity within each class. Finally, these are obtain final classification map. Compared...

Recently, more and multi-layer perceptron (MLP) -like models have been proposed. Among them, CycleMLP is good at dense feature prediction tasks, which potentially useful for hyperspectral image (HSI) classification. However, the receptive field of tends to be cross-shaped, will lead insufficient spatial information extraction. Additionally, most HSI classification methods only use from single data. Lack diversity in features a modality limits performance. To address these issues, novel...

The main challenges of remote sensing image scene classification are extracting discriminative features and making full use the training data. current mainstream deep learning methods usually only hard labels samples, ignoring potential soft natural labels. Self-supervised can take advantage However, it is difficult to train a self-supervised network due limitations dataset computing resources. We propose knowledge distillation (SSKDNet) solve aforementioned challenges. Specifically, feature...

This paper presents a region-based relaxed multiple kernel collaborative representation method for the spatial-spectral classification of hyperspectral images. The proposed consists three steps. In first step, multiscale achieved by extending superpixel segmentation algorithm is designed to capture information For each scale, image can be segmented into several nonoverlapping spectrally similar regions that consist some spatially adjacent pixels. second two criteria (i.e., moments) are...

This paper presents a spatial-spectral method for hyperspectral image classification in the regularization framework of kernel sparse representation. First, two constraint terms are appended to recovery model The first one is graph-based spatially-smooth which utilized describe contextual information images. second spatial location constraint, exploited incorporate prior knowledge training pixels. Then, an efficient alternating direction multipliers developed solve corresponding minimization...

Kernel Fisher discriminant analysis (KFD) can map well-log data into a nonlinear feature space to make linear nonseparable problem of fracture identification separable one. Commonly, KFD uses one kernel. However, the prediction capacity based on kernel is limited some extent, especially for complex classification problem, such as in tight sandstone reservoirs. To alleviate this we have used multiple (MKFD) method recognize zones. MKFD multiscaled Gaussian functions instead single realize...