Deep-learning (DL) algorithms, which learn the representative and discriminative features in a hierarchical manner from data, have recently become hotspot machine-learning area been introduced into geoscience remote sensing (RS) community for RS big data analysis. Considering low-level (e.g., spectral texture) as bottom level, output feature representation top level of network can be directly fed subsequent classifier pixel-based classification. As matter fact, by carefully addressing...

In hyperspectral remote sensing image classification, multiple features, e.g., spectral, texture, and shape are employed to represent pixels from different perspectives. It has been widely acknowledged that properly combining features always results in good classification performance. this paper, we introduce the patch alignment framework linearly combine optimal way obtain a unified low-dimensional representation of these for subsequent classification. Each feature its particular...

Convolutional neural network (CNN) has demonstrated impressive ability to represent hyperspectral images and achieve promising results in image classification. However, traditional CNN models can only operate convolution on regular square regions with fixed size weights, thus, they cannot universally adapt the distinct local various object distributions geometric appearances. Therefore, their classification performances are still be improved, especially class boundaries. To alleviate this...

Deep networks have achieved excellent performance in learning representation from visual data. However, the supervised deep models like convolutional neural network require large quantities of labeled data, which are very expensive to obtain. To solve this problem, paper proposes an unsupervised network, called stacked denoising auto-encoders, can map images hierarchical representations without any label information. The optimized by layer-wise training, is constructed stacking layers...

Existing inpainting methods have achieved promising performance for recovering regular or small image defects. However, filling in large continuous holes remains difficult due to the lack of constraints hole center. In this paper, we devise a Recurrent Feature Reasoning (RFR) network which is mainly constructed by plug-and-play module and Knowledge Consistent Attention (KCA) module. Analogous how humans solve puzzles (i.e., first easier parts then use results as additional information...

Hyperspectral image (HSI) contains a large number of spatial-spectral information, which will make the traditional classification methods face an enormous challenge to discriminate types land-cover. Feature learning is very effective improve performances. However, current feature approaches are mostly based on simple intrinsic structure. To represent complex HSI, novel algorithm, termed hypergraph discriminant analysis (SSHGDA), has been proposed basis and learning. SSHGDA constructs...

In this paper, we propose a method for the dimensionality reduction (DR) of spectral-spatial features in hyperspectral images (HSIs), under umbrella multilinear algebra, i.e., algebra tensors. The proposed approach is tensor extension conventional supervised manifold-learning-based DR. particular, define organization scheme representing pixel's feature and develop discriminative locality alignment (TDLA) removing redundant information subsequent classification. optimal solution TDLA obtained...

In hyperspectral remote sensing data mining, it is important to take into account of both spectral and spatial information, such as the signature, texture feature, morphological property, improve performances, e.g., image classification accuracy. a feature representation point view, nature approach handle this situation concatenate features single but high dimensional vector then apply certain dimension reduction technique directly on that concatenated before feed subsequent classifier....

Dimensionality reduction (DR) is an important way of improving the classification accuracy a hyperspectral image (HSI). Graph learning, which can effectively reveal intrinsic relationships data, has been widely used in case HSIs. However, most them are based on simple graph to represent binary data. An HSI contains complex high-order among different samples. Therefore, this article, we propose hybrid-graph learning method HSI, termed enhanced discriminant (EHGDL). In EHGDL, intraclass...

How can we find a general way to choose the most suitable samples for training classifier? Even with very limited prior information? Active learning, which be regarded as an iterative optimization procedure, plays key role construct refined set improve classification performance in variety of applications, such text analysis, image recognition, social network modeling, etc. Although combining representativeness and informativeness has been proven promising active sampling, state-of-the-art...

In this paper, we propose a novel nonlocal patch tensor-based visual data completion algorithm and analyze its potential problems. Our consists of two steps: the first step is initializing image with triangulation-based linear interpolation second grouping similar patches as tensor then applying proposed technique. Specifically, treating group matrices tensor, impose low-rank constraint on through recently nuclear norm. Moreover, observe that after step, gets blurred and, thus, have found...

Inpainting methods aim to restore missing parts of corrupted images and play a critical role in many computer vision applications, such as object removal image restoration. Although existing perform well on with small holes, restoring large holes remains elusive. To address this issue, paper proposes Progressive Reconstruction Visual Structure (PRVS) network that progressively reconstructs the structures associated visual feature. Specifically, we design novel (VSR) layer entangle...

Recently, deep learning based video super-resolution (SR) methods have achieved promising performance. To simultaneously exploit the spatial and temporal information of videos, employing 3-dimensional (3D) convolutions is a natural approach. However, straight utilizing 3D may lead to an excessively high computational complexity which restricts depth SR models thus undermine In this paper, we present novel fast spatio-temporal residual network (FSTRN) adopt for task in order enhance...

Road segmentation from remote-sensing images is a challenging task with wide ranges of application potentials. Deep neural networks have advanced this field by leveraging the power large-scale labeled data, which, however, are extremely expensive and time-consuming to acquire. One solution use cheap available data train model deploy it directly process specific domain. Nevertheless, well-known domain shift (DS) issue prevents trained generalizing well on target In article, we propose novel...

Image inpainting is a challenging computer vision task that aims to fill in missing regions of corrupted images with realistic contents. With the development convolutional neural networks, many deep learning models have been proposed solve image issues by information from large amount data. In particular, existing algorithms usually follow an encoding and decoding network architecture which some operations standard schemes are employed, such as static convolution, only considers pixels fixed...

Hyperspectral anomaly detection (HAD) is an important hyperspectral image application. HAD can find pixels with anomalous spectral signatures compared their neighbor background without any prior information. While most of the existed researches are related to statistic-based and distance-based techniques, by summarizing samples certain models, then, finding very few outliers various distance metrics, this review focuses on based machine learning methods, which have witnessed remarkable...

Transformer has been extensively explored for hyperspectral image (HSI) classification. However, transformer poses challenges in terms of speed and memory usage because its quadratic computational complexity. Recently, the Mamba model emerged as a promising approach, which strong long-distance modeling capabilities while maintaining linear representing HSI is challenging due to requirement an integrated spatial spectral understanding. To remedy these drawbacks, we propose novel...

Recently, remote sensing image captioning has gained significant attention in the community. Due to differences spatial resolution of images, existing methods this field have predominantly concentrated on fine-grained extraction features, but they cannot effectively handle semantic consistency between visual features and textual features. To efficiently align image-text, we propose a novel two-stage vision-language pre-training-based approach bootstrap interactive image-text alignment for...

The detection and identification of target pixels such as certain minerals man-made objects from hyperspectral remote sensing images is great interest for both civilian military applications. However, due to the restriction in spatial resolution most airborne or satellite sensors, targets often appear subpixels image (HSI). observed spectral feature desired pixel (positive sample) therefore a mixed signature reference spectrum background spectra (negative samples), which belong various land...

Target detection is one of the most important applications in hyperspectral remote sensing image analysis. However, state-of-the-art machine-learning-based algorithms for target cannot perform well when training samples, especially are limited number. This because data and test drawn from different distributions practice given a small-size set high-dimensional space, traditional learning models without sparse constraint face over-fitting problem. Therefore, this paper, we introduce novel...