Recent studies have shown that deep domain adaptation (DA) techniques good performance on cross-domain hyperspectral image (HSI) classification problems. However, most existing HSI DA approaches directly use networks to extract features from the data, which ignores detailed information of in spectral and spatial dimensions. To effectively exploit spectral–spatial joint for HSIs, we propose a two-branch attention adversarial (TAADA) network this article. In TAADA network, feature extraction...

Traditional remote sensing (RS) image classification methods heavily rely on labeled samples for model training. When are unavailable or have different distributions from that of the to be classified, may fail. The cross-domain cross-scene is developed this case where an existing training and unknown scenes domains classification. distribution inconsistency problem caused by differences in acquisition environment conditions, scene, time or/and changing sensors. To cope with problem, many...

Recently, prototypical network based few-shot learning (FSL) has been introduced for small-sample hyperspectral image (HSI) classification and shown good performance. However, existing prototypical-based FSL methods have two problems: prototype instability domain shift between training testing datasets. To solve these problems, we propose a refined contrastive (RPCL-FSL) in this paper, which incorporates supervised into an end-to-end to perform HSI classification. stabilize refine the...

Graph convolutional network (GCN) has recently attracted great attention in hyperspectral image (HSI) classification due to its strong ability aggregate information of neighborhood nodes. However, a GCN model usually suffers from the over-smoothing problem (i.e., all nodes' representations converge stationary point) when number layers is increased. In addition, GCNs always work on superpixel-level nodes reduce computational cost, so pixel-level features cannot be well captured. To deal with...

Abstract Recently, transformer‐based networks have been introduced for the classification of hyperspectral image (HSI). Although methods can well capture spectral sequence information, their ability to fuse different types information contained in HSI is still insufficient. To exploit rich spectral, spatial and semantic HSI, a novel spatial‐spectral feature fusion transformer (S3FFT) network proposed this study. In S3FFT method, attention efficient channel (ECA) modules are employed...

Deep learning has been extensively used for hyperspectral image (HSI) classification with significant success, but the of high-dimensional HSI datasets a limited amount labeled samples is still great challenge. Few-shot (FSL) shown excellent performance in solving small-sample problems. However, most existing FSL methods usually suffer from prototype instability and domain shift. In order to address these problems, this paper proposes category-specific self-refinement contrastive (CPSRCL)...

Recently, the adversarial domain adaptation (ADA) methods have been widely investigated and applied in cross-domain hyperspectral image (HSI) classification. However, most ADA algorithms aim to align distribution without focusing on class separability of aligned target features information samples within domain. To address these issues, a new framework based calibrated prototype dynamic instance convolution (CPDIC) is proposed this paper for cross HSI The CPDIC composed generator,...

By means of a sparse collaborative representation mechanism, sparse-representation-based classifiers show superior performance in hyperspectral image (HSI) classification. Exploiting the similarity and distinctiveness HSI neighboring pixels, we propose new nearest regularized joint (NRJSR) classification method this letter. In process central test pixel, weights different pixels coefficients training samples are optimized simultaneously within sparsity model, which can obtain adaptive with...

Recently, the Transformer-based model has shown superior performance in hyperspectral image classification (HSIC) due to its excellent ability long-term dependencies on sequence data. An important component of Transformer is tokenizer, which can transform features into semantic token sequences (STS). Nonetheless, Transformer's tokenization strategy hardly representative local relatively high-level semantics because global receptive field. Mamba-based methods have even stronger spatial...

In recent years, deep neural networks have been widely used for hyperspectral image (HSI) classification and shown excellent performance using numerous labeled samples. The acquisition of HSI labels is usually based on the field investigation, which expensive time consuming. Hence, available are limited, affects efficiency methods. To improve while reducing labeling cost, this article proposes a semisupervised learning (DL) method classification, named pyramidal dilation attention...

Due to the large volume of sewage in China, efficiency water consumption evaluated by traditional model may be inaccurate. This paper evaluates more scientifically. First, this uses CCR evaluate resource and environmental separately. The latter is generally lower than former, which means issue pollution serious problem consumption. Then, integrally an eco-inefficiency focuses on undesirable outputs. results are good agreement with model: (1) Only Beijing, Tianjin, Shanghai eco-efficient...

In recent years, with the continuous development of deep learning (DL), neural networks have demonstrated good results in large-sample hyperspectral image (HSI) classification. However, practice, labels are often limited. order to use fewer labeled samples without degrading classification performance, this letter proposes a new semi-supervised method named prototype and active network (PALN), which integrates DL, (AL) (PL) into framework. After training DL small number available labels, high...

Dimensionality reduction with prior information is considered. The semi-supervised Laplacian eigenmap algorithm proposed. It shown that the performance of dimensionality algorithms can be improved by taking into account label data. data analysis and experiments show validity our algorithm.

In a multi-projector display system, geometric calibration is usually time-consuming and tedious. this paper, we propose tiled method with geometry parameters estimation. The used for analytic curved surface, such as cylindrical or spherical shaped display. process automatic by using computer vision techniques; in order to estimate the screen projector parameters, camera stereo pair given internals reconstruct 3D points on screen. parametric approach, doesn't have cover entire projectors can...

In the last few years, many known works in learning theory stepped over classical assumption that samples are independent and identical distribution investigated performance based on non-independent samples, as mixing sequences (e.g., [Formula: see text]-mixing, text]-mixing etc.), they derived similar results with investigation sample assumption. Negative association (NA) sequence is a kind of significant dependent random variables plays an important role sequences. It widely applied to...

Abstract Transformer has achieved outstanding performance in hyperspectral image classification (HSIC) thanks to its effectiveness modelling the long‐term dependence relation. However, most of existing algorithms combine convolution with transformer and use for spatial–spectral information fusion, which cannot adequately learn fusion features images (HSIs). To mine rich spatial spectral features, a two‐branch global (GSSFT) model is designed this paper, (SSIF) module fuse branches. For...

In this paper, we propose a learning scheme for regression generated by atomic norm regularization and data independent hypothesis spaces. The spaces based on polynomial kernels are trained from finite datasets, which is of the given sample. We also present an error analysis proposed algorithm with kernels. When dealing algorithms kernels, main difficulty. estimate local reproduction formula. Better estimates derived applying projection iteration techniques. Our study shows that has fast...

Known as input in the Numerical Weather Prediction (NWP) models, Microwave Radiation Imager (MWRI) data have been widely distributed to user community. With development of remote sensing technology, improving geolocation accuracy MWRI are required and first step is estimate error accurately. However, traditional method, such coastline inflection method (CIM), usually has disadvantages low poor anti-noise ability. To overcome these limitations, this paper proposes a novel ℓ p iterative...

The MPEG video data includes three types of frames, that is: I-frame, P-frame and B-frame. However, the I-frame records main information data, B-frame are just regarded as motion compensations I-frame. This paper presents approach which analyzes stream in compressed domain, find out key frame by extracting Experiments indicated this method can be automatically realized it will lay foundation for processing future.

Compared with [Formula: see text]-regularization algorithm, greedy algorithm has great advantage in computational complexity. In this paper, we consider the penalized empirical relaxed and analyze its efficiency fixed design Gaussian regression problem. Through a careful analysis, provide oracle inequalities case of finite infinite dictionary, respectively via choosing appropriate number iterations. Relying on those inequalities, obtain learning rate when target function lies convex hull...

Identifying and detecting pavement cracks quickly accurately for traffic safety is one of the important problems in field automatic driving. This study presents a framework crack detection on BEV (Bird's Eye View). Firstly, based binocular parallax information, captured road image transformed from perspective to as input network. The Unet with attention mechanism used selectively fuse deep shallow features identify pavement. In addition, further processing performed according results...