Convolutional neural network (CNN) for hyperspectral image classification can provide excellent performance when the number of labeled samples training is sufficiently large. Unfortunately, a small are available in images. In this letter, novel semi-supervised convolutional proposed image. The automatically learn features from complex data structures. Furthermore, skip connection parameters added between encoder layer and decoder order to make suitable learning. Semi-supervised method...

In this study, we develop a novel deep hierarchical vision transformer (DHViT) architecture for hyperspectral and light detection ranging (LiDAR) data joint classification. Current classification methods have limitations in heterogeneous feature representation information fusion of multi-modality remote sensing (e.g., LiDAR data), these shortcomings restrict the collaborative accuracy data. The proposed utilizes both powerful modeling capability long-range dependencies strong generalization...

Deep learning has achieved great success in hyperspectral image classification. However, when processing new images, the existing deep models must be retrained from scratch with sufficient samples, which is inefficient and undesirable practical tasks. This paper aims to explore how accurately classify images only a few labeled i.e., few-shot Specifically, we design classification model based on relational network train it idea of meta-learning. Firstly, feature module relation can make full...

This article proposes a novel hierarchical residual network with attention mechanism (HResNetAM) for hyperspectral image (HSI) spectral-spatial classification to improve the performance of conventional deep learning networks. The straightforward convolutional neural network-based models have limitations in exploiting multiscale spatial and spectral features, this is key factor dealing high-dimensional nonlinear characteristics present HSIs. proposed can extract features at granular level, so...

In recent years, the wide use of deep learning based methods has greatly improved classification performance hyperspectral image (HSI). As an effective method to improve convolution networks, attention mechanism is also widely used for HSI tasks. However, majority existing mechanisms are on layer, and accuracy still margins improvement. Motivated by latest self in natural language processing, a transformer proposed this paper. Specifically, along spectral dimension spatial explored...

In this article, we propose a novel multiscale deep learning network with self-calibrated convolution (MSNetSC) for hyperspectral and light detection ranging (LiDAR) data collaborative classification. Conventional methods have limitations in extracting features at granular level from multimodality fusing these context-awareness way, which will severely restrict the performance of LiDAR joint The proposed utilizes hierarchical residual structure combined to extract different receptive fields,...

The difficulties of obtaining sufficient high-quality labelled samples have always been one the important factors hindering practical application hyperspectral images (HSI) classification. regular deep learning models only attempt to mine discriminant and informative features in target HSI. Therefore, satisfactory results cannot be obtained with a few because their huge parameter space fully trained. To this end, simple effective framework is proposed utilizing idea meta-learning improve HSI...

The existing hyperspectral image (HSI) classification encounters the obstacle of improving accuracy with limited labeled samples. In this context, as a typical implementation meta-learning, few-shot learning (FSL) makes model learn by episodic training on source HSI, which has achieved significant improvements in small sample target HSI. However, FSL methods lack explicit consideration and exploration association between pixels, especially intraclass interclass pixels support set query set....

Recently, hyperspectral image (HSI) classification has become a focus of research. However, the complex structure an HSI makes feature extraction difficult to achieve. Most current methods build classifiers based on handcrafted features computed from raw inputs. The design improved 3-D convolutional neural network (3D-CNN) model for is described. This extracts both spectral and spatial dimensions through application convolutions, thereby capturing important discrimination information encoded...

Recently, deep learning models based on convolutional neural networks (CNN) remain dominant in hyperspectral image (HSI) classification. However, there are some problems CNN models, such as not good at modeling the long-distance dependencies and obtaining global context information. Different from existing CNN-based an innovative classification method transformer model is proposed to further improve accuracy of HSI. Specifically, first extracts extended morphological profile (EMP) features...

Although deep learning-based approaches have made significant progress in remote sensing image classification, the supervised learning paradigm has shortcomings under a limited number of labeled samples, which restricts classification performance to great extent. In this article, we investigate an effective self-supervised feature representation architecture (SSFR) for multimodal images few-shot land cover classification. Specifically, exploit multiview strategy construct multiple views from...

Formulae display:?Mathematical formulae have been encoded as MathML and are displayed in this HTML version using MathJax order to improve their display. Uncheck the box turn off. This feature requires Javascript. Click on a formula zoom.

Object detection in urban scenarios is crucial for autonomous driving intelligent traffic systems. However, unlike conventional object tasks, urban-scene images vary greatly style. For example, taken on sunny days differ significantly from those rainy days. Therefore, models trained day may not generalize well to images. In this paper, we aim solve the single-domain generalizable task scenarios, meaning that a model one weather condition should be able perform any other conditions. To...

Formulae display:?Mathematical formulae have been encoded as MathML and are displayed in this HTML version using MathJax order to improve their display. Uncheck the box turn off. This feature requires Javascript. Click on a formula zoom.

Recently, deep learning models have achieved remarkable results in hyperspectral imagery (HSI) classification. However, most tend to struggle due the serious overfitting problem under condition of small sample. For this purpose, we design an end-to-end classification framework and optimize it with idea meta-learning, improve accuracy robustness HSI sample Specifically, embedding network based on 3D convolution is used extract spatial-spectral features HSI, temporal adopted analyse abstract...

Abstract To address the existing problems of capsule networks in deep feature extraction and spatial‐spectral fusion hyperspectral images, this paper proposes a image classification method that combines residual 3D network Markov random field. Based on method, features images are extracted using convolutional structure, vector capsules obtained by initial layer mapped into probability via dynamic routing mechanism to construct map, spatial structure results is regularised field further...

This paper introduces a self-adaptive weighted average method of image fusion for hyperspectral imagery that utilizes recently developed theory Compressive Sensing. In the proposed algorithm, images are transformed into Fourier Domain and sampled in Double-star shaped sampling pattern. Then fused with principle. Finally reconstructed by Minimum Total Variation algorithm. Results presented on real data collected Shandong, China multispectral obtained London. Experimental comparison these...

Recent research on hyperspectral image (HSI) classification has primarily focused deep learning methods. Although these methods can automatically mine HSI features, they typically require many labelled samples to ensure sufficient performance. When there are fewer training samples, the manual feature extraction design rules critical for classification. Considering small sample problem, a slow spatial–spectral method is proposed in this study. The achieve high accuracy by using features....

The genetic factors causing cleft lip with or without palate (CL ± P) are still unclear. SNPs in FOXE1 gene were associated CL P. However, the results have been inconsistent.We explored associations of four and P by a family based study.128 children their parents recruited. rs3758249 rs1867277 genotyped high-resolution melting curve (HRM) method, whereas rs1443434 rs907577 Sequenom MassARRAY® method. software PLINK, FBAT FAMHAP used for analyzing data.rs1867277 was (Pm = 0.0395). patients...

This work was to investigate TiO2 nanocrystalline film material in heart valve replacement (HVR) and the effect of papaverine infusion through aortic root before cardiac self-recovery during HVR. films were prepared by radio frequency (RF) reactive sputtering. The crystallization characteristics surface morphology observed X-ray diffraction scanning electron microscopy, anti-platelet adhesion anti-coagulation properties analyzed. 86 patients with disease selected all underwent They randomly...

With the rapid development of remote sensing data acquisition technology, there are multimodal images over same observed scenes. These could provide complementary valuable information for land cover classification. In this article, we propose a novel self-supervised feature learning and few-shot classification model images, called S2FL. Specifically, contrastive architecture is investigated to learn spatial representations from very high resolution (VHR) image. And spectral features...