One of the challenges in hyperspectral image (HSI) classification is that there are limited labeled samples to train a classifier for very high-dimensional data. In practical applications, we often encounter an HSI domain (called target domain) with few data, while another source may have enough Classes between two domains not be same. This article attempts use class data help classify classes, including same and new unseen classes. To address this paradigm, meta-learning paradigm few-shot...

Hyperspectral image (HSI) classification has drawn increasing attention recently. However, it suffers from noisy labels that may occur during field surveys due to a lack of prior information or human mistakes. To address this issue, article proposes novel dual-channel residual network (DCRN) resolve HSI with labels. Currently, the influence is reduced by simply detecting and removing those anomalous samples. Different such specifically designed noise cleansing method, DCRN easy implement but...

Cross-domain hyperspectral image classification is one of the major challenges in remote sensing, especially for target domain data without labels. Recently, deep learning approaches have demonstrated effectiveness adaptation. However, most them leverage unlabeled only from a statistical perspective but neglect analysis at instance level. For better alignment, existing employ entire unevaluated an unsupervised manner, which may introduce noise and limit discriminability neural networks. In...

Recently, few-shot learning (FSL) has been introduced for hyperspectral image (HSI) classification with few labeled samples. However, existing FSL-based HSI methods mainly focus on the meta-knowledge transfer between HSIs. Compared HSIs, natural images have sufficient annotated data. To utilize (base class data) to achieve accurate of HSIs (novel data), we propose a novel framework SSL (FSCF-SSL) in this article. The orientation objects is relatively unitary, whereas patches each pixel...

A novel O-(carboxymethyl)-chitosan (OCMC) nanofiltration (NF) membrane is developed via surface functionalization with graphene oxide (GO) nanosheets to enhance desalting properties. Using ring-opening polymerization between epoxy groups of GO and amino OCMC active layer, are irreversibly bound the membrane. The NF membranes surface-functionalized characterized by Fourier-transform infrared spectroscopy, X-ray photoelectron scanning electron microscopy, atomic force contact angle analyzer,...

Deep spectral-spatial features fusion has become a research focus in hyperspectral image (HSI) classification. However, how to extract more robust is still challenging problem. In this article, novel deep multilayer dense network (MFDN) proposed improve the performance of HSI The MFDN simultaneously extracts spatial and spectral based on different sample input sizes, which can abundant correlation information. First, principal component analysis algorithm performed data low-dimensional data,...

Recently, few-shot object detection based on fine-tuning has attracted much attention in the field of computer vision. However, due to scarcity samples novel categories, obtaining positive anchors for categories is difficult, which implicitly introduces foreground–background imbalance problem. It difficult identify foreground objects from complex backgrounds various sizes and cluttered backgrounds. In this article, we propose a context information refinement detector (CIR-FSD) remote sensing...

Deep domain adaptation has achieved promising results in cross-domain hyperspectral image (HSI) classification. However, existing methods often focus on aligning data distributions without sufficient consideration of separability source and target themselves. In addition, current adversarial aim to achieve similar between domains by confusing the discriminator, rather than obtaining a more compact distribution. particular, are not discriminative enough for due difficulty high-confidence...

Recently, networks consider spectral-spatial information in multiscale inputs less, even though there are some that this factor, however these cannot guarantee to get optimal features, which extracted from each scale input. Furthermore, do not the complementary and related among different features. To address issues, a deep middle-level feature fusion network (MMFN) is proposed paper for hyperspectral classification. In MMFN, fully fuses strong features extract more discriminative The...

Deep learning has achieved great success in hyperspectral image (HSI) classification. However, its relies on the availability of sufficient training samples. Unfortunately, collection samples is expensive, time-consuming, and even impossible some cases. Natural datasets that are different from HSI, such as Image Net mini-ImageNet, have abundant texture structure information. Effective knowledge transfer between two heterogeneous can significantly improve accuracy HSI In this letter, few-shot...

Hyperspectral image (HSI) has hundreds of continuous bands that contain a lot redundant information. Besides, spatial patch hyperspectral cube often contains some pixels different from the center pixel category, which are usually called interference pixels. The existence such negative effect on extracting more discriminative Therefore, in this letter, multiattention fusion network (MAFN) for HSI classification is proposed. Compared with current state-of-the-art methods, MAFN uses band...

<title>Abstract</title> In response to the challenge of insufficient target detection accuracy for UAVs in adverse environments(such as low light, dense fog, and extreme weather), this paper proposes a lightweight multi-modalfusion network, MMT-NET, UAV such conditions. The network significantlyimproves performance complex environments by fusing complementary characteristicsof infrared visible light images, combined with design. MMT-NET is based on RT-DETR framework uses MobileNetV4 backbone...

Deep learning has attracted extensive attention in the field of hyperspectral images (HSIs) classification. However, supervised deep methods heavily rely on a large amount label information. To address this problem, paper, we propose two-stage domain adaptation method for image classification, which can minimize data shift between two domains and learn more discriminative embedding space with very few labeled target samples. A is first learned by minimizing distance source based Maximum Mean...

The lack of training samples remains one the major obstacles in applying convolutional neural networks (CNNs) to hyperspectral image (HSI) classification. In this letter, accurate classification HSI with limited is investigated. Due advantages minimizing distance between same class and maximizing different classes, siamese CNN used for samples. After that, improve performance, data augmentation investigated CNN-based Specifically, pair based on CutMix proposed generate pairs or classes new...

Chengcheng Chen1*Muyao Bai1Tairan Wang1Weijia Zhang1Helong Yu2*Tiantian Pang3Jiehong Wu1Zhaokui Li1Xianchang Wang1,3,4