Search

In recent years, significant advances in unmanned aerial vehicle (UAV) technology and hyperspectral remote sensing have spurred rapid and innovative developments in UAV-based hyperspectral image (HSI) classification across a range of fields, including environmental monitoring, precision agriculture, forest health assessment, and disaster management. Compared to spaceborne platforms, the spectra of ground objects observed by UAV platforms exhibit notable variations, presenting more pronounced challenges for accurate classification. This article provides an in-depth and systematic review of UAV HSI classification techniques, systematically examining the evolution from traditional machine learning approaches, such as sparse coding, compressed sensing, and kernel methods, to cutting-edge deep learning frameworks, including convolutional neural networks, Transformer models, recurrent neural networks, graph convolutional networks, generative adversarial networks, and hybrid models. Although traditional methods demonstrate effectiveness in certain scenarios, their limitations become increasingly apparent when dealing with high-dimensional, nonlinear spectral data. In contrast, deep learning-based models excel at capturing intricate relationships between spectral and spatial features, significantly boosting classification accuracy and emerging as the dominant paradigm in the field. The WHU-Hi hyperspectral remote sensing dataset is utilized as a case study to elucidate the advantages and limitations of various deep learning methods through rigorous qualitative and quantitative comparisons. The potential of UAV hyperspectral image classification techniques in addressing high-dimensional data and complex scenarios is also thoroughly described. Furthermore, this article delves into cutting-edge research trends, such as lightweight model development, hyperspectral large models, multisource data fusion, and model interpretability, while also highlighting future trends for UAV hyperspectral remote sensing classification technology, particularly in real-time monitoring and intelligent applications.

Under special conditions such as night, rain, and fog, visible light imaging technology often performs poorly, while radar and other imaging technologies are limited due to their high costs. Infrared imaging technology, based on the principle of thermal radiation, can provide clear imaging effects under these extreme conditions and has a relatively low cost of use. Therefore, it has been widely applied in various fields, including military, medical, industrial, and agricultural applications. However, due to the limitations of infrared wavelengths and imaging technology, traditional infrared imaging devices struggle to capture rich texture information, leading to infrared images that lack texture details and have low resolution, which significantly restricts the further research and application of infrared imaging technology in various fields. In recent years, with the widespread attention to infrared imaging technology, researchers have proposed various new infrared image enhancement techniques. Despite this, the lack of texture information in ordinary infrared images leads to the enhancement effect of existing technologies being unsatisfactory. Therefore, we have conducted a systematic investigation of the research advancements in the field of infrared image enhancement, encompassing infrared image enhancement methods, related datasets, and evaluation metrics, with the aim of exploring a research solution that could potentially break through current technological limitations. Based on these investigations, we have evaluated the performance of various representative infrared image enhancement methods, with the hope of providing a reference for future research. In addition, this article also provides a comprehensive introduction to the potential applications of infrared image enhancement technology and discusses significant research directions for the future.

Currently, RGB-thermal (RGB-T) object detection algorithms have demonstrated excellent performance, but issues such as modality failure caused by fog, strong light, sensor damage, and other conditions can significantly impact the detector's performance. This article proposes a multimodal object detection method named diffusion enhanced object detection network (DENet), aiming to address modality failure problems caused by nonroutine environments, sensor anomalies, and other factors, while suppressing redundant information in multimodal data to improve model accuracy. First, we design a multidimensional incremental information generation module based on a diffusion model, which reconstructs the unstable information of RGB-T images through the reverse diffusion process using the original fusion feature map. To further address the issue of redundant information in existing RGB-T object detection models, a redundant information suppression module is introduced, minimizing cross-modal redundant information based on mutual information and contrastive loss. Finally, a kernel similarity-aware illumination module (KSIM) is introduced to dynamically adjust the weighting of RGB and thermal features by incorporating both illumination intensity and the similarity between modalities. KSIM can fine-tune the contribution of each modality during fusion, ensuring a more precise balance that improves recognition performance across diverse conditions. Experimental results on the DroneVehicle and VEDAI datasets show that DENet performs outstandingly in multimodal object detection tasks, effectively improving detection accuracy and reducing the impact of modality failure on performance.

Multitemporal remote sensing images offer multiple views of Earth observation, enabling the integration of temporal information to effectively mitigate issues caused by clouds, shadows, seasonal changes, and atmospheric reflection. This integration significantly reduces the uncertainty and noise of images, facilitating a more comprehensive and accurate extraction of essential ground object features. However, most land cover classification models based on deep learning focus on interpreting single-temporal remote sensing images, and multitemporal remote sensing datasets are relatively scarce, leading to challenges in model training. To address these issues, this article proposes a cross-temporal self-supervised learning method with superpixel mask (CTS2L-sm) for multitemporal land cover classification. Specifically, we design a superpixel-based masking strategy to replace the fixed-size patch masking strategy used in masked image modeling (MIM), which prevents the complete occlusion of small-scale remote sensing scenes and more effectively preserves structural information. To further enhance the integration of multitemporal information, we introduce a cross-temporal attention module (CTAM) in both forward and backward to extract cross-temporal features. The resulting multitemporal denoised homogeneous features, when fused with each temporal feature, significantly improve classification accuracy. CTS2L-sm combines these denoised homogeneous features extracted by CTAM with single-temporal features and simultaneously reconstructs the original remote sensing images across all temporal, extending the MIM strategy from single-temporal to multitemporal self-supervised training. Experimental results on the WUSU and FUSU datasets demonstrate the effectiveness and superiority of the proposed CTS2L-sm compared to several state-of-the-art methods.

Multispectral point cloud, with spatial and multiple-band spectral information, provides the data basis for finer land cover 3-D classification. However, spectral information is not well utilized by traditional methods of point cloud classification. Benefiting from the excellent performance of graph neural networks on non-Euclidean data, it is well suited to the joint use of spatial and spectral information from multispectral point clouds to achieve better classification performance. However, existing graph-based methods for point cloud classification rely on manual experience to construct input graph and cannot adapt to the complexity of remote sensing scenes. In this article, we propose a novel multikernel graph structure learning (MKGSL) framework for multispectral point cloud classification. Specifically, we explore the high-dimensional feature distribution properties of multispectral point clouds in Hilbert space through the use of kernel method. An innovative multiple-kernel learning mechanism is embedded into our network, which allows to obtain better mappings adaptively. Simultaneously, a series of prior constraints designed based on land cover distribution characteristics are imposed on the network training process, which leads the learned graph of the multispectral point cloud to facilitate better classification. Our method is dedicated to adaptively constructing task-oriented graph structures to improve the performance of multispectral point cloud classification. Experimental comparisons demonstrate that the proposed MKGSL performs better than several state-of-the-art methods on two real multispectral point cloud datasets.

Recently, hybrid networks, combining graph convolutional networks (GCNs) and convolutional neural networks into a unified framework, have garnered significant attention in hyperspectral image (HSI) classification. However, existing hybrid networks have the following limitations. 1) Existing methods primarily utilize simple fusion strategies such as concatenation or direct addition, resulting in the ineffective utilization of advantageous features. 2) Traditional GCNs only consider the relationship between pairs of vertices, limiting their ability to capture complex high-order and long-range correlations. In this work, a novel differential feature fusion network (DF2Net) is proposed for HSI classification. Specifically, DF2Net utilizes two subnetworks to learn features at different abstraction levels: 1) the spectral–spatial hypergraph convolutional network for capturing complex high-order and long-range correlations, and the spectral–spatial convolution network for pixel-level local information extraction. Subsequently, we introduce an advantageous feature differential enhancement fusion module, in which mutual enhancement of advantageous features from different network structures is performed, thereby improving the classification robustness of different regions in HSI. The experiments on four HSI benchmark datasets demonstrate that our DF2Net exhibits superior advantages over state-of-the-art models, particularly when the training samples are limited.

With the rapid advancement of remote sensing technology, bitemporal remote sensing change detection (CD) techniques have also seen significant progress. However, existing CD tasks still face two challenges: 1) Variations in lighting and seasonal factors complicate imaging conditions, causing pseudovariation interference, and 2) the spatial distribution and shapes of building are diverse, leading to difficulties in extracting and utilizing effective change features. In this article, we propose the spatial-temporal evolution guided change detection network (STEGNet) to capture and fully utilize rich spatial-temporal information. Specifically, we develop the chrono colorizer to mitigate pseudovariant interference by standardizing color styles and enriching time series information. In addition, we introduce the temporal-spatial guidance module, which combines spatial-temporal information to guide the decoding operation and mitigate information loss during spatial-temporal fusion, resulting in finer prediction results. Experimental results on three benchmark datasets demonstrate that STEGNet effectively suppresses pseudovariation interference and significantly improves the integrity of detection boundaries.

In recent years, UAV object tracking has provided technical support across various fields. Most existing work relies on convolutional neural networks (CNNs) or visual transformers. However, CNNs have limited receptive fields, resulting in suboptimal performance, while transformers require substantial computational resources, making training and inference challenging. Mountainous and jungle environments-critical components of the Earth's surface and key scenarios for UAV object tracking-present unique challenges due to steep terrain, dense vegetation, and rapidly changing weather conditions, which complicate UAV tracking. The lack of relevant datasets further reduces tracking accuracy. This article introduces a new tracking framework based on a state-space model called TrackingMamba, which uses a single-stream tracking architecture with Vision Mamba as its backbone. TrackingMamba not only matches transformer-based trackers in global feature extraction and long-range dependence modeling but also maintains computational efficiency with linear growth. Compared to other advanced trackers, TrackingMamba delivers higher accuracy with a simpler model framework, fewer parameters, and reduced FLOPs. Specifically, on the UAV123 benchmark, TrackingMamba outperforms the baseline model OSTtrack-256, improving AUC by 2.59% and Precision by 4.42%, while reducing parameters by 95.52% and FLOPs by 95.02%. The article also evaluates the performance and shortcomings of TrackingMamba and other advanced trackers in the complex and critical context of jungle environments, and it explores potential future research directions in UAV jungle object tracking.

Multisource remote sensing (RS) image classification is a significant research area in Earth observation, aiming to achieve more comprehensive and accurate classification of land cover by integrating data from different sensors. Due to differences in imaging mechanisms and information imbalance between multisource data, multisource RS image classification faces two major challenges as follows. 1) Synergistically capturing features from different modalities to fully exploit complementary information. 2) Adaptively fusing multisource features to overcome the imbalance between modalities and avoid redundant information. This article proposes a spectral–spatial feature extraction network with SSM–CNN (SSFNet) for the collaborative classification of hyperspectral images (HSI) and multispectral images (MSI). Specifically, SSFNet captures long-range spectral correlations in HSI through a bidirectional state–space model (SSM) and learns local correlations between adjacent channels through spectral grouping, achieving global–local spectral information mining in HSI. Simultaneously, joint spatial feature extraction for HSI and MSI data is performed using embedded weight-shared residual feature extractor based on convolutional neural network. This process involves adaptively identifying the importance of features through privatized factors in batch normalization and accurately replacing redundant features. In addition, a spatial attention module is used to further enhance spatial feature representation. Finally, to better accommodate feature distributions and enhance classification outcomes, the extracted spectral–spatial features are combined using weighted fusion, allowing for dynamic integration. Experimental results on two datasets demonstrate that the proposed SSFNet significantly outperforms other competing methods.

The compensation process for the state of polarization (SOP) from quantum signal plays an important role in the practical implementation of continuous-variable quantum key distribution (CV-QKD). In contrast to the previous scheme using an expensive digital dynamic polarization controller, we choose a cheaper manual polarization controller interfaced with a digital algorithm based on Kalman filter to achieve compensation of the polarization loss under the condition that the SOP of the fiber is relatively stable. This paper also enumerates the details of other digital signal processing method to achieve final secret key rate. And a pilot-sequential Gaussian modulated coherent state (GMCS) continuous-variable quantum key distribution system with a local local oscillator (LLO) is experimentally implemented based on the analysis of excess noise. Finally, the experimental results show that the method of polarization loss compensation takes advantages in maintaining the stability of the final secret key rate. And a secret key rate of 110 kbps is achieved within 20 km optical fiber (with 4 dB loss) under the finite-size block of $1 \times {10^{6}}$ during a more extended time in the laboratory.

A critical ideology of the existing Material Genome Project refers to the application of data and artificial intelligence to facilitate material innovation. The lack of data hinders the development of novel materials. The figures and captions in the material literature cover essential information regarding the entire document and have sufficient image sample data for research. Accordingly, how to extract figures and captions from the literature is critical to solve the lack of data. Though some PDF parsing tools are capable of extracting information from documents, they generally identify a document's figures by parsing the document into a concrete structure. As impacted by the inconsistency of the form of different journals, they commonly achieve wrong recognition results. Thus, an efficient figure and caption extraction network FCENet is proposed in the present study. Inconsistent with other extraction tools, this study first attempts to adopt instance segmentation models to detect figures and their captions, and then extract them. FCENet developed in this study builds upon BlendMask and introduces a horizontal and vertical attention module. This study splits the BlendMask detection head into two branches, i.e., figure detection and caption detection, which increases final detection accuracy and speed. This study collects nearly 3000 material documents for model training and testing. As revealed from the last experiments and results, the performance of FCENet is significantly compared with that of other existing instance segmentation models. Its box and mask mAP (mean Average Precision) are 8.51% and 12.59% higher than those of BlendMask, respectively. This study hopes that considerable material image data can be acquired via FCENet and sufficiently support image data for machine learning and data mining in the material area.

Today’s development focuses on the integration of affective meaning into products and services. Kansei Engineering (KE) is a technology that establishes the relationship between customers’ kansei and design elements to optimize the design for customers. This study aims to identify the research status and emerging trends in KE research by using visualization analysis with CiteSpace. We retrieve 2830 articles from the core collection of the Web of Science (1995–2021). First, based on the chronological distribution and background of research institutions, journals, literatures, countries and core authors, this study analyzes the main contributions, knowledge sources, interdisciplinary characteristics and major research domains of KE. Second, the research hotspots and cutting-edge technologies are intuitively presented and analyzed to reveal the knowledge structure and research frontier. Finally, this research expounds upon the evolution and developmental process of each stage and proposes the emerging trends of KE research, which are the use of big data in the Internet era, multidisciplinary cooperation, and the integration of multiple mathematical algorithms to serve multiapplication practice. This is also the orientation for further research on KE in the succeeding stages.

A twin-core sensor with high birefringence employing a photonic crystal fiber (PCF) is presented for simulated temperature measurements based on surface plasmon resonance (SPR). The internal wall of the central hole of the PCF is plated with gold and coated with graphene oxide (Au-GO) to develop an SPR sensing channel. The results of a numerical analysis with COMSOL Multiphysics show that the birefringence of this sensing structure can reach 0.0052. In addition, an average sensitivity of -12.695 nm/ °C can be achieved in the range of 0 °C~80 °C. Compared with uncoated GO sensors and existing sensing units, this sensor shows improved stability, sensitivity and birefringence.

Distributed energy trading has become an essential part of the energy trading market and provides a useful supplement to traditional centralized energy trading, but there are still problems such as opaque trading information and asymmetric user data. The blockchain technology has the advantages of traceability, trade openness, and data transparency, which is naturally suitable for distributed energy transactions. The electricity information data transmission represented by distributed energy transaction has the characteristics of real-time, which has a high-efficiency requirement on the selected blockchain technology. The consensus algorithm is the core of blockchain technology and affects the efficiency of the blockchain system. The efficiency of the existing consensus algorithms for energy transaction-oriented blockchain still needs to be improved. In this paper, a consensus resource slicing model(CRSM) is designed to meet the requirements of consensus efficiency in energy trading scenarios. Specifically, CRSM divides consensus nodes into different consensus domains for concurrent consensus, and the storage domain only stores block information without consensus. By building an experimental platform, the efficiency of CRSM was verified, the communication pressure of the blockchain system was reduced, and the consensus speed was effectively improved.

Karyotype analysis is one of the main techniques of cytogenetics through medical image processing, which has an important role in modern medical treatment and diagnosis. The process of human karyotype analysis contains two key components: Firstly, chromosomes are segmented from metaphase chromosome digital images taken under a microscope. Chromosomes then are analyzed, compared, ordered and classified one by one carefully. Under this procedure, the operation on segmentation and classification is cumbersomely time-consuming, where traditional geometric or statistical methods only have limited effect due to low accuracy. Thus, in most conditions, human effort is still heavily required to monitor the workflow and correct the errors. In this paper, we present an integrated workflow to segment out and classify chromosomes automatically using a combination of multiple input convolutional neural networks (CNN) and geometric optimization, called mCNN_GO. We investigate Mask R-CNN to segment out chromosomes from metaphase chromosome images and train the mCNN_GO to classify the sub-images. To improve the performance of the segmentation network, we adapt a new feature-based approach to synthesize images on the labeled data. Furthermore, we develop a geometric algorithm to straighten the chromosomes before classification to ensure the consistency of the data. Experimental results demonstrate that our approach significantly outperforms the state-of-the-art methods on automatic karyotype analysis.