Search

Rice is pivotal in global food security and Rice seeding detection in precision agriculture is essential for optimizing crop productivity and efficient resource use. Currently, the spatial distribution and detection of rice seeding are manually done, which is time-consuming. As technology advances, the rapid progress in precision agriculture calls for innovative solutions like lightweight edge-computing vision models for unmanned aerial vehicle (UAV) intelligence. However, the limited computational capabilities of UAVs make it challenging to deploy complex object detection models onboard. This article proposes Li-YOLOv9, an efficient, lightweight, and precise object detection model for detection of rice seedlings from large-scale UAV remote sensing data. The proposed Li-YOLOv9 includes a 3-D feature adaptation module, convolutional block attention modules, coordinate attention module, and deepwise point convolution modules, which offer significant advantages over the original YOLOv9. The compact Li-YOLOv9 with approximately 9 million parameters is significantly lighter than the original YOLOv9 with 60 million parameters, making it ideal for resource-efficient onboard intelligence. Li-YOLOv9 is tested and validated against a real UAV rice seedling dataset. It showcases exceptional detection capabilities, achieving impressive metrics across various performance indicators, including a mean average precision of 99.60%, an $F1$-score of 96.95%, an average recall of 96.70%, and an average precision of 97.20% in comparison to the state-of-the-art. Li-YOLOv9 is highly efficient and ideal for UAV deployment for real-time detection of rice seedlings from large-scale remote sensing data.

Noise suppression in a high-level noise environment using a low-complexity method is challenging. This study proposes a low-complexity noise suppression algorithm for robot vacuum cleaner processors. We collected working noise from a robot vacuum cleaner along with speech signals and developed a method to extract the desired speech signal while estimating the noise. Our approach estimates the noise in the existing signal and converts it into the desired signal. In addition, we designed a low-complexity neural network capable of operating on mobile processors. The evaluation results demonstrate that our method achieves a performance comparable to that of highly computational methods. Notably, our method maintains superior performance when the intensity of the desired signal is low, and its performance is less degraded than that of other methods. It exhibits less degradation than existing methods, and in contrast to other neural networks, it avoids generating incorrect signals. Furthermore, we simplified the neural network architecture reducing its size by approximately 25% with minimal performance loss.

In the field of machine tools, it is general to use facilities from various vendors with different interfaces. However, manuals that vary from machine to machine and are less portable and usable, are adding the burden. In this case, the digital intelligent assistant can be used as an easy and unified interface to communicate with users and increase work efficiency. However, applying a digital assistant requires a large amount of dataset and effort separately, and there is a limitation that it only plays a role in a specific situation designed. In this paper, we propose a smart and cognitive chatbot with a dataset-constructing pipeline to solve the problems. The proposed pipeline aims to build a dataset and update source data for the chatbot automatically. It provides relevant manual context, machine status, current operation, and operation sequence, depending on the users’ intention. It utilizes various AI technologies including optical character recognition, reinforcement learning, and natural language preprocessing technologies to minimize human intervention in the dataset construction. Also, it can be applied to various heterogeneous facilities with different interfaces. Thanks to the proposed data extraction pipeline, the proposed chatbot interacts with users intuitively using conversational language, replacing cumbersome manipulation and unusable documentation. In the paper, one of the equipment, commonly used in the real world, is chosen to be applied the proposed system to the chosen facility, dealing with possible problems during the operation.

Predicting bank failures is a critical task requiring balancing the need for model explainability with the necessity of preserving data privacy. Traditional machine learning models often lack transparency, which poses challenges for stakeholders who need to understand the factors leading to predictions. In this study, we employ differentially private glass-box models, namely Explainable Boosting Machine (EBM) and Neural Additive Models (NAM), to address these issues. We analyzed data from 21,243 American banks spanning from 1969 to 2021, focusing on key financial ratios. By applying Differential Privacy (DP) to these models, we aimed to protect sensitive financial data while evaluating the trade-offs between privacy, accuracy, and explainability. Our main findings are as follows: 1) In the absence of privacy constraints, the models consistently identified Asset Turnover, Total Debt / Invested Capital, and ROE ratios as the most influential factors in predicting bank failure, in that order; 2) When the privacy budget $\epsilon \leq 1$ , only EBM maintained significant performance; 3) The reduction in explainability due to privacy protection was more pronounced for variables with initially lower explanatory power, while Asset Turnover retained its explanatory power even at $\epsilon = 0.01$ . These findings provide valuable insights for banks, policymakers, and investors, suggesting that glass-box models can offer a promising solution for reliable and explainable bank failure prediction under privacy constraints.

This document describes the adaptation of a clamshell inductive current coupler for online reflectometry testing (both frequency-domain reflectometry and spread spectrum time-domain reflectometry) to evaluate cable anomalies. The life span of safety-critical nuclear power plant cables is initially qualified for 40 years in accordance with IEEE 383 without additional testing. As plants extend their operating licenses to 60 and 80 years, justification for continued safe operation includes cable test and condition monitoring (CM) programs. Test programs traditionally involve manual interventions to disconnect the cables, perform one or several tests, then reconnect the systems, usually during refueling outages occurring only every two years. Offline testing poses an operational burden that can be minimized by online testing. This work investigates the adaptation of a clamshell inductive current coupler to inject a signal onto a cable conductor and listen for a reflected signal indicative of a damaged condition. The coupler provides >60 dB of protection, thereby allowing tests on cables up to 10 kV or more. Although the clamshell coupler is a known commercial product for cable performance testing, its use for energized cable CM constitutes a novel use case.

Industrial anomaly detection involves identifying abnormal regions in products and plays a crucial role in quality inspection. While 2D image-based anomaly detection has been extensively explored, combining two-dimensional (2D) images with three-dimensional (3D) point clouds remains less studied. Existing multimodal methods often combine features from different modalities, leading to feature interference and degraded performance. To overcome this, we propose a novel framework for unsupervised industrial anomaly detection that leverages both visual and geometric information. Specifically, we use pre-trained 2D and 3D models to extract visual features from color images and geometric features from 3D point clouds. Instead of directly fusing these features, we propose a geometric feature reconstruction network that predicts 3D geometric features from the 2D visual features. During training, we minimize the difference between the predicted geometric features and the extracted geometric features, enabling the model to learn how 2D appearance correlates with 3D structure in anomaly-free images. During inference, this learned relationship allows the model to detect anomalies: significant discrepancies between the reconstructed and actual geometric features indicate abnormal regions. Evaluated on the MVTec 3D-AD dataset, our method achieves state-of-the-art performance with an average image-level AUROC score of 0.968, surpassing previous approaches. Additionally, it provides fast inference at 8.2 frames per second with a memory footprint of only 1045 MB, making it highly efficient for industrial applications.

This paper proposes a hybrid system comprising multiple unmanned aerial vehicles (UAVs) to monitor nature. The proposed tributary mapping system consists of perception and control systems. A perception system is established for tributary mapping using three-dimensional light detection and ranging (LiDAR) semantic segmentation for surface water recognition. The system defines a water point cloud and segments the surface water area using singular value decomposition, applying the water-absorbing property of LiDAR. A path is generated through random sample consensus by calculating the center point from the segmented area. Each UAV is modeled using a continuous-time dynamic-based low-level model and a discrete event-time dynamic-based high-level model to control the multiple UAVs system. The desired plant behavior is designed considering the control objectives of the plant, and a supervisory controller is developed accordingly. A physics-based robot simulator is employed to verify the proposed tributary mapping system in perception and path generation and the performance of the hybrid system-based supervisory controller.

With the growing global emphasis on environmental protection policies, renewable energy generation systems have become widely adopted. In particular, photovoltaic (PV) systems have gained popularity for their ease of management, while digital twin (DT) systems are being developed to enable real-time monitoring and management of the systems. Furthermore, the DT systems simulate the operations of the physical systems in real-time based on the data collected from various sensors. To this end, a novel sensor data generation model based on numerical weather prediction (NWP) data is proposed to forecast the future operations of PV systems using DT systems. The proposed model utilizes a hybrid data-driven model structure combining supervised learning-based long short-term memory (LSTM) and unsupervised learning-based generative adversarial network (GAN) to enhance both average and variance accuracy. Specifically, TransTimeGAN is proposed, which combines TimeGAN with Transformer to effectively capture 15-min variability. For practical applicability, the proposed model is trained and validated using data from a self-developed PV DT system. To evaluate the effectiveness of the proposed model, the similarities between normalized real and generated data are compared using a range of error metrics and statistical metrics. For representative error metrics, the proposed model achieves a mean squared error (MSE) of 7.84e-3 and a dynamic time warping (DTW) score of 1.3769. Regarding representative statistical metrics, the model achieves a Kullback-Leibler divergence (KLD, max-normalized) of 0.9591 and a standard deviation similarity (SDS) of 0.9671. The experimental results demonstrate that the proposed model delivers superior performance in generating data compared with various data-driven models across a range of numerical metrics and visual assessments.

One of the first steps in the retinal image preprocessing is cropping the Field of View (FOV) area and scaling it into a template of a predefined size. Fundus cameras of different producers record digital images of the retina of various sizes, and the FOV area containing helpful information can be from 43 to 98% of the image area. For automated analysis of retinal images and detection of DR, it is necessary to segment the FOV region and cut it out from the image. This is important to preserve microaneurysms and small capillaries in the retinal image as much as possible, since neural network methods always reduce the original images to a predefined size. In this paper, we propose a universal method for FOV segmentation based on the ideas of histograms and thresholds. We compared 11 methods for segmenting FOV regions on the four most commonly used retinal image grayscale representations. In total, we compared 35 variants of segmentation and evaluated the obtained results by four functions: Jaccard index, Matthews correlation coefficient (MCC), accuracy and balanced accuracy. All options were tested on 7000 images from nine of the largest databases. The images were generated by 100 different fundus cameras. The following observations have been extracted through extensive comparative experiments namely: 1) segmentation of the FOV area should be performed on the grayscale image obtained from the red channel; 2) for more accurate segmentation, a logarithmic transformation should be applied to the grayscale image; 3) the FOV area mask can be segmented by a global threshold calculated by Otsu’s method; 4) global thresholding based on analysis of histogram peaks does not provide advantages over binarization by Otsu’s method applied to the logarithmic transformation of the image.

Chinese characters are an essential means of communication in the East Asian cultural regions. Chinese characters are characterized by many strokes and complex structures, some of which are very similar. However, the misrecognition of messy writing can significantly decrease the accuracy of Optical Character Recognition (OCR) systems. Therefore, filtering the messy writing is critical. We propose a novel federated learning-based deep learning model to classify messy Chinese handwriting, addressing the challenges posed by complex character structures and variations in writing quality. To validate the proposed method, we conducted experiments to compare our approach with global, local, FedAVG, and IPA federated learning methods. We evaluated our approach using the CNN, ResNet50V2, NASNetLarge, EfficientNetV2B0, EfficientNetB0, and swin transformer models. The results showed a 301.2% improvement in the test accuracy and a 241.54% increase in the area under the curve (AUC) score with the proposed method, particularly when EfficientNetB0 was trained using the global method. These findings confirm that the proposed model effectively classifies Chinese messy handwriting. Additionally, when the data were fed into the OCR system, the match rate between the image and the recognized characters improved by up to 7.14%.

Odd-bit quadrature amplitude modulations (QAMs) can achieve higher data rates than even-bit QAMs in some signal-to-noise (SNR) regions but they have not been used in wireless communication standards due to its high demapping complexity. In this paper, we propose a novel constellation for odd-bit QAM, called dual square-QAM (DSQAM). It consists of two regular even-bit square-QAMs spaced apart by a shift value in the diagonal direction. We also propose a two-layer coding which effectively exploits the structural property of the constellation and a successive demapping to use the decoding result of the first layer for the second layer. The proposed DSQAM is enhanced by the two-layer coding and successive demapping, which consists of multi-level coding and bit-interleaved coded modulation (BICM). The asymptotic information rate (AIR) under the coding scheme is maximized for a given channel state by adjusting the shift value. The AIR analysis shows that the DSQAM with the two-layer coding and the successive demapping is the best among conventional QAMs with the BICM for a wide range of SNR values. Based on the structure of the constellation, we propose a low-complexity implementation of the max-log approximation to calculate the log-likelihood ratio for practical use. It is shown via simulations over the additive white Gaussian channel that the proposed DSQAMs with the coding scheme have significant block error rate gains over the conventional odd-bit non-square-QAM (NSQAM) and cross-QAM (XQAM) with bit-interleaved coded modulation. The computational complexity for demapping is reduced by up to 10% and 59% compared to NSQAM and XQAM, respectively.

Rolling bearing faults frequently cause rotating equipment failure, leading to costly downtime and maintenance expenses. As a result, researchers have focused on developing effective methods for diagnosing these faults. In this paper, we explore the potential of Machine Learning (ML) techniques for classifying the health status of bearings. Our approach involves decomposing the signal, extracting statistical features, and using a feature selection employing Binary Grey Wolf Optimization. We then employ four different classifiers such as Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Decision Tree (DT), and Random Forest (RF) to diagnose faults based on the reduced set of features. To evaluate the performance of our methods, we utilize several performance indicators. Our results demonstrate that four Machine Learning methods can achieve a high-accuracy fault classification result of 99.85%, better than state-of-the-art methods, highlighting their potential for use in predictive maintenance applications.

A 1-kS/s 12-bit successive approximation register (SAR) analog-to-digital converter (ADC) which performs burst conversion is proposed to reduce the loss of the sampled analog signal due to leakage current in the capacitors of the capacitor-based digital-to-analog converter (CDAC). In order to perform burst conversion despite having a low sampling rate of 1 kHz, the proposed 12-bit SAR ADC is synchronized to 2 MHz, performing data conversion for 20 cycles and entering sleep mode for 1980 cycles to wait for the next data conversion. The capacitor-resistor DAC (CRDAC) used in the proposed SAR ADC reduces the area required for 12 bits of resolution, and the voting process allows for accurate quantization of the low 3 bits. The proposed SAR ADC is implemented using a 180-nm 1-poly six-metal CMOS process with a supply of 1.8 V. The proposed burst conversion for low-speed SAR ADCs improves the dynamic performance of ENOB from 10.82 bits to 11.59 bits for an input signal with a frequency of 450.0 Hz at a sampling rate of 1 kHz. The sleep mode operation between the burst conversions reduces the average power consumption of the SAR ADC from $47.2~\mu $ W to $9.0~\mu $ W by reducing static power consumption.

The main objective of this study is to develop and evaluate an effective integrated learning approach for the automatic classification of Municipal Solid Waste using the TrashBox dataset, comprising 17,785 images, to improve the sorting of recyclable waste materials, reduce landfill usage, and promote sustainable environmental practices. Initially, four deep learning models—DenseNet161, ResNet152, and MobileNetV3 variants—are explored to determine the most suitable feature extraction method. During the feature selection phase, three metaheuristic algorithms—Whale Optimization Algorithm (WOA), Salp Swarm Algorithm (SSA), and Harris Hawk Optimization (HHO)—are applied to filter out irrelevant features and retain significant ones. These selected features are then fed into machine learning classifiers—Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR), and K-Nearest Neighbor (KNN)—for final predictions. The DenseNet161-HHO-SVM combination outperforms other models in this study, achieving the highest accuracy and lowest execution time. This integrated approach also demonstrates superior performance (97.45%) compared to previous state-of-the-art models on the same dataset, with the data processing and method integration phases having substantial impacts.

In server consolidation environments, co-locating I/O-intensive and other workloads on a single physical machine often leads to significant performance degradation for memory-intensive workloads due to the memory copies required during copy-based I/O (buffered I/O) operations. Zero-copy I/O has been proposed as a solution to this problem by eliminating unnecessary memory copying between user space and kernel space, thereby improving performance. However, existing zero-copy I/O schemes fail to provide both transparent copy avoidance via standard read/write system calls and the benefits of kernel-level caching, which limits their general applicability. We introduce the z-READ framework, a solution that enables transparent zero-copy read I/O operations while leveraging kernel-level caching. It addresses two primary practical challenges: minimizing page remapping overhead by reducing the number of TLB shootdowns, and ensuring robust I/O performance under high copy-on-write fault conditions using adaptive hybrid I/O mode switching based on a performance loss model and historical I/O patterns. We implement a z-READ prototype on Linux Kernel 4.12.9 and conduct extensive evaluations. Our results demonstrate that, while traditional copy-based I/O configurations in co-located environments lead to an average performance slowdown of 84% for memory-intensive workloads, z-READ restricts this impact to only an 11% average slowdown.

This paper considers wireless power transfer (WPT) for powering low-power devices in massive Machine Type Communication (mMTC) using a distributed massive multiple-input multiple-output (MIMO) system. Each Internet of Things (IoT) device can be served by one or more access points (APs) which is equipped with a massive antenna array. During each time slot, each IoT device transmits pilot sequences to enable APs to perform channel estimation. This process is followed by the WPT using conjugate beamforming. The approach to transmission power control is formulated as a non-convex optimization problem aiming to maximize the total accumulated power achieved by all IoT devices while taking into account the power weights at the APs, pilot power control at the IoT devices, and the non-linearity of practical energy harvesting circuits. An alternating optimization approach is adopted to solve it iteratively, achieving convergence within just a few iterations. Furthermore, since the number of IoT devices might be enormous in mMTC networks, we propose a pilot sharing algorithm allowing IoT devices to reuse pilot sequences effectively. Numerical results are provided to validate the effectiveness of the proposed power control algorithms and the pilot sharing scheme. It is shown that by allowing IoT devices to share the pilot sequences instead of employing the orthogonal pilots, the per-user accumulated performance is enhanced considerably, especially when the number of IoT devices is large relative to the coherence interval. The advantage of using distributed massive MIMO compared to its collocated counterpart is demonstrated in terms of the per-user accumulated power.

Service provisioning is of paramount importance as we are now heading towards a world of integrated services giving rise to the next generation of service ecosystems. The huge number of service offerings that will be available to customers in future scenarios require a novel approach to service registry and discovery that allows customers to choose the offerings that best match their preferences. One way to achieve this is to introduce the provider’s reputation, i.e., a quality indicator of the provisioned service, as an additional search criterion. Now, with blockchain technology in our hands, automated regulation of service-level agreements (SLAs) that capture mutual agreements between all involved parties has regained momentum. In this article, we report on our full-fledged work on the conception, design, and construction of a platform for SLA-minded service provisioning called PenChain. With our work, we demonstrate that penalty-aware SLAs of general services–if represented in machine-readable logic and assisted by distributed ledger technology–are programmatically enforceable. We devise algorithms for ranking services in a search result taking into account the digitized values of the SLAs. We offer two scenario-based evaluations of PenChain in the field of precision agriculture and in the domain of automotive manufacturing. Furthermore, we examine the scalability and data security of PenChain for precision agriculture.

Photovoltaics (PV) have attracted considerable attention owing to their longer lifespans and higher generation potentials compared with other renewable energy sources. However, the intermittent nature of PV systems can degrade the power quality, hindering their widespread adoption. To mitigate the power-quality degradation resulting from the proliferation of PV, high forecasting accuracy is essential. However, missing data during the development of forecasting models can degrade performance. Therefore, appropriate imputation procedures are required. Typically, linear imputation is used. However, there is a tendency for the performance of the forecasting model to decline owing to errors between the actual and imputed values. In this study, we addressed missing PV power data using direct deletion, linear imputation, k-nearest neighbors imputation, and Generative Adversarial Imputation Nets. Subsequently, to assess the impact of weather variability on the imputation performance, we employed the “sky status” to categorize the replaced data and analyze whether differences in imputation performance emerged. Finally, we developed a PV forecasting model using the replaced data and evaluated its forecasting performance.

This paper describes a design of dual-polarized composite patch-monopole antenna with reconfigurable radiation pattern. The design consists of a double differential-feed patch loaded with four vertical monopoles symmetrically. The monopoles are connected to or disconnected from the ground plane (GND) by changing the ON/OFF states of p-i-n diodes. By adjusting the connections between the monopoles and GND, the pattern reconfigurability (i.e., one widebeam and three narrow-beam modes) can be achieved for each polarization. The proposed antenna can switch the beam in two dimensions separately $(xz-$ and $yz-$ plane). The proposed antenna is characterized computationally and validated experimentally. The measurements result in an overlapping 10-dB return-loss bandwidth of $2.40 - 2.51$ GHz and a port-to-port isolation ≥20 dB in all the modes. In the far field, the prototype can deploy switched beam flexibility. Its radiation characteristics are confirmed by the far-field measurements, which shows the reconfigurable patterns for both polarizations.

Recommender systems (RS) are substantial for online shopping or digital content services. However, due to some data characteristics or insufficient historical data, may encounter considerable difficulties impacting the quality of their recommendations. This study introduces the clustering-based frequent pattern mining framework for recommender systems (Clustering-based FPRS) - a novel RS constituting several recommendation strategies leveraging agglomerative clustering and FP-growth algorithms. The developed strategies combine the generated frequent itemsets with collaborative- and content-filtering methods to address the cold-start problem, which occurs whenever a new user or item enters the system. In such cases, the RS has limited information about the new user or object. Thus, the recommendations may be inaccurate. The experimental evaluation on several benchmark datasets showed that Clustering-based FPRS is superior to state-of-the-art and could effectively alleviate the cold-start problem.

This paper presents the design and implementation of an X-band horn antenna, which operates over a frequency range of 7 – 9 GHz for small satellite communication systems in Low Earth Orbit (LEO), employing metamaterials and 3D printing technology. It was fabricated using a polymer-based material through a 3D printer and subsequently coated with copper. The horn antenna with a metamaterial lens (meta lens) and septum polarizer is fabricated as a single structure, eliminating the need for additional mechanical structures and parts for assembly of the antenna. This results in a significant reduction in weight compared to a conventional antenna manufactured using traditional methods. To enhance the directivity of the conical horn antenna, the meta lens horn antenna incorporates a periodic lattice-shaped metamaterial. Additionally, a septum polarizer is integrated to achieve circular polarization. The metamaterial lens used in this paper exhibits Near-Zero Refractive Index (NZRI) characteristics within the operating frequency band. It is positioned inside the conical horn antenna to compensate for the field phase difference between the horn’s edge and vertex. This compensation leads to a uniform phase distribution across the aperture, resulting in improved antenna directivity. The simulated and measured results show that the meta lens enhances the gain by over 1 dB within the frequency band of interest compared to the conventional horn of the same size. Furthermore, it exhibits a return loss below 20 dB. The proposed antenna was fabricated as a monolithic lightweight structure using the Fused Deposition Modeling (FDM) technique and subsequently metalized through the electro-less plating process. The stability of electrical properties has been verified in a thermal vacuum environment test to ensure its functionality in an LEO environment, characterized by significant temperature fluctuation.

In this paper, we investigate the impact of two pulse shapes on the performance of a real-time free-space optical communication link. The two candidate pulse shapes are the square-root raised cosine and Xia pulse, respectively which are tested as the basis function for multi-band carrier-less amplitude and phase modulation. We first develop a real-time system based on a Xilinx Zynq ZCU102 system-on-chip platform utilising a high-resolution analogue-to-digital-converter. We then generate multi-band carrier-less amplitude and phase modulation formats using it and test the error vector magnitude whilst varying parameters. We emulate the fog environment utilising neutral density filters and evaluate the error performance of the link under increasingly poor visibility conditions. We show that contrary to previous reports, the SRRC pulse shape offers superior performance over the first-order Xia pulse in the FSO environment operating at data rates exceeding 1 Gb/s.

This work investigates the energy-outage performance of aerial unmanned aerial vehicles (UAV)-enabled simultaneous wireless information and power transfer (SWIPT) deployments. Focus on the energy preservation challenges for battery-powered UAVs to maintain hovering in the sky, we propose two energy-based antenna approaches, including energy-based single antenna selection (ESAS) and energy-based maximal antenna transmission (EMAT), to improve the energy-outage probability (EOP). In order to evaluate EOP, we derive closed-form approximations for two proposed ESAS and EMAT schemes as well as the asymptotic EOP formulations. In addition, we deduce the system diversity gain, the performance gap between the two proposed schemes, and guidelines for some useful information in system designs. Moreover, we also carry out an optimization of aerial UAV position to minimize the EOP when dealing with certain service areas. Through the Monte-Carlo method, extensive numerical results show that: 1) Simulation results align well with our analytical approximations; 2) For achieving the same EOP requirement, the use of EMAT consumes less than 3 dB of the transmit power compared to ESAS; 3) Compared to ESAS, the adoption of EMAT with more than four antennas at the UAV supply and aerial hovering UAV nodes reduced EOP by more than 100 times; 4) The considered SWIPT-PS mechanism results in a 10 times better EOP improvement than the SWIPT-TS one; and 5) Optimizing the position of aerial hovering UAVs lead to the EOP minimization significantly.

Battery electric vehicles (BEVs) are increasingly recognized as the key to the power industry’s shift towards decentralization, carbon neutrality, and electrification because of their potential as flexible assets. Advances in grid integration, communication standards, smart charging, vehicle-to-grid technology, and electric vehicle service platforms are crucial for effectively leveraging BEVs. However, realizing these potential hinges on the active participation of electric vehicle owners, which is a vital yet challenging aspect. This study proposes a gamification-based incentive model to boost the participation of BEV owners in demand response programs, focusing on enhancing grid flexibility. A pilot project involving BEV drivers and charging-point operators was conducted on Jeju Island, South Korea. The significant contributions of this study include the development of both hardware and software systems for the gamification model and an ANOVA analysis to assess its impact. Compared with conventional economic incentives, the application of gamification-based incentives had effects of 13.8–69.2 percent from a demand increase perspective and 2.0–91.4 percent from a demand reduction perspective.

This paper proposes a novel opportunistic full-duplex (FD) scheme switching among three duplex modes (DMs) of half duplex (HD), three node FD (TNFD) and bidirectional FD (BiFD) for the cellular systems. To this end, we first derive the conditions, by which one of the three DMs achieves larger average spectral efficiency (ASE) than the other two DMs. Specifically, we derive the minimum required self inteference (SI) cancellation factors of user equipments (UEs)/base station for BiFD/TNFD to achieve the highest ASE among the three duplex modes respectively. We also show that they are given as an explicit function of the static system parameters: cell radius, path loss exponent and target SNR. Notably, for the typical small cell radii, BiFD with the state-of-art SI cancellation techniques significantly outperforms TNFD (which undergoes high inter-UE interference in small cell systems). Compared to the single mode or two mode duplex schemes, substantial ASE gains are achieved by three mode duplex switching according to the aforementioned static system parameters. Moreover, the proposed slow (i.e., a static system parameter-dependent) switching scheme achieves almost identical ASE to that of the fast switching scheme that requires high complexity to calculate and compare the instantaneous spectral efficiencies (SEs) of the three DMs.

A source localization of a chemical gas dispersion is important to understand the situation and facilitate effective responses to incidents. The source point can be simply identified on a map assuming an infinite number of chemical sensors in every region. However, it is limited to manufacture and install a number of sensors due to the several conditions. Consequently, when dispersion begins in a region without sensors, tracking the movement of chemical gas cloud and pinpointing its origin is challenging since obtaining information during the early stages of an incident is difficult. Therefore, this study proposes a recursive tracking model based on a transformer for the source localization of a chemical gas dispersion. This machine learning model recursively tracks the past spatial dispersion distribution using sequential frames from a chemical concentration map. When N sequential image frames $(C_{1}, C_{2}, \ldots, C_{N})$ are input, the model outputs a single image frame $(C_{0})$ , which represents the image frame prior to the input time sequence. Thus, the model predicts the origin by successively predicting past image frames. The gas concentration data for this study were sourced from a gas dispersion model based on Fick’s law. For each scenario, 15000 datasets were generated to train and test the model, each dataset comprising 15 image frames that describe various aspects of gas dispersion. The performance of source localization was assessed based on accuracy. The model was validated using 1500 test data obtained from the model based on Fick’s law and 100 test data obtained from the Nuclear Biological Chemical Reporting and Modeling System (NBC_RAMS).

Quantifying the intensity and frequency of climatic extremes under the impacts of climate change is crucial for effective water resource management. In this study, we utilize the soil and water assessment tool hydrological model, robust indices, e.g., standardized precipitation index and standardized streamflow index as well as the interquartile range method for a comprehensive analysis of the river flow response to future climate scenarios towards 2090. Four general circulation models under two shared socioeconomic pathways have been used, including BCC-CSM2-MR, CanESM5, MIROC6, and MRI-ESM2-0. We aim to reveal the future impacts of extreme events and their potential consequences for local livelihoods and human well-being in the Srepok River basin—a major tributary of the Mekong River basin in Southeast Asia. Our findings include the following: 1) a significant discrepancy between extreme events found with more flood events projected toward 2090; 2) a shift in precipitation patterns with an increase in intensity is observed; and 3) a correlation between climatic extremes and regional characteristics has been identified. This work provides valuable insights into the anticipated changes in climatic extremes under the impacts of climate change and serves as the scientific basis for stakeholders and decision-makers to develop adaptative strategies and sustainable plans to enhance the region's resilience.

Quantifying muscle strength is an important measure in clinical settings; however, there is a lack of practical tools that can be deployed for routine assessment. The purpose of this study is to propose a deep learning model for ankle plantar flexion torque prediction from time-series mechanomyogram (MMG) signals recorded during isometric contractions (i.e., a similar form to manual muscle testing procedure in clinical practice) and to evaluate its performance. Four different deep learning models in terms of model architecture (based on a stacked bidirectional long short-term memory and dense layers) were designed with different combinations of the number of units (from 32 to 512) and dropout ratio (from 0.0 to 0.8), and then evaluated for prediction performance by conducting the leave-one-subject-out cross-validation method from the 10-subject dataset. As a result, the models explained more variance in the untrained test dataset as the error metrics (e.g., root-mean-square error) decreased and as the slope of the relationship between the measured and predicted joint torques became closer to 1.0. Although the slope estimates appear to be sensitive to an individual dataset, >70% of the variance in nine out of 10 datasets was explained by the optimal model. These results demonstrated the feasibility of the proposed model as a potential tool to quantify average joint torque during a sustained isometric contraction.

A compact dual-band tripolarized antenna with simple structure and high isolation operating at 2.45 GHz and 3.5 GHz bands is presented. The design is composed of a slotted patch and a monopolar patch connected together by four vias. The antenna uses a double differential-fed scheme for x- and y-horizontally polarized broadside radiations and a single-ended port at the center for vertically-polarized omnidirectional radiation. The combination of slotted patch, monopolar patch, and vias yields several interesting features, which are exploited in the design to achieve dual-band tripolarized operation. Thank to the differential feed scheme and structural symmetry, the proposed antenna achieves a very high isolation among all ports. For verification, the final design is fabricated and measured. The double differential-fed scheme are realized by using two wideband out-of-phase power dividers, whose operational bandwidth covers both 2.45 and 3.5 GHz bands. The antenna with profile of $0.09\lambda _{2.45-{\mathrm { GHz}}}$ yields a measured 10-dB return loss bandwidth of $2.43-2$ .49 GHz and $3.23-3$ .66 GHz and isolation of $\ge $ 35 dB among all ports. Tripolarized radiation is verified with far-field measurement, showing highly symmetrical pattern and low cross-polarization in all three operational modes. The proposed design is a good candidate for dual-band communication systems which require polarization and pattern diversity antennas.

This paper proposes dual-band dual linearly-polarized (LP)/circularly-polarized (CP) antennas using combined metasurface and patch for ultra high frequency (UHF) and 2.45-GHz RF identification (RFID) readers. A metasurface and two stacked patches are employed for the lower and upper bands, respectively. The metasurface and patches are incorporated with a common feeding structure which is composed of two orthogonal microstrip lines and a crossed slot to produce the dual-wideband dual-LP radiation. To reduce the back radiation and improve the gain, the antenna is backed by a metallic cavity. Taking advantage of its low-profile structure, the dual-LP antenna is further integrated with a dual-band branch-line coupler to realize dual-CP radiation. Both proposed antennas yield stable radiation performances across dual operating bands which completely cover the assigned RFID spectrum at the UHF ( $860-960$ MHz) and 2.45-GHz ( $2400-2500$ MHz) bands.

Private 5G networks for industrial users are emerging as one of the leading advanced 5G use cases. This timely work presents a comprehensive experimental analysis of a private 5G network conducted in sparse and dense industrial environments at sub-6 GHz. Measured results of the over-the-air error vector magnitude (EVM) are provided, considering signal-to-noise ratio (SNR) for different 5G new radio modulation and coding schemes (MCSs), bandwidths (BWs) and numerologies (subcarrier spacings) using omnidirectional or directional antenna configurations at the transmitter (TX) and the receiver (RX). Channel sounding measurements are also conducted to characterise the channels in terms of root mean square (RMS) delay spread. The measurement results show that channels in the dense industrial environment have greater RMS delay spreads than in the sparse industrial environment due to strong reflected or scattered multipath components with significant delays. This results in higher EVMs and bit error rates (BERs), i.e., as the RMS delay spread increases, a higher SNR is required to meet the EVM limits. It is also observed that using directional antennas at the TX and RX in both environments reduces the RMS delay spread and hence the inter-symbol interference and the EVM. This allows higher MCS modes (e.g., 64 QAM and 256 QAM) to be used for reliable data transmission, significantly improving the bandwidth efficiency and reducing the latency. When evaluating system performance for different BWs and numerologies, using a lower BW and numerology provides a better system performance (lower EVMs and BERs), especially in dense industrial environments.

Achieving reliable low-latency streaming on real-time immersive services that require seamless interaction has been of increasing importance recently. To cope with such an immersive service requirement, IETF and 3GPP defined Low Latency, Low Loss, and Scalable Throughput (L4S) architecture and terminologies to enable delay-critical applications to achieve low congestion and scalable bitrate control over 5G. With low-latency applications in mind, this paper presents a cloud-based streaming system using WebRTC for real-time communication with an adaptable L4S congestion control (aL4S-CC). aL4S-CC is designed to prevent the target service from surpassing a required end-to-end latency. It is evaluated against existing congestion controls GCC and ScreamV2 across two configurations: 1) standard L4S (sL4S) which has no knowledge of Explicit Congestion Notification (ECN) marking scheme information; 2) conscious L4S (cL4S) which recognizes the ECN marking scheme information. The results show that aL4S-CC achieves high link utilization with low latency while maintaining good performance in terms of fairness, and cL4S improves sL4S's performance by having an efficient trade-off between link utilization and latency. In the entire simulation, the gain of link utilization on cL4S is 1.4%, 4%, and 17.9% on average compared to sL4S, GCC, and ScreamV2, respectively, and the ratio of duration exceeding the target queuing delay achieves the lowest values of 1% and 0.9% for cL4S and sL4S, respectively.

This study addresses the challenges of effectively reflecting maritime traffic flows in Korea’s MSP. Maritime traffic intensity zones were identified through a spatio-temporal density analysis and the application of standardization techniques. This study proposes unified extraction standards for maritime traffic intensity zones and provides quantitative indicators for MSP based on the occupancy rates of Automatic Identification System (AIS) data. The results of the spatio-temporal density analysis were polygonized and represented at 10% intervals. A suitable maritime traffic intensity zone was then selected based on the occupancy rates of ship counts and ship points within each polygon, followed by a statistical review. Consequently, the maritime traffic intensity zone with an upper 20% density was identified, and the zone exhibiting an upper 10% density was designated as the primary maritime traffic intensity zone. This approach facilitates improved navigation and port area reflections, as anticipated from the implementation of unified extraction standards and methods for maritime traffic intensity zones established by local governments in Korea.

As new forms of industries emerge in marine spaces due to global environmental protection trends and the promotion of recreational activities, traditional navigation areas for vessels are also being affected. To ensure safe vessel passage and effective vessel traffic management in response to evolving maritime environments, navigational routes for ships must be exclusively established. Numerous studies have attempted to derive shipping routes from historical vessel-traffic data; however, the final forms of polygon shapes representing routes have not been produced. Hence, this study aimed to extract polygonal routes based on densely trafficked maritime areas. Data collected from the automatic identification system (AIS) onboard maritime vessels were utilized to analyze dense navigational areas, which were then divided into 1 km grids. Through a spatial-temporal analysis method utilized from The European Marine Observation and Data Network (EMODnet), the occupancy time of the vessels in each grid was calculated to extract dense traffic areas. The dense traffic areas extracted in grid form were processed using the geographical identification system to create polygonal routes by smoothing and simplification. The resulting polygons exhibited different shapes depending on the analysis period. To extract a unique representative route polygon, the CRiteria Importance Through Intercriteria Correlation (CRITIC) method was employed to calculate the shape similarity based on the centroid, shape index, and overlapping area ratio of the polygons. The extracted representative polygon demonstrated the highest shape similarity compared with the other polygons and was utilized by 95.93% of the vessels navigating the analyzed area. The study results can contribute to identifying essential areas for vessels in maritime zones by proposing representative shipping routes.

This paper presents a gripper structure that improves the known potential limitations of curved fingertip movements and the restricted range of objects that can be grasped during power grasp in two-finger adaptive grippers. In the proposed gripper, a Grasshopper mechanism is applied to improve the movement of the fingertips, making it closer to a straight line. Additionally, a grasping method is implemented that allows the fingertips to move towards the palm of the gripper after grasping an object, thereby expanding the size range of objects that can be adaptively grasped. This movement of the fingertips also results in an increase in gripping force. The performance theoretically confirmed through kinematics and force analysis is verified through the development of a prototype. This prototype is used to validate the basic functions of the adaptive gripper and to assess improvements in the fingertip trajectory, demonstrating movement closer to a straight line. Additionally, the improvements in the size range of objects that can be adaptively grasped are verified. The proposed power grasp range ratio shows that the gripper can grasp objects up to 76.7% of the stroke, demonstrating an improved range compared to other two-finger adaptive grippers. Furthermore, the grasping motion of the proposed gripper demonstrates an increase in gripping force in the experiments.

Currently, it is a norm to design a semiconductor fab using building information models (BIMs), which refer to a digital representation of a building’s physical and functional characteristics. The comprehensive data provided by BIMs include 3D geometric models. This paper presents a 3D model-based camera tracking method, which is targeted at navigating a fab’s wide indoor environment. The key observation made in designing the method is that there are a number of fixed objects in such an indoor environment. The columns are the representative among them. Our method extracts the columns from the input image and matches them to their BIMs to estimate the camera pose. The estimation accuracy is significantly increased by adopting an instance segmentation network. It is trained with a dataset, which is extracted from the target indoor environment and processed by our own data engine. The test results show that our tracking method is drift-free, accurate and robust. We envision that it can be used in many applications such as AR-based visual inspection.

The critical role of a remote chemical sensing using a Fourier Transform Infrared (FT-IR) spectrometer has been emphasized for detecting lethal chemicals in the atmosphere. To enhance standoff detection capabilities, acquiring adequate gas spectral data is crucial for training and optimizing detection algorithms across diverse outdoor scenarios. However, the collection of outdoor infrared spectra with a number of conditions is constrained owing to uncontrolled weather factors including a temperature and humidity, leading to impaired reliability of the data. In addressing outdoor data acquisition challenges, we introduced a data augmentation method using a conditional CycleGAN. This technique utilizes spectral data obtained exclusively under controlled laboratory conditions. The proposed deep generative model takes as input the background spectrum, which is concatenated with two critical attributes: the temperature difference between the target substance and the background, and pathlength concentration. Subsequently, the model computes a brightness temperature spectrum for a gas against a specific background, employing SF6 as the target chemical gas. The validity of the generated data was assessed using two detection algorithms: the Pearson Correlation Coefficient and Adaptive Subspace Detector. In addition, the accuracy performance of detectors trained with the augmented dataset was compared and evaluated against those trained with the pure dataset. The results demonstrated that the model can simulate gas spectra onto unseen background spectra and enhance the chemical sensing database, and it can contribute to data augmentation for improving the performance of chemical gas detection systems.

The rapid expansion of Advanced Air Mobility has highlighted its significant dependence on Global Navigation Satellite Systems, particularly in the maritime context. This study introduces a novel adaptation of the eLoran system, which, while traditionally utilized for 2D positioning via ground-based radio frequency signals, is now being enhanced to enable 3D position estimation. A key innovation is the development of a 3D positioning method employing a geomatrix and distance calculations specifically designed for non-line-of-sight signals, facilitating eLoran’s application in aviation and providing continuous position estimation at various elevations. In addition, this study addresses the challenge of cycle slips in aviation, which is common in dynamic flight conditions. It proposes an algorithm that considers the uncertainty of multiple error sources to effectively mitigate such problems. This approach significantly improves positional accuracy by precisely detecting and correcting cycle slips, ensuring consistent and reliable navigation in rapidly changing flight dynamics. The performance of this advanced eLoran positioning algorithm was evaluated against that of traditional commercial receivers through real-flight experiments. The findings illustrate substantial enhancements in the accuracy and reliability of the eLoran system, offering a dependable navigation alternative that lessens reliance on Global Navigation Satellite Systems and boosts safety and efficiency in maritime and coastal air travel.

In recent years, with the advancement of information technology and digital transformation, the number of IoT devices has been rapidly increasing. Despite efforts to raise awareness and encourage users to enhance security for these IoT devices, a large number of them remain connected to the Internet with poor security, using simple passwords or default credentials. These insecure IoT devices unintentionally become useful tools for malicious actors. By utilizing Worm-type Malware, a type of malware capable of rapid infection, attackers can infiltrate and control numerous devices in a very short period. Once compromised, these IoT devices are used in network attacks such as DDoS or malware propagation. Representing modern Worm-type Malware, two Malware have continuously attracted attention since their appearance: Mirai and Hajime. Specifically, Mirai is a highly infectious malware that has caused numerous DDoS attacks, resulting in significant financial and reputational damage to several corporations and businesses. Similarly, Hajime is also a powerful malware with a propagation principle similar to Mirai. Similar to past investigations into malware such as Code-red, SQL Slammer and Conficker, previous studies focused on these two malware were based on statistical analysis of the number of infected IP addresses within fixed time frames or geographical locations. However, relying on this simple statistical method, previous studies didn’t assess the persistence, intermittency, or variations in the infection process of each IoT device. This led to incomplete reflections of the infection process and accurate changes in the numbers of Mirai and Hajime. Intending to collect comprehensive information and make accurate assessments reflecting the infection status, variations, and characteristics of Mirai and Hajime during the infection process, this study applies a new investigation method and evaluation index using data collected from the Network Telescope in the Darknet space. The investigative results of this study have revealed the changes and assessed the infection scale of Mirai and Hajime since their appearance. Additionally, our research has identified the update time of Hajime, estimated the current number of infected devices, and quantitatively evaluated the impact of Hajime on the infection activities of Mirai.

The concept of Mobile Edge Computing (MEC) has been recently highlighted as a key enabling technology for the deployment of sixth-generation (6G) wireless network services. On the other hand, the possibility of combining Unmanned Aerial Vehicles (UAV) with Reconfigurable Intelligent Surfaces (RIS) has also been recognized as a powerful communication paradigm able to provide improved propagation characteristics of wireless communication channels, as well as increased capacity and extended coverage. Then, the possibility of merging the characteristics of such a communication paradigm with the one provided through MEC represents a valid solution to fulfill the main requirements of 6G networks. In this paper, we consider the combination of computation offloading and resource allocation in an MEC-based system where the MEC server is hosted by a massive MIMO base station, which serves multiple macro-cells assisted by a UAV-equipped RIS. In this context, we focus on minimising the latency for executing tasks of all user equipment (UE) within the considered scenario. To tackle this problem, we formulate an optimisation problem that jointly optimises computation offloading from user equipment (UE) towards the MEC server, and communication resources in the underlying UAV-assisted and RIS-aided network. The extensive simulation results demonstrate how the proposed method outperforms in terms of providing reduced latency for the considered system when compared with other conventional schemes.

We present a novel concept of a millimeter-wave patch antenna that integrates the functions of a dual-polarized antenna and a capacitive proximity sensor to detect the state of a hand grip on a smartphone and compensate for the effect. The antenna has a dual-polarized, proximity-coupled, square patch structure, and acts as a proximity sensor by utilizing the change in capacitance of the patch when the user’s hand approaches. By observing the change in capacitance at the center of a square patch operating in two fundamental TM modes, antenna and sensor functions can be performed concurrently and without mutual interference. The antenna is designed using a low-cost four-layer FR-4 substrate and fabricated using a cost-effective standard printed circuit board (PCB) process. The antenna is configured as a $1 \times 8$ array, and after mounting it on a smartphone model, the gain, beam steering performance, and change in capacitance according to hand proximity are predicted and measured. The peak gain measured at 28 GHz exceeds 11.8 dBi, with a gain variation of less than 2.2 dB within the 5 G 257 band ( $26.5-29.5 \mathrm {GHz}$ ). The measured capacitance change due to hand proximity is 130 fF, which is sufficient for real-time detection of the hand grip state. An example of improvement in the gain of the proposed antenna by detecting the hand grip state is presented through a simulation.

Today, unmanned aerial vehicles (UAVs) are extensively employed in wireless communication systems due to their mobility and exceptional maneuverability. With the advantage of flying high to increase the line of sight (LOS) probability with ground user equipment (UE), many studies have demonstrated that using UAVs to replace ground base stations (BS) can reduce transmission power loss. However, most research results are still limited to simulations with implicit formulas. This paper provides a closed formula for calculating the communication power gain of a cell based on environmental parameters. To achieve this, we propose an approximately linear mathematical model that allows for determining the formula for the optimal elevation angle, thereby calculating the coverage area of the UAV-BS and communication power gain of the coverage area. Simulations show that the proposed formula can be applied over a wide range of environmental parameters, and the quantitative results enable rapid cell planning for communication designers.

We demonstrate that the shorter channel self-aligned top-gate (SA TG) coplanar indiumgallium- zinc oxide (IGZO) thin-film transistors (TFTs), with negative voltage applied to the back-gate, exhibit superior characteristics as driving transistors in organic light-emitting diode (OLED) pixels compared to their longer channel counterparts. The shorter channel IGZO TFTs (with a channel length (L) of 3 μm) biased with a back gate voltage of −3.5 V showed a larger subthreshold swing (SS = 0.21 V/dec) than the longer channel ones (with L = 5 μm, SS = 0.16 V/dec) with a similar threshold value (VTH = 0.7–0.8 V). A large SS is beneficial for controlling grayscale levels, especially at low gray levels, when IGZO TFTs are used as driving transistors in OLED pixels. Furthermore, the negatively back-gate-biased shorter channel SA TG coplanar IGZO TFTs exhibited significantly enhanced electrical stability compared to the longer channel ones under both positive gate bias and hot carrier stresses. The findings of this study are expected to be useful in expanding the utility of IGZO TFTs in OLED displays.

Monocular Depth Estimation (MDE) and Object Pose Estimation (OPE) are important tasks in visual scene understanding. Traditionally, these challenges have been addressed independently, with separate deep neural networks designed for each task. However, we contend that MDE, which provides information about object distances from the camera, and OPE, which focuses on determining precise object position and orientation, are inherently connected. Combining these tasks in a unified approach facilitates the integration of spatial context, offering a more comprehensive understanding of object distribution in three-dimensional space. Consequently, this work addresses both challenges simultaneously, treating them as a multi-task learning problem. Our proposed solution is encapsulated in a Unified Framework for Depth-Assisted Monocular Object Pose Estimation. Leveraging Red-Green-Blue (RGB) images as input, our framework estimates pose of multiple object instances alongside an instance-level depth map. During training, we utilize both depth and color images, but during inference, the model relies exclusively on color images. To enhance the depth-aware features crucial for robust object pose estimation, we introduce a depth estimation branch supervised by depth images. These features undergo further refinement through a cross-task attention module, contributing to the innovation of our method in significantly improving feature discriminability and robustness in object pose estimation. Through extensive experiments, our approach demonstrates competitive performance compared to state-of-the-art methods in object pose estimation. Moreover, our method operates in real-time, underscoring its efficiency and practical applicability in various scenarios. This unified framework not only advances the state of the art in monocular depth estimation and object pose estimation but also underscores the potential of multi-task learning for enhancing the understanding of complex visual scenes.

Semantic segmentation of aerial and satellite images plays a crucial role in a wide range of applications and services, catering to the increasing needs of environmental resource management, urban planning, and traffic safety. Many efficient semantic segmentation methods have been proposed by exploiting deep learning techniques with convolution neural networks (CNNs) architectures and self-attention mechanisms to achieve superior accuracy if compared with conventional machine learning-based approaches. In this article, we introduce a hybrid bilateral segmentation network (HBSeNet), a novel semantic segmentation architecture. Inspired by the success of dual-path network models that replace conventional single-branch encoder–decoder architectures, we construct a model with the core idea of combining the context path and the spatial path to optimize both the accuracy and the complexity of deep learning model in the field of remote sensing image segmentation. Moreover, HBSeNet innovates with auxiliary modules designed to enhance its performance, such as sequential atrous convolution, information synthesis module, and bridge for efficient multiscale feature extraction, fusion, and integration. In simulations, our model achieves a global accuracy of 92.04%, a mean intersection-over-union of 83.57%, and a mean boundary-F1-score of 90.23% when evaluated on the ISPRS Potsdam dataset, surpassing the state-of-the-art segmentation models, such as DeepLabV3+, SwinCNN, and ST-UNet.

Hand pose estimation is vital for various applications, including virtual reality (VR), augmented reality (AR), gesture recognition, human-computer interaction (HCI), and robotics. Achieving accurate and real-time hand pose estimation is challenging due to factors such as the high degree of articulation in the human hand and the variability in hand shapes and sizes. While multimodal data offers advantages, developing a fast and resource-efficient hand pose estimation system remains challenging. Current state-of-the-art methods often require powerful graphics processing units (GPUs) for high performance, limiting deployment on edge platforms with limited computational resources. There is a critical need for higher efficiency without compromising accuracy, especially in real-world applications like mobile devices and embedded systems. Additionally, real-time performance is essential for practical applications, where systems must respond immediately to user interactions. Unfortunately, most current methods struggle to achieve real-time speeds, even on powerful GPUs, let alone on resource-constrained devices. To address these challenges, we propose an efficient hand pose estimation system that leverages both red-green-blue (RGB) and depth (RGBD) data through a unified fusion strategy. Our method combines appearance and geometric data early in the processing pipeline, significantly reducing computational complexity while maintaining real-time performance on resource-constrained devices. Experimental results show that the proposed model runs at over 110 fps on GPU, and 30 fps on the edge platform of NVidia Jetson NX Xavier, which is 4 to 5 times faster than existing methods, while achieving competitive accuracy.

Label sparsity in multivariate time series (MTS) makes using label information for practical applications challenging. Thus, unsupervised representation learning methods have gained attention to learn effective representations suitable for various MTS tasks without relying on labels. Recently, contrastive learning has emerged as a promising approach to generate robust representations by capturing underlying MTS information. However, the existing methods have some limitations, such as insufficient consideration of cross-variable relationships of MTS and high sensitivity to positive pairs. Therefore, we proposed a novel spatio-temporal contrastive representation learning method (STCR) designed to address these limitations. STCR focuses on learning robust representations by encouraging spatio-temporal consistency, which comprehensively considers spatial information as well as temporal dependencies in MTS. The results of extensive experiments on MTS classification and forecasting tasks demonstrate the efficacy of STCR in generating high-quality representations, achieving state-of-the-art performance on both tasks.

Rain streaks impede image feature extraction, hindering the performance of computer vision algorithms such as pedestrian and lane detection in adverse weather conditions. Image deraining is crucial for enhancing reliability of such algorithms. However, detail and texture information of objects in background areas are often lost during the deraining process due to their structural similarity with rain streaks. To remove rain streaks effectively while preserving image details, we propose a novel layer decomposition learning network (LDLNet) to separate rain streaks and object details in rainy images. LDLNet consists of two parts: the discriminative group feature split (DGFS) and the group feature merging (GFM). DGFS utilizes sparse residual attention modules (SRAM) to capture spatial contextual features of rainy images, enhancing the network’s ability to understand the complex relationships between rain streaks and object details. In addition, DGFS employs the bottom-up intergroup feature fusion (BIFF) approach to aggregate multi-scale context information from continuous SRAMs, facilitating the decomposition of rainy images into discriminative feature groups. Subsequently, GFM integrates these feature groups by concatenating them, preserving the interdependent characteristics of clean backgrounds and rain layers. Experimental results reveal that the proposed approach achieves superior rain removal and detail preservation in both synthetic datasets and real-world rainy images compared to the state-of-the-art rain removal models.

Recent advancements in Artificial Intelligence (AI) have greatly impacted cybersecurity, particularly in detecting phishing websites. Traditional methods struggle to address evolving vulnerabilities, but research shows that Machine Learning (ML), Ensemble Learning (EL), and Deep Learning (DL) are effective in developing defenses. However, these methods face challenges with adversarial examples (AEs). The multimodal model (MM) is a promising solution, yet there is a significant lack of research using multimodal techniques specifically for phishing website detection (PWD) against adversarial websites. To tackle this challenge, this paper assesses 15 learning-based models, particularly multimodal ones, for phishing and adversarial detection, aiming to enhance their defense capabilities. Due to the scarcity of adversarial websites, training and testing models are limited. Therefore, this study proposes an innovative attack framework, AWG - Adversarial Website Generation that employs Generative Adversarial Networks (GAN) and transfer-based black box attacks to create AEs. This framework closely mirrors real-world attack scenarios, ensuring high effectiveness and realism. Finally, we present defense strategies with straightforward implementation and high effectiveness to enhance the resistance of models. The models underwent training and testing on a dataset collected from reputable sources such as OpenPhish, PhishTank, Phishing Database, and Alexa. This approach was chosen to ensure the dataset’s diversity and relevance to reflect real-world conditions. Experimental results highlight that the Generator’s effectiveness is demonstrated by a domain structure generation rate exceeding 90%. Moreover, AEs generated by this Generator effectively bypass most state-of-the-art ML, DL, and EL models with an evasion rate of up to 88%. Notably, the Support Vector Machine (SVM) model is the most vulnerable, with a detection rate of only 10.02%. On the other hand, the MM Shark-Eyes demonstrates outstanding resistance against AEs, with a detection rate of up to 99%. Upon applying our defense strategy, the resistance of models is significantly boosted, with all detection rates surpassing 90%. These findings underscore the robustness of our methods and pave the way for further exploration into advanced attack and defense strategies in the context of phishing and adversarial website detection.

In this study, we designed FogFusionNet, a multimodal sea fog prediction model, that used closed-circuit television (CCTV) images and multivariate time series observation (MTSO) data to predict three visibility classes—Normal visibility, Low visibility, and Sea fog—at 1-h intervals from the current time to 6-h in the future for a specific region. We applied weighted sampling and weighted loss to overcome the imbalance of each visibility class, and additionally evaluated the effect of replacing missing MTSO data. A total of 4 years of data regarding Incheon Port, which faces the Yellow Sea and is prone to sea fog, were collected for training and verifying FogFusionNet. Of these, 3 years of data was used for training FogFusionNet, and the remaining 1 year of data were used for verifying the performance of FogFusionNet. The prediction performance of FogFusionNet at 1-h intervals was 86.2% (0-h), 79.1% (1-h), 73.4% (2-h), 70.7% (3-h), 64.7% (4-h), 59.6% (5-h), and 49.3% (6-h), showing an average prediction performance of 69.0%. FogFusioneNet is expected to promote coastal safety and reduce economic losses due to coastal sea fog.