Search

A novel D-band self-packaged silicon-based air cavity-backed transition from grounded coplanar waveguide to air-filled rectangular waveguide was investigated, fabricated, and measured in this work. The equivalent circuit model was established and analyzed in detail, and design procedures are given. The calculated, simulated, and measured S-parameters of the transition show some agreement. The minimum measured insertion loss of the proposed transition is 1.1 dB at 147 GHz with a fractional 3-dB bandwidth of 10.2%. This transition demonstrates outstanding performance of low loss and profile compared with state-of-the art works in our in-house silicon-based MEMS photosensitive composite film fabrication process. It can be further used in a high-performance joint radar communication system in packaging.

Automatic driving technology has high accuracy and real-time requirements for pedestrian identification and localization. Pedestrian detection is a basic and necessary function in vision-based pedestrian detection systems and collision warning, which can effectively avoid traffic accidents and improve road driving safety to a certain extent. In this paper, a lightweight solution based on the FCOS-DEFPN model is proposed for real-time pedestrian detection. Based on the FCOS model, this paper proposes the FCOS-DEFPN model, which achieves the lightweight of the network by replacing the ResNet50 backbone network with the MobilenetV3 network and using the depth separable convolution instead of the ordinary convolution for parameter compression. While maintaining the detection accuracy, this paper introduces data enhancement methods such as Random Erasing and Morsia to simulate pedestrian occlusion and small target scenarios to improve the robustness of the model. For the pedestrian occlusion scenario, this paper introduces a lightweight attention network ECA, which helps to extract pedestrian features better. For small-target multi-scale pedestrians, the DEFPN feature pyramid network is proposed, which acquires feature information at multiple scales by attentional fusion of feature layers at different scales from top-down, bottom-up, and front-back. The experimental results show that the proposed model is enhanced in terms of detection accuracy for occluded and small-target pedestrians, and satisfies real-time pedestrian detection under the premise of robustness in complex scenes.

Pedestrian trajectory prediction is a key technology in surveillance systems and autonomous driving. However, due to the high uncertainty and dynamic spatial-temporal dependence of pedestrian movement, timely and accurate pedestrian trajectory prediction, especially long-term prediction, is still an open challenge. However, the existing models lack an effective modeling method for temporal dependence and spatial interaction modeling. To solve these problems, in this paper, we propose a novel paradigm of Sparse Transformer Networks with Spatial-Temporal Graph (STGSTN), which captures complex spatial-temporal interactions by stacking spatial-temporal Transformer blocks and improves the accuracy of pedestrian trajectory prediction by combining dynamic spatial dependence and long-range temporal dependence. We propose a new variant of sparse spatial transformer combined with graph neural networks, which uses the self-attention mechanism to dynamically model spatial dependencies to capture the state of pedestrian movement. The multi-head attention mechanism is also used to jointly model various patterns of spatial dependence. In addition, the sparse temporal transformer is used to learn the sparse attention map for time modeling, and then the long-range temporal dependence between pedestrians is modeled. Compared with the existing work, STGSTN can effectively and efficiently train long-range spatial-temporal dependencies. Experimental results show that STGSTN is competitive with state-of-the-art techniques on the ETH-UCY dataset, especially for long-term prediction. The performance of STGSTN may be limited by the quality of the training data, and its generalization to highly dynamic or crowded environments remains a challenge. Future work will focus on addressing these limitations and exploring additional contextual information to further enhance prediction accuracy.

The orthogonal time-frequency space (OTFS) modulation is resistant to double selective fading, making it a promising modulation scheme for next-generation integrated sensing and communication (ISAC) systems. Nonetheless, there exists a problem of mutual interference between uplink communication signals and sensing echo signals, which leads to the degradation of both information transmission and sensing performance. For this reason, this paper investigates a successive interference cancellation (SIC) method based on the delay-Doppler (DD) domain to suppress the interference between the sensing signals and communication signals, thereby ensuring the accuracy of communication signal detection and sensing parameter estimation. After receiving the mixed signals at the receiver, these signals are transformed from the time-frequency (TF) domain to the DD domain. Firstly, the sensing echo signal is treated as interference to the communication signal, which is detected using a threshold-based channel estimation method and the Message Passing (MP) algorithm. The communication signal is then reconstructed and subtracted from the mixed signal. At last, the remaining signal is then subjected to parameter estimation with the Maximum Likelihood (ML) algorithm. Simulation results demonstrate that this method allows the ISAC system to operate effectively in both communication and sensing tasks. Compared to the scenario without considering signal interference, the sensing parameter estimation is improved by approximately 6 dB. Additionally, compared to traditional interference suppression methods belonging to TF domain, the performance of sensing parameter estimation is improved by approximately 2 dB.

Multiple infrastructure RGB-D cameras can be used for localizing autonomous vehicles in Automated Valet Parking. The accurate calibration of these cameras’ extrinsic parameters is crucial. However, due to the sparse and distributed placement of the cameras, the field of view (FOV) between them is very small. This makes the calibration process complex and dependent on human expertise. To address this, this paper proposes an automatic extrinsic calibration method for multiple infrastructure cameras with a small FOV. The method introduces an auxiliary camera to enhance the association between the multiple infrastructure cameras. A moving checkerboard placed within the public FOV is utilized as a reference for calibration. The optimization method involves constructing a pose graph to store the poses of the cameras and checkerboard, and it solves the pose graph by calculating the reprojection errors of the checkerboard. The experimental results demonstrate that the proposed method achieves a calibration accuracy of two centimeters. It outperforms other calibration methods when applied to a constructed multiple RGB-D camera system. Furthermore, the proposed method is simple and efficient in the real calibration procedure.

This article presents novel methods for synthesizing distributionally robust stabilizing neural controllers and certificates for control systems under model uncertainty. A key challenge in designing controllers with stability guarantees for uncertain systems is the accurate determination of and adaptation to shifts in model parametric uncertainty during online deployment. We tackle this with a novel distributionally robust formulation of the Lyapunov derivative chance constraint ensuring a monotonic decrease of the Lyapunov certificate. To avoid the computational complexity involved in dealing with the space of probability measures, we identify a sufficient condition in the form of deterministic convex constraints that ensures the Lyapunov derivative constraint is satisfied. We integrate this condition into a loss function for training a neural network-based controller and show that, for the resulting closed-loop system, the global asymptotic stability of its equilibrium can be certified with high confidence, even with Out-of-Distribution (OoD) model uncertainties. To demonstrate the efficacy and efficiency of the proposed methodology, we compare it with an uncertainty-agnostic baseline approach and several reinforcement learning approaches in two control problems in simulation. Open-source implementations of the examples are available at https://github.com/KehanLong/DR_Stabilizing_Policy.

The advent of the sixth generation of mobile communications (6G) ushers in an era of heightened demand for advanced network intelligence to tackle the challenges of an expanding network landscape and increasing service demands. Deep Learning (DL), as a crucial technique for instilling intelligence into 6G, has demonstrated powerful and promising development. This paper provides a comprehensive overview of the pivotal role of DL in 6G, exploring the myriad opportunities and challenges that arise. Firstly, we present a detailed vision for DL in 6G, emphasizing areas such as adaptive resource allocation, intelligent network management, robust signal processing, ubiquitous edge intelligence, and endogenous security. Secondly, this paper reviews how DL models leverage their unique learning capabilities to solve complex service demands in 6G. The models discussed include Convolutional Neural Networks (CNN), Generative Adversarial Networks (GAN), Graph Neural Networks (GNN), Deep Reinforcement Learning (DRL), Transformer, Federated Learning (FL), and Meta Learning. Additionally, we examine the specific challenges each DL model faces within the 6G context. Moreover, we delve into the rapidly evolving field of Artificial Intelligence Generated Content (AIGC), examining its development and impact within the 6G framework. Finally, this paper culminates in a detailed discussion of ten critical open problems in integrating DL with 6G, setting the stage for future research and development in this field.

Decoding continuous human motion from surface electromyography (sEMG) in advance is crucial for improving the intelligence of exoskeleton robots. However, incomplete sEMG signals are prevalent on account of unstable data transmission, sensor malfunction, and electrode sheet detachment. These non-ideal factors severely compromise the accuracy of continuous motion recognition and the reliability of clinical applications. To tackle this challenge, this paper develops a multi-task parallel learning framework for continuous motion estimation with incomplete sEMG signals. Concretely, a residual network is incorporated into a recurrent neural network to integrate the information flow of hidden states and reconstruct random and consecutive missing sEMG signals. The attention mechanism is applied for redistributing the distribution of weights. A jointly optimized loss function is devised to enable training the model for simultaneously dealing with signal anomalies/absences and multi-joint continuous motion estimation. The proposed model is implemented for estimating hip, knee, and ankle joint angles of physically competent individuals and patients during diverse exercises. Experimental results indicate that the estimation root-mean-square errors with 60% missing sEMG signals steadily converges to below 5 degrees. Even with multi-channel electrode sheet shedding, our model still demonstrates cutting-edge estimation performance, errors only marginally increase 1 degree.

As a promising alternative to traditional underwater communication methods, underwater wireless optical communication (UWOC) offers higher data rates, lower latency, and enhanced security. However, the practical implementation of high-speed UWOC systems typically faces challenges arising from bandwidth-limited optical devices and adverse effects of absorption, scattering, and oceanic turbulence. To tackle these challenges, we propose a novel joint probabilistic shaping and precoding scheme that simultaneously optimizes the mutual information (MI) and reduces the peak-to-average power ratio (PAPR) of multiple-input single-output (MISO) orthogonal frequency-division multiplexing (OFDM) UWOC systems. Specifically, the proposed approach probabilistically shapes the transmitted symbols and appropriately allocates weights to both multi-light sources and multi-carriers based on precoding to match the channel characteristics. Numerical results demonstrate its superiority over existing schemes, particularly showcasing its potential to significantly improve the performance of UWOC systems via MI optimization and PAPR reduction. Moreover, the results verify that the proposed approach exhibits satisfactory robustness against oceanic turbulence.

Cardiac motion tracking from echocardiography can be used to estimate and quantify myocardial motion within a cardiac cycle. It is a cost-efficient and effective approach for assessing myocardial function. However, ultrasound imaging has the inherent characteristics of spatially low resolution and temporally random noise, which leads to difficulties in obtaining reliable annotation. Thus it is difficult to perform supervised learning for motion tracking. In addition, there is no end-to-end unsupervised method currently in the literature. This paper presents a motion tracking method where unsupervised optical flow networks are designed with spatial reconstruction loss and temporal-consistency loss. Our proposed loss functions make use of the pair-wise and temporal correlation to estimate cardiac motion from noisy background. Experiments using a synthetic 4D echocardiography dataset has shown the effectiveness of our approach, and its superiority over existing methods on both accuracy and running speed.

In this paper, a variational Bayesian adaptive Kalman filter (VBAKF) was used to solve the impact of unknown non-Gaussian measurement noise (NGMN) and sensor measurement loss in Wireless Sensor Networks (WSN) communication. First, the inverse Wishart (IW) distribution was used as the conjugate prior distribution of multiple nominal noise covariance matrices, and a Gaussian mixture model (GMM) was introduced to construct a measurement likelihood probability density function (PDF) to model the effect of the NGMN. The sensor measurement loss problem was modeled using the Bernoulli distribution as a statistical property of the packet loss parameter. Second, the proposed algorithm achieves iterative refinement of latent variable estimates through the application of variational Bayesian (VB) methods, thereby adjusting the GMM weights and the probability of sensor measurement loss accordingly. Third, we present the floating-point operations of the algorithm and compare them with those of other mixture model algorithms. Finally, the effectiveness of the algorithm will be verified by experimental simulation.

Traditionally, radio frequency detection or ambient spectrum sensing has required high-performance spectrum analyzers and RF signal analyzers, leading to relatively high costs due to the need for a local oscillator and signal mixer. To overcome this challenge, we propose a low-cost, substantially simplified solution utilizing a stretchable rectenna, a microcontroller unit (MCU), and feedback control systems. By exploiting the dynamic correlation between the resonant frequency and the tensile ratio of the stretchable antenna, the incoming frequency can be determined by recording the maximum rectifier DC power output as a function of the electromechanically controlled tension ratio of the stretchable antenna. Our measured results indicate that a frequency measurement range of 1.8 GHz to 2.5 GHz can be achieved through careful design of the stretchable antenna and broadband rectifier. We have experimentally demonstrated an over-the-air far-field frequency sensing system based on this concept, showcasing significant advantages in power consumption, cost-effectiveness, and simplicity when compared to state-of-the-art RF spectrum analyzers.

The enormous throttling losses are the crucial reason for the low energy efficiency of non-road mobile machinery. To achieve energy saving, a parallel electro-hydraulic hybrid drivetrain that combines an electric-hydraulic energy recovery system with a valve-controlled system is proposed. With a parallel electric-hydraulic energy recovery system, both the energy lost caused by load imbalances and overrunning loads can be recovered and reutilized. To achieve satisfactory operation performance and minimize energy consumption, a rule-based energy management strategy is implemented for real-time control. Several experiments are conducted to verify the feasibility and energy-saving features of the proposed scheme. Additionally, taking a 37-ton hydraulic excavator as an example, a co-simulation platform is established to demonstrate its working characteristics under various conditions. Compared with the hydraulic excavator without energy recovery, throttling losses and energy consumption of the proposed system are reduced by 60%-75% and 24%-27%. Moreover, the proposed system integrates the energy-saving pressure compensator, and the response lag between the heavy-load actuator and light-load actuator is shortened by 47% during multi-actuator motion, effectively improving coordination and maneuverability.

Aiming at the problem that cracks with different shapes and scales on the exterior walls of high buildings are difficult to detect, this paper proposed a wall crack detection method for high-altitude cleaning robots by hybridizing the GAM attention mechanism and RepGFPN. First, the GAM attention mechanism was incorporated into the YOLOV5 backbone network to reduce information and amplify global features to improve the accuracy of feature extraction. Then, the neck network incorporated the RepFPN method to improve the descriptive ability of fused multi-scale features and to increase computational efficiency. Public datasets Concrete Crack Images for Classification, Mixed VOC2007, CrackForest-dataset-master, and UCMerced_LandUse were used for experimental validation. The ablation experiment results show that the average accuracy of mAP is improved by 13.5% after introducing the GAM attention mechanism under the yolov5 s original model, while the method in this paper (GR-YOLO) continues to improve by 4.7%. The experimental results show that the average accuracy mAP of the proposed method (GR-YOLO) is 24.0%, 47.1% and 41.0% higher than that of the model yolov5s + involution, yolov5s + p2 + involution and yolov5s + p2 + involution + CBAM, respectively. The method proposed in this article can more effectively improve the accuracy of crack detection and has important application prospects.

In order to overcome spectrum scarcity and provide higher data rates, the sixth-generation (6G) wireless communication network is expected to perform data transmission using terahertz (THz) frequencies. However, the effective implementation of these communication systems is hampered by severe levels of signal degradation to which the THz bandwidth is subject to. Recent improvements and advancements in the fabrication process of electromagnetic (EM) metamaterials have made reconfigurable intelligent surfaces (RIS) a very promising solution to address these THz-related attenuation issues. Additionally, the adoption of non-orthogonal multiple access (NOMA) transmissions represents an effective way to improve spectrum efficiency for 6G networks. In this paper, we investigate the problem of downlink aggregated sum-rate maximisation for a multiple-input multiple-output (MIMO) system assisted by a RIS panel in performing NOMA transmission within the THz bandwidth. More specifically, we propose an optimization algorithm that jointly optimizes the transmitting power at the access point (AP) and the phase-shift coefficients for the RIS elements iteratively. Through simulation results, we demonstrate that the proposed method outperforms conventional benchmark schemes in terms of achieved aggregated throughput.

Purpose: We propose an FDK-based neural network to directly learn the filter from an iterative reconstruction (IR) algorithm and apply the learned filter in the FDK algorithm to obtain a high-quality CBCT reconstruction. Methods: The FDK algorithm is transformed into a linear expression of several matrix multipliers and embedded into neural network layers. Then, the FDK-based neural network framework is built including two fundamental modules and four core network layers. This network model can learn a filter directly from the iteratively reconstructed CBCT images by cascading the network layers of cosine weighting, filtering, backprojection, and leaky rectified linear unit and setting filter as the only trainable parameter. Preliminary and simulation studies performed on abdominal CT datasets are conducted to explore the correctness and effectiveness of the learned filter. Then, the head and neck CT data and Catphan phantom are utilized to demonstrate the generalization performance of the learned filter. Results: Preliminary study shows that the learned filter is consistent with the target filter, and the mean absolute difference is around 0.001. Compared with conventional FDK, the FDK-based neural network shows a better image quality with the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) increasing by 67% and 6%, respectively. In terms of the line-pair slice in the Catphan phantom, the SSIM and PSNR are improved by 13.75% and 42.78%, respectively. Conclusions: The FDK-based neural network can reconstruct high-quality images by directly learning a filter from the label images and provides a new perspective on solving the time-consuming problem of IR methods.

Magnetorheological (MR) dampers possess the ability to alter the damping force of the damper by modifying the viscosity of Magnetorheological Fluid (MRF). The prevalent MR dampers mostly feature a single damping channel structure, which necessitates adjusting the magnetic field strength to attain asymmetric damping force during the compression and recovery stroke. However, it amplifies the influence of the control circuit’s time-delay on damping force. This article proposes the design of a dual-damping-channel MR damper to structurally achieve the asymmetric damping force. The dimensions of key components were determined using damping calculation formulas and electromagnetic simulation. A prototype MR damper was manufactured, and experimental verification was conducted. The results indicate that the damper operates optimally, and the damping force value and adjustable range of damping force meet the design objectives. The indicator characteristic curve of the shock absorber shows that the MR shock absorber designed in this paper can effectively reduce the influence of MRF time-delay on the damping force value of the shock absorber.

In the field of remote sensing, it is not feasible to obtain high spatial resolution multispectral (HRMS) images from a single satellite sensor. The existing methods use pansharpening techniques to obtain HRMS images by fusing panchromatic (PAN) and multispectral (MS) images. However, due to the scale difference between PAN and MS images, most pansharpening methods often use explicit sampling methods to integrate features at different scales. These explicit-based sampling techniques represent pixels as discrete points through predefined functions, rendering it challenging to fit the distribution among diverse modal data, this results in the loss of image texture details during the fusion process. Implicit neural networks can enhance the generative capability of images by incorporating pixel coordinate information, which is crucial for the fusion of remote sensing images with different spatial resolutions. Inspired by implicit neural representation, we propose a progressive implicit neural feature fusion network (PINFNet) for remote sensing images. A progressive implicit neural feature fusion is proposed; it establishes a coordinate modal relationship between the spatial and spectral information through the guidance of the high spatial features in PAN images. This enables the proposed PINFNet to progressively learn and integrate spatial and spectral information at different scales. Our method, as opposed to discrete sampling techniques, is capable of establishing a continuous representation between diverse modal data, which in turn preserves more texture detail information. Extensive experiments have shown that this approach outperforms state-of-the-art methods while maintaining high efficiency.

Electroencephalogram (EEG) signals play an important role in brain-computer interface (BCI) applications. Recent studies have utilized transfer learning to assist the learning task in the new subject, i.e., target domain, by leveraging beneficial information from previous subjects, i.e., source domains. Nevertheless, EEG signals involve sensitive personal mental and health information. Thus, privacy concern becomes a critical issue. In addition, existing methods mostly assume that a portion of the new subject’s data is available and perform alignment or adaptation between the source and target domains. However, in some practical scenarios, new subjects prefer prompt BCI utilization over the time-consuming process of collecting data for calibration and adaptation, which makes the above assumption difficult to hold. To address the above challenges, we propose Online Source-Free Transfer Learning (OSFTL) for privacy-preserving EEG classification. Specifically, the learning procedure contains offline and online stages. At the offline stage, multiple model parameters are obtained based on the EEG samples from multiple source subjects. OSFTL only needs access to these source model parameters to preserve the privacy of the source subjects. At the online stage, a target classifier is trained based on the online sequence of EEG instances. Subsequently, OSFTL learns a weighted combination of the source and target classifiers to obtain the final prediction for each target instance. Moreover, to ensure good transferability, OSFTL dynamically updates the transferred weight of each source domain based on the similarity between each source classifier and the target classifier. Comprehensive experiments on both simulated and real-world applications demonstrate the effectiveness of the proposed method, indicating the potential of OSFTL to facilitate the deployment of BCI applications outside of controlled laboratory settings.

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.

Multivariate Time series forecasting finds numerous applications across various fields, including society, industry, market, etc. Recently, gated recurrent unit neural networks (GRU) have shown high efficiency in processing sequential time series data in recent years. While traditional GRUs can learn and understand time series data, with the explosion and increasing complexity of data, there has not been much research on GRU networks that considers the fuzziness and periodicity of the data’s nature. Thus, the novel developed complex fuzzy-gated recurrent neural network (CFGRU) is proposed in this study to improve the ability of GRU networks to resolve multivariate time series forecasting issues. Complex fuzzy theory, which represents the uncertainty and periodicity of the data space from the input data, is integrated with GRU regression neural networks and the proposed CFGRU network. Furthermore, this paper also suggests a stacked residual complex fuzzy-gated recurrent neural network architecture for multivariate time series data forecasting. An experiment was carried out on multivariate time series data sets comprising 05 multivariate time series datasets (weather, sunspots, PM2.5, air quality, and power consumption) to validate the success and efficiency of the suggested model. Comparison results on three indices—MAE, RMSE, and SMAPE—indicate that the proposed model performs forecasting better than both complex fuzzy forecasting models and conventional GRU models.

Focusing on the challenges of vehicle detection in foggy weather, especially the algorithm of low accuracy caused by small and incomplete targets in adverse weather conditions, a foggy weather vehicle detection algorithm based on improved lightweight YOLOv8 was proposed. Firstly, the dataset was processed through a combination of data transformation, Dehaze Formers and dark channel preprocessing. Secondly, in the main body of YOLOv8, the C2f component was replaced with the dynamic convolution C2f- DCN, enhancing its adaptability to geometric changes in the image. To further improve the detection performance of the classifier, an improved S5attention module based on S2-MLP was introduced. This module utilizes contextual information to capture long-range dependencies and assign weights to different channels based on their relevance to the task at hand. By considering non-local features, the S5attention module helps the model better capture important spatial relationships within the image. Additionally, the feature extraction module was updated to FasterNext, improving the differential convolution’s feature extraction capabilities. The Involution module was also introduced to reduce FLOPs during feature channel fusion and reduce the model’s parameter count. Experimental results show that on the RESIDE foggy weather dataset, the improved algorithm has an mAP50 increase of 4.1% compared with the original algorithm, and the model’s parameter quantity is only 9.06m, with a computational cost reduced from 28.7G to 28.1G. The research model in this article will provide technical support for detecting vehicle targets in foggy weather, ensuring fast and accurate operation.

This article presents an efficient and effective design approach of the phase distribution calculation across metasurface for significant radar cross section (RCS) reduction of a circular polarization (CP) and liner polarization (LP) radar waves. The RCS reduction using the proposed design approach is achieved by imposing a novel cusp phase mask (which is usually used to generate 3D self-accelerating and self-healing cusp beams) at each geometric phased anisotropic unit cell composing the proposed cusp phased metasurface. By solving the cusp phase formula using MATLAB, it is found that the cusp phase mask required to achieve more than 10 dB of RCS reduction over a wide frequency band can be calculated without the need of significant lengthy optimizations or huge computer resources. The ability of such cusp phase mask metasurfaces to achieve significant backward scattering and RCS reduction has been rigorously investigated by means of simulations and measurements. When illuminated by a far-field radar CP or LP plane wave, the proposed cusp phased metasurface realizes more than 10 dB of RCS reduction from 10.9 to 26 GHz, corresponding to a fractional bandwidth of FBW = 81.8%. The 10 dB RCS reduction bandwidth of the cusp metasurface is maintained under both normal and wide angular incidence up to 75o. The proposed cusp metasurfaces have potential applications to make objects stealthy where the incidence radar signal has an unknown frequency, polarization, or angle of incidence.

Bohai Sea is China's inland sea, its complex marine and atmospheric optical properties pose a challenge to the application of satellite data to retrieve chlorophyll-a (CHLA) concentration with high accuracy. The high accuracy retrieval results of CHLA require simultaneous consideration of the adaptability of atmospheric correction algorithms and CHLA retrieval models. In this study, four atmospheric correction methods [the standard atmospheric correction algorithms of GDPS1.4.1 and GDPS2.0, the standard near-infrared atmospheric correction algorithm of NASA (Seadas_Default), and management unit of the North Sea mathematical models (Seadas_MUMM)] and four CHLA retrieval models (OC2, YOC, OC3G, and OC2M-HI) were selected in the process of applying geostationary ocean color imager (GOCI) data to retrieve CHLA in Bohai Sea. Based on the in situ data, the adaptability of their pairwise combinations in retrieval of CHLA in Bohai Sea was evaluated. The results indicate that the OC2 and OC3G models significantly overestimated the CHLA. The combination of the Seadas_Default atmospheric correction algorithm with the YOC CHLA retrieval model, or the combination of the Seadas_MUMM atmospheric correction algorithm with the YOC CHLA retrieval model, is more suitable for the retrieval of CHLA using GOCI data in Bohai Sea. In addition, this study shows that the CHLA obtained based on the data from eight-scene GOCI data were different to the data obtained based on single-scene GOCI data (approximating traditional polar-orbiting satellite sensor data) in daily, monthly, and yearly average results. The monthly mean difference between the two is the most significant, ranging from -0.66 to 1.49 μg/l.

Abstract Selective-area p-type doping has been regarded as one of the primary challenges in vertical GaN junction-based power devices. Nickel oxide (NiO), serving as a natural p-type semiconductor without the requirement for sophisticated activation and enabling adjustable charge concentration, is potentially feasible to form pn hetero-junction in GaN power devices. In this work, a vertical GaN Schottky barrier diode (SBD) featuring hybrid p-NiO junction termination extension (HP-JTE) with fluorine (F)-implanted buried layer (FIBL) has been demonstrated. With FIBL incorporated underneath p-NiO in the termination region, the reverse leakage current can be effectively reduced by approximately 3 orders of magnitude. By virtue of photon emission microscopy measurements, it has also been verified that the light emission and leakage current through p-NiO termination region can be effectively suppressed by FIBL. Thanks to the HP-JTE structure as well as the nearly ideal Schottky interface, the vertical GaN SBD exhibits a high current swing of $\sim 10^{13}$ , a low ideality factor of $\sim 1.02$ , a low differential $R_{O N}$ of $\sim 0.89 \mathrm{~m} \Omega \cdot \mathrm{cm}^2$ , a low forward voltage drop of $\sim 0.8 \mathrm{~V}$ (defined at $100 \mathrm{~A} / \mathrm{cm}^2$ ), and a breakdown voltage of $\sim 780 \mathrm{~V}$ (defined at $0.1 \mathrm{~A} / \mathrm{cm}^2$ ). The characterizations and findings in this work can provide valuable insights into the p-NiO/GaN hetero-junction-based power devices.

Prosthetic hands have significant potential to restore the manipulative capabilities and self-confidence of amputees and enhance their quality of life. However, incompatibility between prosthetic devices and residual limbs can lead to secondary injuries such as skin pressure ulcers and restricted joint motion, contributing to a high prosthesis abandonment rate. To address these challenges, this study introduces a data-driven design framework (D3Frame) utilizing a multi-index optimization method. By incorporating motion/ pressure data, as well as clinical criteria such as pain threshold/ tolerance, from various anatomical sites on the residual limbs of amputees, this framework aims to optimize the structural design of the prosthetic socket, including the Antecubital Channel (AC), Lateral Epicondylar Region Contour (LC), Medial Epicondylar Region Contour (MC), Olecranon Region Contour (OC), Lateral Flexor/ Extensor Region (LR), and Medial Flexor/ Extensor Region (MR). Experiments on five forearm amputees verified the improved adaptability of the optimized socket compared to traditional sockets under three load conditions. The experimental results revealed a modest score enhancement on standard clinical scales and reduced muscle fatigue levels. Specifically, the percent effort of muscles and slope value of mean/ median frequency decreased by 19%, 70%, and 99% on average, respectively, and the average values of mean/ median frequency in the motion cycle both increased by approximately 5%. The proposed D3Frame in this study was applied to optimize the structural aspects of designated regions of the prosthetic socket, offering the potential to aid prosthetists in prosthesis design and, consequently, augmenting the adaptability of prosthetic devices.

Assessing plantar pressure is crucial for fabricating diabetic insoles and preventing diabetic foot ulcers (DFUs), which are caused by increased plantar pressure. However, the commonly used methods for assessing plantar pressure distribution involve professional sensor-based equipment and expertise, which are costly and time-consuming. Given the qualitative association between ink footprint images and plantar pressure, this study proposes using the footprint images to predict the quantitative values of dynamic plantar pressure in barefoot and 4 different insole conditions (including Nora Lunalastik EVA, Nora Lunalight A fresh, Pe-Lite, and PORON® Medical 4708) based on a multilayer perceptron (MLP) neural network model. To provide more precise insole material recommendations for specific foot regions for better plantar pressure distribution, the plantar of the foot is divided into 5 regions: the toes, metatarsal heads, medial midfoot, lateral midfoot, and heel. Patch-based MLP with localization embedding is introduced to learn the correspondence between ink density and plantar pressure information. Ground-truth data collected from 52 diabetes patients is constructed as a dataset named diabetes-footprint-to-pressure and used to train and validate the model. The mean absolute error (MAE) of the models for the barefoot and 4 insole conditions is 5.51% (33.06 kPa), 3.99% (23.94 kPa), 4.85% (29.10 kPa), 4.25% (25.50 kPa), and 3.57% (21.42 kPa) of the sensing range, respectively. Compared to traditional methods for plantar pressure assessment, this approach streamlines the process of acquiring the overall and regional dynamic plantar pressure with barefoot and 4 different insole materials. Clinicians can quickly provide recommendations on the type of insole material for individual patients.

This paper introduces a novel hybrid parking system that integrates stationary charging piles (SCPs) and mobile charging robots (MCRs) to optimize urban parking utility. The model categorizes parking spaces based on the presence of SCPs, considering customer behavior including improper parking. It also introduces an operational algorithm—Earliest Available Device First (EADF)—to manage real-time scheduling of MCRs efficiently. Through strategic planning and operational management, the system aims to enhance social welfare by balancing cost-efficiency with flexible charging solutions. We evaluate our approach based on real-world data, demonstrating how MCRs significantly improve both the strategic and accumulated operational aspects of urban parking facilities. The results showcase the potential of hybrid systems in urban environments, promoting higher utility and cost-effective management.

With the development of Earth observation technology, it becomes easier and easier to acquire multimodal image data at the same time. To improve the performance of a multimodal remote-sensing detection algorithm, a new fusion feature optimization detection network is proposed. The method is designed to solve the problem of performance degradation caused by the unreliability of single-modal data in multimodal remote-sensing data. The key to obtain high-quality fusion features from multimodal data with interference is to suppress single-modal redundant features and fully integrate multimodal features. The proposed method mainly includes two improvements. First, a novel joint expression optimization module is designed to enhance the target features and suppress the redundant and interference features that affect the fusion effect. In addition, we propose a novel specific information enhancement module to further enhance the discriminative feature information of targets within each modal image. Experiments on the DroneVehicle dataset show that our proposed method is state of the art on this dataset.

Predicting cutterhead torque is essential for optimizing TBM construction strategies and minimizing jamming risks. This study presents a novel hybrid model (IEWOA-TSVD-ITELM), developed using data from 7635 tunneling cycles from the Yinsong Project, to enhance cutterhead torque prediction accuracy. The EWOA enhances its global search capability by introducing new position updating and adaptive adjustment strategies (IEWOA). In addition, by leveraging the Softsign function for the nonlinear transformation of the expected output matrix of the TELM, a third hidden layer is added to enhance the feature extraction capabilities (ITELM), whereas truncated singular value decomposition (TSVD) is employed to reduce the noise in the output matrix of the third hidden layer of the ITELM (TSVD-ITELM). Furthermore, the IEWOA optimized the number of neurons and randomly generated weights and biases in the TSVD-ITELM. This study comprehensively evaluates and compares six optimization algorithms using 25 standard test functions. Additionally, the IEWOA-TSVD-ITELM is compared with eight classical machine learning models. This study examines the impact of different timing lengths of the rising phase and rock mass grades on model performance. The results demonstrate the outstanding performance of the IEWOA as an optimization algorithm. The IEWOA-TSVD-ITELM model achieves an R2 value of 0.644 on the test set, with an MAE of 326.623 and an RMSE of 435.821, outperforming the other algorithms. Increasing the timing length from 30 to 60 s reduces the MAE and RMSE by 11.82% and 9.56%, respectively, but the gains diminish when the timing length increases from 60 to 90 s.

This paper introduces a new device concept and outlines the fabrication process of a bipolar junction transistor based on an IGZO/NiO/PbSe nanoband array heterostructure. We performed comprehensive electrical property testing and characterization analysis on the device to thoroughly assess the device's performance. The heterojunction structure efficiently amplifies the opto-electric responsivity and enhances the transmission efficiency of photogenerated carriers. Significantly, the phototransistor demonstrates a high photoresponsivity of 4000 A/W under an incident light power of 1 μW/cm2 and a wavelength of 850 nm. Hence, a solution detection system integrated with a bipolar phototransistor is designed. Through the analysis of the output signals, the system accurately determines the identity of the solute present in the solution.

In emotion recognition tasks, electroencephalography (EEG) has gained significant favor among researchers as a powerful biological signal tool. However, existing studies often fail to fully utilize the high temporal resolution provided by EEG when combining spatiotemporal and frequency features for emotion recognition, and do not meet the needs of effective feature fusion. Therefore, this paper proposes a multilevel multidomain feature fusion network model called MMF-Net, aiming to obtain a more comprehensive representation of spatiotemporal-frequency features and achieve higher accuracy in emotion classification. The model takes the original EEG two-dimensional feature map as input, simultaneously extracting spatiotemporal and spatial-frequency domain features at different levels to effectively utilize temporal resolution. Next, at each level, a specially designed fusion network layer is employed to combine the captured temporal, spatial, and frequency domain features. In addition, the fusion network layer contributes positively to the convergence of the model and the enhancement of feature detectors. In subject-dependent experiments, MMF-Net achieved average accuracy rates of 99.50% and 99.59% for valence and arousal dimensions on the DEAP dataset, respectively. In subject-independent experiments, the average accuracy rates for these two dimensions reached 97.46% and 97.54%, respectively.

With the rapid development of mobile networks and widespread use of mobile devices, there is an increasing focus on assigning location-based tasks to mobile users in the context of Mobile Crowd Sensing (MCS). One of the primary challenges in MCS is task assignment, i.e., distributing tasks to suitable users for completion. However, existing work often assumes static offline scenarios where the spatiotemporal information of all users and tasks is pre-determined and known. Neglecting the dynamic spatiotemporal distribution of users and tasks can lead to suboptimal assignment results. In this study, we investigate a novel task assignment problem called Community Task Assignment (CTA). The objective is to enhance the effectiveness and precision of task distribution by considering the sociality of current users and distributing location-based tasks through communities. Initially, we partition users into different communities by abstracting and identifying behavior patterns through the computation of minimum spanning trees, connectivity parameters, and community cohesion. Subsequently, we calculate the match between perception tasks and community behavior pattern features, and task distribution is carried out by the central nodes of the communities based on this match. Experimental validation first confirms the effectiveness of the community partitioning algorithm. Compared to existing algorithms, the proposed method more accurately detects community structures with similar behavioral features in the network. Furthermore, a comparison with existing task assignment algorithms verifies the superiority of the proposed method in terms of average task completion time, task matching rate, and overall utility of task assignments.

In this study, a backstepping sliding mode control method (NDO-ABSC) based on nonlinear disturbance observer (NDO) and adaptive line-of-sight guidance (ALOS) is proposed to address the three-dimensional path following control problem of underactuated autonomous underwater vehicles (AUVs) in the presence of ocean currents, unmodeled dynamics, and other unknown disturbances. First, the following error equations in the current environment are established in the Serret-Frenet coordinate system. Then, ALOS is designed to estimate the variations changes in angle-of-attack and crab angle caused by time-varying currents. In the kinematic controller, a parameter adaptive law, a control law for the virtual target point, and a virtual desired angular velocity are designed by backstepping. Subsequently, the composite uncertainty disturbance is observed and compensated by constructing a nonlinear disturbance observer, and a dynamic controller is designed by backstepping method, and a sliding mode control term is introduced in the control law to enhance the robustness of the system to the uncertainty disturbance. Finally, the good performance and strong robustness of the proposed method in 3D path following control are verified by numerical simulation.

The unique mining process of China's ion-adsorption rare earth (RE) mines has changed the structure of the mine ecosystem, and the interplay between the natural red soil characteristics and economic and social activities has exacerbated environmental problems such as the degradation of regional vegetation cover and soil erosion. These issues have had a profound and detrimental impact on the ecological security (ES) of the mining areas. The existing static evaluation study cannot comprehensively assess the ES status and dynamic evolution trend of the mining area, and cannot meet the needs of the complex ecosystem in the mining area. Therefore, this article constructs an ES evaluation index system based on the driver-pressure-state-impact-response-management causal framework model, and uses the variable weight (VW) theory to formulate a penalty-dominated state VW function to calculate the weight values of the indicators in different contexts of each year and evaluation unit. Finally, a dynamic evaluation of the spatial and temporal evolution trend of the ES of the Lingbei RE mining area is carried out during the period from 2000 to 2020. The geodetector model is then applied to reveal the driving factors impacting the ES of the mining area in different time periods. The results show that 1) Compared to the constant weight method, VW can provide a more detailed distribution of the ES level in the mining area, which has good application value in the small and dispersed ionic RE mining area. 2) The overall ES status of the Lingbei mining area shows a dynamic trend of deterioration followed by improvement and finally stabilization. 3) The vegetation health status is one of the most important driving factors of ES in the mine site, and the interaction between any two factors is greater than the explanatory power of the individual factors. This study provided insights into the ES and sustainable development of mining areas.

This work introduces a novel approach to improve the precision of distance estimation in localization systems by using existing LoRaWAN and RSSI-based techniques. Despite the benefits of range and power efficiency, these systems exhibit limited accuracy in practical situations. To address the limitation, this study provides an innovative technique that greatly improves the precision of distance estimations, particularly in urban environments. The fundamental basis of this approach lies in the use of a dynamic path loss model. An additional element is to accommodate the varied and dynamic conditions of signal transmission in metropolitan areas. A better Kalman filter is also used in the study. This is important because it reduces the effects of multipath fading and environmental noise that often make RSSI-based localization in LoRaWAN networks less accurate. The study further examines the influence of the environmental exponent, also known as the path loss exponent, on the RSSI results and the precision of the distance measurements. This methodology achieves the average error under 1 meters for indoor environments and under 7 meters for outdoor environments. Finally, the Cumulative Density Function (CDF) shows 90 % of the distance estimation algorithm error for indoor environment is lower than 1.08 meters while for outdoor environment is lower than 7.55 meters. Based on these improvements, the introduced methodology not only enhances and improves existing approaches but also optimizes the precision and dependability of urban localization technologies, with substantial implications for a variety of practical applications.

Automatic detection of mechanical failure is an essential technology in the fourth industrial revolution, which involves artificial-intelligence-based factory automation. The primary challenge in encoder-based machine sound anomaly detection lies in ensuring high-quality reconstruction of feature maps, as this directly impacts the precise definition of reconstruction error thresholds for normal and abnormal sound feature maps. This study proposes an improved deep convolutional generative adversarial network combined with the GANomaly method to introduce a unique anomaly detection model. This model leverages a residual deep convolutional generative adversarial network with an integrated attention mechanism as the generator and a multi-scale, multi-layer convolutional neural network as the discriminator to address the issue of information loss in reconstruction feature maps with increasing network depth, enhancing model generalization capabilities. The proposed approach introduces custom hyperparameters and tailored loss functions, utilizing Wasserstein distance to measure sample differences and promote model convergence. Researching the DCASE Challenge 2023 Task 2 development dataset, improvements are observed in experimental metrics such as AUC and pAUC, demonstrating the superiority of the model. We also analyze aspects such as feature map quality, parameter settings, and experimental ablation, and compare with other state-of-the-art methods to showcase the contributions of our model.

Determining threshold values in model based fault detection (MBFD) is a longstanding challenge, which is often addressed through empirical and ambiguous adjustments. To tackle this issue, this paper proposes an uncertainty-informed framework for quantitative threshold assessment. The framework comprises three stages: 1) identifying uncertainties by explicitly understanding the implemented MBFD method, 2) quantifying fault detection residual through uncertainty propagation, and 3) determining and optimizing threshold values based on the quantified misdiagnosis rates. To validate the effectiveness of the proposed approach, a case study of a modular multilevel converter is selected. The proposed method not only enables a quantified threshold assessment but also enhances the robustness of the fault detection by accounting for uncertainties.

Compact and high current inductors determine the power density and efficiency of voltage regulators (VRs) for computing ICs. Conventional moulding structures using powder cores and gapped ferrite cores are reaching the limits of size miniaturization. In this paper, a sandwich-structured inductor using 2D magnetic nanocrystalline flake ribbon (NFR) is developed to achieve high current density and low height. The analytical models of inductance and conduction losses are developed and validated by finite element methods and experiments. A 5 to 1 V, 50 A, VR is built to compare the performance of the fabricated NFR sandwich inductor to the latest commercial counterpart. Thermal performance is investigated, stability tests under solder iron heat and reflow temperature are performed. The fabricated sandwich inductor is featured as simple structure, low profile, low DC resistance (DCR), high saturation current, soldering heat stable, but in the modest compromise of the AC conduction and core losses.

Current differential protection (CDP) typically requires a reliable communications to implement the correct actions. Power-line communication (PLC) thanks to its natural advantage becomes the most preferred communication medium for the CDP service; however, it always undergoes to the deep multipath fading and the strong additive noises, which impede its deployment in the CDP. In this paper, we delicate to the performance improvement of CDP service by designing a reliable and robust PLC signal in the PLC transceiver. Firstly, a quadrature amplitude modulation (QAM) based OFDM signal employing frequency-hopping (FH) technique is proposed into the classic PLC, which can boost the capability of suppressing the noise and fading in PLC channel. On the basis of the multipath fading following Manfred-Klaus model, and additive noises (background noise and impulsive noise) following Bernoulli-Gaussian model, respectively, the impact of the proposed FH/OFDM PLC signal on the protection action performance is studied via the step-by-step analysis. Finally, based on the signal derivation, the mal-operation rate and refusal operation rate of CDP are evaluated by extensive simulations. The analysis and simulations reveal that the proposed CDP system can attain the superior correct actions (e.g., lower mal-operation rate and refusal operation rate), which are benefited from the reliability and robustness of the proposed FH/OFDM PLC signal, compared to the previous CDP employing classic OFDM PLC scheme.

Breast cancer often necessitates surgical interventions such as breast-conserving surgery or mastectomy. In these surgeries, sentinel lymph node (SLN) samples are often excised for histopathological examination to ascertain the presence of cancer metastasis. Despite its importance, traditional hematoxylin and eosin (H&E) staining, considerably prolongs the operation because of its complex processing requirements. Ultraviolet photoacoustic microscopy (UV-PAM) has emerged as a solution for bypassing the necessity of tissue staining or sectioning. However, its clinical application has been hindered by imaging speed. To overcome this challenge, we have developed a fast-scanning, reflection-mode UV-PAM designed for histopathology without staining based on high-sensitivity, wide-vision optical ultrasound detection. A specimen area of around 12 mm2 can be scanned in 8 min with a lateral resolution of 1.5 μm. To enhance imaging speed, multi-focal PAM was implemented, resulting in a fourfold acceleration. This PAM technique has been utilized in SLN biopsy to differentiate cancerous tissue in breast cancer patients.

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.

The seismic resilience has become a critical concern for power systems for high-impact, low-probability disasters such as severe earthquakes. In order to enhance seismic resilience for power systems, a novel protective device, the Passive Over-Current Trip (POCT), has been designed in this paper. In the design, the POCT integrates a fuse and a mechanical switch, able to trigger the breaker to interrupt the circuit to ensure effective protection. The POCT has the trip-free feature as it fulfills its protective function without the need for an electric relay requiring a continuous power supply. The proposed POCT provides an immediate and reliable solution for the protection in high and ultra-high voltage applications. To validate the effectiveness of the proposed POCT, shockproof and tripping tests were conducted. A seismic test waveform from the Fukushima earthquake that occurred on March 11, 2011 was used to test POCT. Under such a severe earthquake, no structural or mechanical damage on the associated components in the POCT was found. Also, the POCT is also in compliance with IEEE 693 standard. Finally, various tripping tests show that the POCT can operate within 18 ms when a fault occurs in a power system, which is much shorter than that of a typical high-voltage fuse.

While computer vision and computing technology advances have facilitated advanced driver assistance applications, systems with multi-task design remain highly demanding to operate at high speed on resource-constrained devices. Our study addresses this challenge by proposing a real-time driver assistance solution specifically developed for a single Jetson AGX Xavier embedded device. It simultaneously performs lane detection, traffic object detection and recognition, and rule-based scene analysis. To achieve high throughput (up to 43 frames per second) without reliance on additional hardware or cloud server, the system exploits Jetson device’s specialized AI accelerator and employs various optimization techniques: multithreaded programming, the TensorRT framework, and post-training quantization. The modular design integrates state-of-the-art task-specific methods and ensures adaptation to diverse traffic scenarios across countries as well as future hardware and solutions. Experimental results using a Korean dashcam traffic dataset validated the system’s effectiveness and practicality.

This article tackles the visual localization of unmanned aerial vehicles (UAVs) in the presence of multisource and cross-view images are involved. We present a lightweight end-to-end scene graph encoding and matching network that finds the best matches for the airborne camera views from the reference image maps. The scene graph addresses the challenges of encoding the semantic scene by aggregating the image convolutional features into global and structured semiglobal descriptors. The principal contributions of this article are as follows: First, we develop a new network architecture that embeds a nonlocal block and a modified vector of locally aggregated descriptors network (NetVLAD) into a backbone convolutional neural network. The main component of the modified NetVLAD is a cluster similarity masking graph (CSMG) encoder, which is proposed to replace the feature-cluster residuals computing in NetVLAD with cluster consensus feature aggregation and structure-aware scene graph extraction. In addition, a global descriptor is extracted by a nonlocal block to label each image with a discriminative global feature descriptor. Second, we develop a new triplet loss for the network training procedure to learn the features at different semantic levels. The proposed global descriptor and CSMG encoder are trained together according to a weighted sum of cosine triplet losses. Third, the global descriptor from the nonlocal block and semiglobal descriptor from the CSMG encoder work hierarchically for coarse-to-fine image retrieval and can achieve real-time efficiency and favorable accuracy of image searching and matching from the reference image map. We train and test the model on two challenging benchmark datasets. We also test the pretrained model on a dataset collected by a fixed-wing UAV to further evaluate the model's generalizability. The benchmark evaluations and ablation experiments show that the developed method outperforms state-of-the-art methods and achieves superior performance in the real-time matching of UAV images and reference image maps for UAV visual localization. Open-source code is available on GitHub.

Skin cancer recognition poses a significant challenge in the field of deep learning. While conventional convolutional neural networks have been extensively employed for classifying skin cancer images, their fixed receptive field limits their ability to capture the global features present in such images. Conversely, transformer-based models that rely on self-attention can effectively model long-range dependencies, but they come with high computational complexity and exhibit certain limitations in local feature induction. To address this issue, this paper presents a novel skin cancer recognition network named ECRNet. ECRNet has been designed to effectively capture both global and local information, and it introduces an explicit vision center to accomplish this purpose. Moreover, this paper presents a feature fusion module known as the CCPA block. This module utilizes both coordinate attention and channel attention mechanisms to extract image features and enhance the representation of skin cancer images. To evaluate the performance of ECRNet, extensive experimental comparisons were conducted on the ISIC2018 dataset. The experimental results demonstrate that ECRNet outperforms the baseline model, showing improvements of 1.19% in accuracy (ACC), 1.96% in precision, 4.08% in recall, and 3.28% in the F1 score.

The Metaverse relies on advanced machine learning (ML) techniques to facilitate the seamless mapping between the virtual and physical realms. ML-based technologies also enable metaverse service providers (MSPs) to offer a diverse range of intelligent virtual services to metaverse users (MUs). However, it can be challenging for MSPs to collect sufficient metaverse data to train ML models by themselves. As a result, MSPs can be interested in seeking contributions from MUs in both ML data and models. To address these challenges, we propose MetaAICM, a blockchain-based framework that empowers the metaverse through two key components. Firstly, it incorporates a distributed crowdsourcing system that allows MSPs to gather metaverse data and ML models from MUs. Secondly, it features a decentralized marketplace, enabling MUs to proactively collect data and train ML models for sale using their metaverse devices and computing resources. MetaAICM leverages blockchain and smart contracts to achieve decentralization, ensuring security and privacy without relying on a trusted third-party authority or additional trust assumptions between MUs and MSPs. Numerical studies show that MetaAICM offers high processing performance and cost efficiency, while the framework is implemented on top of a consortium blockchain to show its feasibility.

The task of background suppression in infrared small-target scenarios aims to eliminate irregular noisy backgrounds while preserving targets with high-frequency features. In infrared small-target scenes at long distances, the backgrounds become complex and the target features are degraded, highlighting a significant disparity between the detailed and realistic background and the limited features of the targets. To address these challenges, we propose a patch-based semantically enhanced generative adversarial network (GAN) named PSEnet for background suppression in infrared small-target scenarios. First, we introduce a patch-scale GAN that allows the model to concentrate on local background suppression. This shift from a global to local perspective simplifies the complexity of background suppression. Second, we employ the PSE module consisting multiscale dilated convolution and adaptive weight fusion to extract local semantic information. Third, by segmenting the infrared image into smaller patches and resampling them, we create a more balanced dataset for adversarial training. Experimental results demonstrate that the proposed algorithm significantly improves the signal-to-noise ratio of dim and small targets, reduces the missing detection rate, and achieves a precision of almost 91%. In conclusion, this approach effectively uses GANs for background suppression in complex environments.

Nonlinear discrete control system theory, which is increasingly garnering attention, has important applications in computer control theory. In this study, the robust control problem of nonlinear discrete systems with perturbations is investigated. Using the Lyapunov theory and linear matrix inequality method, the sufficient conditions for fast convergence of the state observer estimation error and fast stability of the state feedback closed-loop system are formulated, thereby showing that the independent solution of the gain matrix of the state observer and robust controller is more convenient. When designing the state observer and observer-based robust controller, the gain matrix was constrained twice, reducing the time required for robust control. The proposed robust control method based on the state observer reduces the conservatism of the system and improves the robust control effect of nonlinear discrete systems. Concurrently, the observer-based robust control method in normal systems is extended to the generalized system form. Finally, the proposed robust control method was simulated using the longitudinal motion model of an aircraft and the DOLPHIN MARK II autonomous underwater vehicle motion model, as research objects. In comparisons with two control methods from the literature, the average error of state estimation was significantly reduced, and the closed-loop control system rapidly achieved robust stabilization, demonstrating the effectiveness and superiority of the control method proposed in this paper.

Large Language Models (LLMs) excel in interactive chat scenarios due to their advanced conversational abilities. However, their training process invariably exposes them to a diverse range of harmful or toxic content, posing significant challenges in ensuring that LLM responses align with human ethical values. Consequently, the detection and quantification of adverse content remains a paramount issue in contemporary research. In this paper, we introduce the SAFE dataset, a novel resource designed to advance safety assessment research in LLMs. Our dataset extends beyond the binary categorization of content into “safe” and “unsafe”. Drawing upon human interpretations of safety, we further delineate unsafe content into six granular categories: Sensitivity, Harmfulness, Falsehood, Information Corruption, Unnaturalness, and Deviation from Instructions. This refined classification aims to enhance LLMs’ ability to discern unsafe data more accurately. In total, we have created a dataset comprising 52,340 instruction-response pairs, each annotated with safety meta-tags. Additionally, we have compiled expert comparative assessments for these indicators. We developed a multi-expert rating model trained on the SAFE dataset, designed to evaluate the responses of LLMs across various dimensions. This approach highlights the potential of our dataset in the realm of safety assessment for LLMs. The model’s capability to provide multi-faceted evaluations reflects an advanced understanding of the nuanced requirements in LLM response assessment. We believe this dataset represents a valuable resource for the community, contributing to the safe development and deployment of LLMs. Our findings and resources are poised to fuel future research endeavors in this domain.