Using sparse sensors to capture human movements and driving virtual humans in computers to reproduce these movements is one of the key technologies in fields such as virtual reality,among which,the relevant calculations for reproducing movements must simultaneously satisfy motion constraints and physical constraints.Currently,nonlinear optimization methods are primarily employed to address the physical constraints in such computations.However,these methods suffer from several drawbacks,including long computation time,high computational complexity,and the necessity of designing dedicated optimizers.To address these issues,a method based on a deep neural network model for solving the physical constraints optimization in sparse sensor motion capture is proposed.Firstly,the model effectively combines the multi-modal feature fusion network and the multi-path refinement network to form a post-fusion multi-level structure,which serves as the basic model of this research.Secondly,through a progressive fusion method,back-projection connections between the layers of the basic model are established,enabling the iteration of the basic model.This allows the information fused at deeper layers to be utilized by the shallower layers.Thirdly,a loss function in the form of combined weighting that incorporates physical constraints is proposed,which is suitable for fine-tuning the deep model in the presence of both implicit and explicit physical constraints.The experimental results demonstrate that the proposed method not only exhibits good feasibility but also improves computational efficiency by approximately 20 percentage points.Compared with other commonly used optimization methods,the proposed method performs better in terms of four mainstream evaluation indicators.Additionally,the method yields favorable results when applied to other datasets,demonstrating its strong generalization capability.This method provides a new perspective for the formation of an end-to-end deep model for sparse sensor motion capture.
Non-negative Matrix Factorization(NMF)is a dimensionality reduction technique based on matrix factorization,used to find linear representations based on latent features.Although Graph Nonnegative Matrix Factorization(GNMF),which is proposed to address the issue of NMF ignoring the local geometric structure of data,can improve the geometric relationship between data in high-dimensional spaces relatively well,the graph information it follows-that is,the adjacency matrix-is constructed based on the distance relationship of the observable space of the original samples and cannot reflect the true distance relationship between objects.To address this issue,we integrate the graph information based on self-representation into the algorithm framework of NMF and propose Graph Non-negative Matrix Factorization based on Self-representation Learning Update(GNMFSLU).This method updates the adjacency matrix of the graph in each iteration,thereby making the graph information closer to the true distance relationship between objects and enhancing the clustering performance of NMF.
Given the uneven global distribution of livestock resources leading to a mismatch in protein supply and demand,and the current lack of clarity regarding the mechanisms and evolution of virtual protein flows embedded within meat trade,this study constructs a global virtual meat protein trade network for the period 1995～2022.Our aim is to deeply analyze its structural evolution,geographical patterns,and inherent inequalities.We converted global meat trade data into virtual protein flows and employed complex network analysis to comprehensively examine the network's topological indicators,including average degree,graph density,clustering coefficient,path length,number of communities,and modularity.Additionally,the Gini coefficient was introduced to quantify import and export inequalities.Results reveal a significant increase in the total volume of global virtual meat protein trade,with poultry experiencing particularly rapid growth.The trade pattern is highly concentrated,with China,the United States,and Germany accounting for 25.3%of global virtual protein imports.Trade relations have evolved from regional to global,with emerging economies playing an increasingly pivotal role.Network structure analysis indicates enhanced connectivity,dynamic changes in efficiency and clustering,and clearer community delineation.The Gini coefficient confirms a significant inequality in global virtual meat protein trade,especially pronounced in exports.This research is the first to construct a global virtual protein trade network for meat and reveals the mechanisms of trade concentration and inequality from the perspective of the coupling of nutritional value flow and network structure.
Domain adaptive retrieval(DAR)has obtained the high-precision and fast retrieval across different domains by reducing domain discrepancies.However,existing works still face two issues:(1)incorrect pseudo labels in the target domain cause error accumulation during training,and(2)most existing methods are limited to single-label works and ignore the requirement for finer-grained retrieval.To address these problems,we propose a method called Dual Semantic Guidance and Cluster Matching for Domain Adaptive Retrieval(DSG-CM).First,samples with high confidence in the target domain are gradually selected to reduce the impact of incorrect pseudo-labels.Second,the distinctiveness of samples is enhanced by incorporating multi-label semantic information from features with the class label information of the samples.Lastly,samples are divided into micro-clusters and the inter-cluster relationships between different domains are explored to enhance the retrieval precision.This method can generate compact and efficient hash codes,leading to superior performance.Experiment results on three benchmark datasets demonstrate that our method outperforms others.
Traffic flow prediction is an important technology in intelligent transportation systems(ITS).Accurate traffic prediction can reduce congestion and improve traffic efficiency.However,traffic flow data contains complex temporal relationships,and capturing dynamic traffic spatial relationships is a challenge.In order to improve the prediction accuracy,a dynamic graph multi temporal perspectives attention network(DGMAN)is proposed,based on the spatiotemporal data of traffic flow.The model uses a dynamic graph learning module(DGLM)to extract the dynamic relationship information between traffic nodes in traffic data by establishing a dynamic graph.In complex temporal data,the multi temporal perspectives attention mechanism(MtpA)captures the temporal dependence of traffic flow and mines potential temporal relationships.Finally,the proposed model is tested on 4 real-world datasets.Compared with the baseline models,DGMAN achieves the best performance in the mean absolute error(MAE),root mean square error(RMSE)and mean absolute percentage error(MAPE)evaluation metrics.
Ramp merging areas,as critical nodes for intelligent vehicle traffic,are characterized by high dynamic interaction and strong uncertainty,making it difficult for traditional path planning algorithms to balance safety,real-time performance,and smoothness in such complex scenarios.To address these challenges,a path planning algorithm is proposed integrating an improved artificial potential field and line-of-sight A*.Firstly,a piecewise potential field model comprising guidance,obstacle avoidance,and boundary constraints is constructed.A dynamic weight adjustment mechanism is also introduced to adapt to the kinematic requirements of vehicles at different merging stages,efficiently resolving the unreachable target problem of traditional potential field methods.Secondly,a physical line-of-sight check mechanism incorporating safety thresholds is embedded into the A* algorithm.By utilizing a non-axial connection strategy to prune redundant nodes,the node expansion direction is optimized to enhance search efficiency.Finally,the improved potential field is mapped as the heuristic cost function of the A*algorithm,achieving a deep coupling of global path planning and local risk perception.Simulation results demonstrate that,compared with traditional A*,the Rapidly-exploring Random Tree Star(RRT*)algorithm,and existing analogous fusion schemes,the runtime of the proposed algorithm is reduced by 70.8%,the minimum obstacle avoidance distance is increased by 106.6%,and lateral jerk is decreased to 0.62 m/s3.The proposed algorithm effectively enhances the safety and stability of trajectory planning,demonstrating its efficiency and robustness in complex merging scenarios.
Existing six-degree-of-freedom(6-DoF)grasp pose detection methods remain weak in explicitly modeling grasp stability and physical feasibility,making it difficult to ensure high grasp success rates for robotic execution in cluttered scenes.To address this limitation,we propose DPCGrasp,an end-to-end 6-DoF grasp detection method that incorporates differentiable physical constraints.The proposed method introduces four physically motivated constraints:antipodal alignment,surface flatness,center-of-mass proximity,and contact tolerance.These constraints are formulated as differentiable regularization terms and integrated into the training objective to promote physically plausible grasp configurations.To enhance local geometric understanding around candidate grasp points,we design a multi-scale cylindrical sampling and feature fusion module.Furthermore,we develop a self-attention-based multi-parameter grasp prediction head to capture latent dependencies among grasp parameters,improving the consistency of parameter outputs under task-decoupled learning.Experimental results show that the proposed method improves the average precision by 4.66 percentage points on the large-scale GraspNet-1Billion dataset compared to state-of-the-art methods.In real-world robotic experiments,it attains a 7.83 percentage points increase in average grasp success rate,confirming its effectiveness and practical feasibility in actual grasping scenarios.
Heat release rate(HRR)is one of the most important parameters in fire dynamics,directly reflecting fire intensity and the energy release rate during combustion.Traditional HRR recognition methods have fixed receptive fields,struggle with multi-scale flame variations,and often ignore critical regions.In this study,a fire HRR recognition method is proposed based on the multi-scale atrous convolution attention fusion module(MSACAF)and the segment anything(SAM),aiming to improve the accuracy of HRR estimation.This algorithm is based on the ResNet-18 backbone and introduces multi-scale atrous convolution to adapt to flames of different sizes and shapes and extract richer features.By combining channel and spatial attention mechanisms,it allocates weight information effectively,allowing the model to focus on key flame regions.In this study,128 combustion videos were selected from the NIST fire calorimetry database(FCD).Flame images from the videos were fed into the SAM large model for segmentation,generating a dataset of 48 841 segmented flame images to reduce background and non-fire feature interference on prediction accuracy.The experimental results show that the proposed model outperforms other deep neural network models.Ablation experiments validate the effectiveness of MSACAF,and the accuracy of HRR prediction improves by 4.4%.The results demonstrate that the proposed method achieves higher accuracy in HRR-based recognition,offering new insights for risk assessment.
Networked electric vehicles rely on wireless communication to achieve cooperative lateral motion control.However,the measurement and control channels are vulnerable to denial-of-service(DoS)attacks,and external disturbances together with communication delays can significantly degrade lateral control performance.In this research,an investigation is conducted on the secure consensus problem for a multi-vehicle lateral dynamics system with time delays and parametric uncertainties under DoS attacks,and an event-triggered resilient distributed controller with switching gains is designed.First,a time-delay model of the vehicle lateral dynamics is established,and the closed-loop system is represented as a switched system composed of a normal mode and a DoS-affected mode.To improve the utilization of communication and control resources,an event-triggered mechanism is introduced to determine the instants for updating the control inputs.Second,to overcome the enlargement of the effective attack duration caused by the misalignment between the termination of each DoS interval and the event-triggered sampling instants,a recovery-time compensation scheme is designed to cover the waiting time associated with the triggering margin.Then,by constructing multiple Lyapunov-Krasovskii functionals,a set of bilinear matrix inequalities(BMI)is derived,and their solution provides the desired feedback gain matrices for the system under different operating modes.Numerical simulations demonstrate that the proposed control scheme effectively withstands DoS attacks,suppresses external disturbances,and enhances the lateral consensus performance of vehicles.
Laser has been widely used in various areas such as advanced manufacturing and modern healthcare,and played a critical role for accurate detection of laser spot.Considering that the detection is difficult as it is often interfered by the noise light from the background,we propose a new method based on multi-scale neighborhood search(MNS)for laser spot detection in this study.Firstly,a multi-scale difference space for laser spot is constructed,and within this space,a neighborhood search method is designed for estimating the spot center with maximum response.Technically,the extremum point with the maximum response value is searched by comparing neighboring points within a multi-scale space,and the center of the spot is estimated by fitting the local grayscale surface in the image.Then,the region of interest for the spot is extracted,and the directions of the long axis and the short axis are determined by accumulating gradient magnitudes within the region.Also,the length of each axis is obtained by using the second-moment method.The effective performance of the proposed method is verified through simulation experiments,significantly improving the accuracy of spot detection in noisy environments.Meanwhile,the detection errors are consistently low in different noise environments.In particular,under uniform light backgrounds,the average detection accuracy improved by at least 30%compared to the existing methods in the market.Even with natural backlight close to 20%of the peak intensity,the parameter detection errors for laser spot remain below 1%.The results indicate that our method maintains strong robustness in detecting laser spots under various lighting conditions,providing a feasible detection method for laser spots in complex environments.
Quadruped robots have the potential to traverse challenging natural environments with payloads.Recent advancements have demonstrated the efficacy of optimization-based control methods for locomotion of quadruped robots;however,these methods rely heavily on precise dynamic models to meet high performance.When robots carry unknown payloads,the resulting payload-induced dynamic uncertainties can significantly degrade performance or even lead to failure in practical applications.To address this issue,a motion control method integrating quadratic programming(QP)with adaptive control techniques is proposed.First,an adaptive parameter estimator based on Lyapunov stability analysis is designed to achieve online identification of the unknown payload mass.Subsequently,a nonlinear disturbance observer(DOB)is introduced to provide real-time compensation for torque disturbances caused by payload variations.Building upon this foundation,a QP-based balance controller is developed to solve optimal foot contact force allocation under friction cone constraints.Finally,simulations and real-world experiments under various payloads and operating conditions are conducted on the Unitree Go1 quadruped robot.Experimental results demonstrate that the proposed method outperforms comparison methods in terms of centroid trajectory tracking accuracy and body posture stability,verifying its effectiveness and robustness.
Aiming at the problems of dynamic environmental changes,a lack of natural environmental characteristics,and the dependence on the number of reflector columns of the traditional trilateral localization algorithm based on reflective columns under the ship segmentation operation environment,an improved positioning strategy by introducing sensors such as Light Detection and Ranging,Inertial Measurement Unit(IMU),and odometer,and fusing multi-source sensor data is proposed.The algorithm fuses trilateral positioning with the iterative closest point(ICP)algorithm for point cloud matching for positioning and introduces a factor graph optimization framework to achieve multi-source data fusion.The experimental platform is built to carry out the localization test,and the static X/Y/heading localization error reaches 12.876 mm,4.273 mm and 0.000 3 rad respectively,and the X/Y/heading localization error under dynamic and complex conditions reaches 33.364 mm,16.95 mm and 0.026 3 rad respectively,which is better than the traditional method in terms of accuracy and robustness.The experimental results show that the proposed localization strategy reduces the static X/Y/heading localization error by more than 5.1%compared with the traditional trilateral localization and Kalman filtering,and reduces the dynamic complex X/Y/heading localization error by more than 14.5%.
Cross-network node classification aims to transfer knowledge extracted from a source network with sufficient labeled nodes to predict labels for nodes in a target network.Existing methods mainly focus on associating two networks in a shared embedding space,where the distribution discrepancy of node embedding across two networks isminimized.However,the relationship across two networks remains underexplored.In this paper,we propose a method that learns cross-network relationship among nodes from different networks and performs information aggregation across networks based on our learned relationship.To achieve this,we introduce a labeled Fused Gromov-Wasserstein model based on optimal transport,which exploits feature,structure and label information to construct node association between two networks.Based on the cross-network relationship,we design a cross-network graph convolutional network to learn node embeddings.Experimental results on several benchmark datasets show the superiority of the proposed method over state-of-the-art methods.
The advancement of shrimp farming is generally hampered by limited technical expertise among practitioners,a shortage of domain specialists and local experts with hands-on experience,and delayed technical guidance,leading to managerial inefficiency and elevated operational risks.To fill this gap,an intelligent question-answering model that integrates knowledge graph with user intent recognition is proposed,aiming to provide scientific and efficient decision-making support for shrimp farming.The proposed model systematically consolidates empirical knowledge from local experts and experimental data from domain specialists to construct a vertical knowledge graph,which encompasses critical aspects of the shrimp farming process,including environmental water quality control,cultivation management,and disease prevention.On this basis,a dual-channel query preprocessing framework is designed,combining intent recognition with semantic enhancement.This designed framework utilizes intent recognition technique to accurately direct ambiguous user queries to pertinent sub-graphs,thereby improving retrieval precision.Meanwhile,it leverages parallel semantic enhancement technique to enrich query context,increasing the information density of sparse queries and activating potential entities and relationships within the knowledge graph.Additionally,community-level graph retrieval is adopted to generate well-structured and context-aware answers.Experimental results show that,the proposed model outperforms four baseline models,in terms of answer completeness and faithfulness.On factual questions,it achieves a 20.73 percentage points improvement in completeness compared with standard retrieval-augmented generation models.On causal questions,it elevates faithfulness by 13.97 percentage points over simple hybrid retrieval models.Ablation studies show that,the synergy between intent recognition and semantic enhancement is critical to the performance of the proposed model.Furthermore,domain adaptation analysis demonstrates that the proposed model maintains stable performance across diverse farming scenarios.
Few-shot knowledge graph completion(FKGC)aims to infer missing triples within long-tail relations by leveraging a limited number of reference instances.Existing FKGC models struggle to effectively distinguish informative neighbors from noisy ones during the aggregation of neighborhood information for central entities.Moreover,in the matching and prediction phase,they typically rely solely on entity pair similarity,which often leads to biased predictions when the reference triples are unevenly distributed.To address these challenges,MhAMM,a novel FKGC model,is proposed based on multi-head attention matching.In the neighborhood aggregation stage,MhAMM introduces a multi-head attention mechanism tailored to the sparsity characteristics of FKGC tasks,which effectively amplifies the attention weights of informative neighbors while suppressing the influence of noisy ones,thereby improving the encoding quality of central entities.In the matching stage,a multidimensional matching network is designed,which integrates both the entity pair similarity score and a triple plausibility score computed via a fully connected neural network.These two complementary scores jointly enhance the overall matching performance.Extensive experiments on public datasets demonstrate that MhAMM consistently achieves significant improvements across multiple evaluation metrics,verifying the effectiveness and robustness of the proposed model.
In this research,an automated screw fastening robot system for container floors is designed and implemented,aiming to resolve the issues of low efficiency,high labor intensity,and safety deficiencies resulting from the long-term dependence on manual operation in the container floor screw fastening process.The system integrates a mobile chassis,a multi-joint robotic arm,a high-precision visual positioning module,an end-effector,and an automatic screw feeder for automated identification,positioning,and fastening of screw holes.A mechanical model is constructed to guide the development of a pneumatic fastening device.Experimental results demonstrate that the visual module achieves a 96.5%recognition rate for unfastened holes,with mean positioning errors of 0.50 mm(x-axis)and 0.47 mm(y-axis).The feeder delivers screws in 0.21 s at pressures above 0.35 MPa,enabling a total fastening cycle time of 4.5 s per screw.The system proves stable,significantly enhancing operational automation,efficiency,and consistency,thereby offering a viable solution for intelligent upgrading in container manufacturing.
High-efficiency organic blue light-emitting materials are the key components to construct high-performance organic light-emitting diode(OLED)devices.In this research,a novel asymmetric pyrene-based blue emitter(TPE-o2Ph)with aggregation-induced emission(AIE)characteristics was synthesized.The pyrene-based AIE luminogen exhibits blue emission with the maximum emission peak(λem)at 460 nm and high fluorescence quantum yield of 0.90 in solid state.In addition,the compound TPE-o2Ph was utilized as the light-emitting layer for non-doped/doped blue OLED devices with excellent electroluminescence(EL)performance.Moreover,the non-doped OLED exhibits excellent EL performance,with a maximum luminescence(Lmax)and the maximum external quantum efficiency(EQE)of 12 560 cd·m-2 and 2.34%.
In many practical applications,faults are frequently caused by the occurrence of specific events in succession(i.e.,pattern faults)rather than a single failure event.The safe diagnosis method based on pattern faults can diagnose the event string that triggers faults.However,the static observation of the original system may not be able to fully capture system faults for complex systems due to the pre-defined and fixed set of observable and unobservable events,leading to the omission of the fault diagnosis.The purpose of this research is to address the limitations of traditional static observation methods.For this reason,a safe diagnosis method based on the dynamic observation of pattern faults in stochastic discrete-event systems(SDESs)is proposed.First,the concepts of S-type and T-type pattern safe diagnosability for SDESs under dynamic observations are formally introduced.Then,a pattern-safe diagnoser and a forbidden language recognizer are constructed to diagnose the pattern faults.Finally,the necessary and sufficient conditions for pattern-safe diagnosability of SDESs under dynamic observations are presented,and an example is provided to illustrate the results.
To address CO2 emissions caused by excessive fossil fuel consumption,a novel Ni-N co-doped biomass-derived carbon catalyst(Ni-NBC-800)is developed using agricultural waste corn stalks,aiming to achieve electrochemical CO2 reduction(CO2 Reduction Reaction,CO2RR)for synthesizing syngas(CO/H2)with tunable ratios and promote CO2 resource utilization.A NH4Cl-assisted pore-forming strategy combined with nickel impregnation and high-temperature pyrolysis is employed to construct a composite catalyst featuring hierarchical porous structures and metal-nitrogen active sites.Experimental results demonstrate that with the optimized mass fraction of the load Ni(2%)and calcination temperature(800℃),the catalyst achieved a CO Faradaic efficiency of 72.8%at-0.8V,while the molar ratio of CO to H2 can be continuously adjusted within the range of 0.75 to 3.15 through potential regulation,meeting downstream syngas process requirements.Compared with commercial activated carbon-based catalyst(Ni-NAC-800),Ni-NBC-800 exhibited superior CO partial current density(-4.75mA/cm2)and stability(the 24 h current retention rate of 90.3%).Characterization analyses revealed that the hierarchical porous structure formed by NH3/HCl gas synergistic etching during NH4Cl pyrolysis significantly enhanced reactant mass transfer and catalytic activity.This work not only validates the feasibility of agricultural waste-derived carbon materials as alternatives to commercial carbon supports,but also provides a new strategy for low-cost and tunable syngas electrosynthesis through metal-support synergistic design.
To address the issue that the presence of dynamic clutter in actual environments will affect the accuracy of human target localization and vital signs detection,a coarse-to-fine point cloud selection strategy and an adaptive variational modal decomposition method based on the quality factor are proposed to achieve the suppression of dynamic clutter and the enhancement of vital signs detection performance.First,coarse point clouds of the human body and dynamic objects are distinguished by autocorrelation analysis.Second,a spectrum-based multi-feature fusion model is proposed to select fine point cloud with strong vital signs.Third,a quality factor-based variational mode decomposition method is proposed to separate the dynamic clutter and weak vital signals.Finally,a harmonic weighting selection algorithm is proposed to adaptively extract the respiratory and heartbeat components.Experiments conducted in cluttered indoor environments show that the proposed method effectively mitigates the effects of dynamic clutter and achieves accurate detection of human vital signs in dynamic environments,achieving respiratory and heart rate accuracies of 98.01%and 98.14%,respectively.
