We show that the torsion module Tor_(j)^(R)(R/a,H_(a)^(i)(X))is in a Serre subcategory for the bounded below R-complex X.In addition,we prove the isomorphism Tor_(s-t)^(R)(R/a,X)≅Tor_(s)^(R)(R/a,H_(a)^(t)(X))in some c...We show that the torsion module Tor_(j)^(R)(R/a,H_(a)^(i)(X))is in a Serre subcategory for the bounded below R-complex X.In addition,we prove the isomorphism Tor_(s-t)^(R)(R/a,X)≅Tor_(s)^(R)(R/a,H_(a)^(t)(X))in some case.As an application,the Betti number of a complex X in a prime ideal p can be computed by the Betti number of the local cohomology modules of X in p.展开更多
This experiment obtained different laser energy density(LED) by changing SLM molding process parameters.The surface morphology, surface quality, and microstructure of as-fabricated samples were studied. The effects of...This experiment obtained different laser energy density(LED) by changing SLM molding process parameters.The surface morphology, surface quality, and microstructure of as-fabricated samples were studied. The effects of scanning speed, hatching space, and laser power on surface quality were analyzed, and the optimal LED range for surface quality was determined. The results show that pores and spherical particles appear on the sample’s surface when low LED is applied, while there are lamellar structures on the sides of the samples. Cracks appear on the sample’s surface,and the splash phenomenon increases when a high LED is taken. At the same time, a large amount of unmelted powder adhered to the side of the sample. The surface quality is the best when the LED is 150-170 J/mm^(3). The preferred hatch space is currently 0.05-0.09 mm, the laser power is 200-350 W, and the average surface roughness value is(15.1±3) μm.The average surface hardness reaches HV404±HV3, higher than the forging standard range of HV340-HV395.Increasing the LED within the experiment range can increase the surface hardness, yet an excessively high LED will not further increase the surface hardness. The microstructure is composed of needle-like α’-phases with a length of about 20μm, in a crisscross ‘N’ shape, when the LED is low. The β-phase grain boundary is not obvious, and the secondaryphase volume fraction is high;when the LED is high, the α’-phase of the microstructure is in the form of coarse slats, and the secondary-phase is composed of a small amount of secondary α’-phase, the tertiary α’-phase and the fourth α’-phase disappear, and the volume fraction of the secondary-phase becomes low.展开更多
The multi-compartment electric vehicle routing problem(EVRP)with soft time window and multiple charging types(MCEVRP-STW&MCT)is studied,in which electric multi-compartment vehicles that are environmentally friendl...The multi-compartment electric vehicle routing problem(EVRP)with soft time window and multiple charging types(MCEVRP-STW&MCT)is studied,in which electric multi-compartment vehicles that are environmentally friendly but need to be recharged in course of transport process,are employed.A mathematical model for this optimization problem is established with the objective of minimizing the function composed of vehicle cost,distribution cost,time window penalty cost and charging service cost.To solve the problem,an estimation of the distribution algorithm based on Lévy flight(EDA-LF)is proposed to perform a local search at each iteration to prevent the algorithm from falling into local optimum.Experimental results demonstrate that the EDA-LF algorithm can find better solutions and has stronger robustness than the basic EDA algorithm.In addition,when comparing with existing algorithms,the result shows that the EDA-LF can often get better solutions in a relatively short time when solving medium and large-scale instances.Further experiments show that using electric multi-compartment vehicles to deliver incompatible products can produce better results than using traditional fuel vehicles.展开更多
With the warming up and continuous development of machine learning,especially deep learning,the research on visual question answering field has made significant progress,with important theoretical research significanc...With the warming up and continuous development of machine learning,especially deep learning,the research on visual question answering field has made significant progress,with important theoretical research significance and practical application value.Therefore,it is necessary to summarize the current research and provide some reference for researchers in this field.This article conducted a detailed and in-depth analysis and summarized of relevant research and typical methods of visual question answering field.First,relevant background knowledge about VQA(Visual Question Answering)was introduced.Secondly,the issues and challenges of visual question answering were discussed,and at the same time,some promising discussion on the particular methodologies was given.Thirdly,the key sub-problems affecting visual question answering were summarized and analyzed.Then,the current commonly used data sets and evaluation indicators were summarized.Next,in view of the popular algorithms and models in VQA research,comparison of the algorithms and models was summarized and listed.Finally,the future development trend and conclusion of visual question answering were prospected.展开更多
In this paper,the optimal control of non-linear switching system is investigated without knowing the system dynamics.First,the Hamilton-Jacobi-Bellman(HJB)equation is derived with the consideration of hybrid action sp...In this paper,the optimal control of non-linear switching system is investigated without knowing the system dynamics.First,the Hamilton-Jacobi-Bellman(HJB)equation is derived with the consideration of hybrid action space.Then,a novel data-based hybrid Q-learning(HQL)algorithm is proposed to find the optimal solution in an iterative manner.In addition,the theoretical analysis is provided to illustrate the convergence and optimality of the proposed algorithm.Finally,the algorithm is implemented with the actor-critic(AC)structure,and two linear-in-parameter neural networks are utilized to approximate the functions.Simulation results validate the effectiveness of the data-driven method.展开更多
Network security problems bring many imperceptible threats to the integrity of data and the reliability of device services,so proposing a network intrusion detection model with high reliability is of great research si...Network security problems bring many imperceptible threats to the integrity of data and the reliability of device services,so proposing a network intrusion detection model with high reliability is of great research significance for network security.Due to the strong generalization of invalid features during training process,it is more difficult for single autoencoder intrusion detection model to obtain effective results.A network intrusion detection model based on the Ensemble of Denoising Adversarial Autoencoder(EDAAE)was proposed,which had higher accuracy and reliability compared to the traditional anomaly detection model.Using the adversarial learning idea of Adversarial Autoencoder(AAE),the discriminator module was added to the original model,and the encoder part was used as the generator.The distribution of the hidden space of the data generated by the encoder matched with the distribution of the original data.The generalization of the model to the invalid features was also reduced to improve the detection accuracy.At the same time,the denoising autoencoder and integrated operation was introduced to prevent overfitting in the adversarial learning process.Experiments on the CICIDS2018 traffic dataset showed that the proposed intrusion detection model achieves an Accuracy of 95.23%,which out performs traditional self-encoders and other existing intrusion detection models methods in terms of overall performance.展开更多
It is difficult for the double suppression division algorithm of bee colony to solve the spatio-temporal coupling or have higher dimensional attributes and undertake sudden tasks.Using the idea of clustering,after clu...It is difficult for the double suppression division algorithm of bee colony to solve the spatio-temporal coupling or have higher dimensional attributes and undertake sudden tasks.Using the idea of clustering,after clustering tasks according to spatio-temporal attributes,the clustered groups are linked into task sub-chains according to similarity.Then,based on the correlation between clusters,the child chains are connected to form a task chain.Therefore,the limitation is solved that the task chain in the bee colony algorithm can only be connected according to one dimension.When a sudden task occurs,a method of inserting a small number of tasks into the original task chain and a task chain reconstruction method are designed according to the relative relationship between the number of sudden tasks and the number of remaining tasks.Through the above improvements,the algorithm can be used to process tasks with spatio-temporal coupling and burst tasks.In order to reflect the efficiency and applicability of the algorithm,a task allocation model for the unmanned aerial vehicle(UAV)group is constructed,and a one-to-one correspondence between the improved bee colony double suppression division algorithm and each attribute in the UAV group is proposed.Task assignment has been constructed.The study uses the self-adjusting characteristics of the bee colony to achieve task allocation.Simulation verification and algorithm comparison show that the algorithm has stronger planning advantages and algorithm performance.展开更多
With limited number of labeled samples,hyperspectral image(HSI)classification is a difficult Problem in current research.The graph neural network(GNN)has emerged as an approach to semi-supervised classification,and th...With limited number of labeled samples,hyperspectral image(HSI)classification is a difficult Problem in current research.The graph neural network(GNN)has emerged as an approach to semi-supervised classification,and the application of GNN to hyperspectral images has attracted much attention.However,in the existing GNN-based methods a single graph neural network or graph filter is mainly used to extract HSI features,which does not take full advantage of various graph neural networks(graph filters).Moreover,the traditional GNNs have the problem of oversmoothing.To alleviate these shortcomings,we introduce a deep hybrid multi-graph neural network(DHMG),where two different graph filters,i.e.,the spectral filter and the autoregressive moving average(ARMA)filter,are utilized in two branches.The former can well extract the spectral features of the nodes,and the latter has a good suppression effect on graph noise.The network realizes information interaction between the two branches and takes good advantage of different graph filters.In addition,to address the problem of oversmoothing,a dense network is proposed,where the local graph features are preserved.The dense structure satisfies the needs of different classification targets presenting different features.Finally,we introduce a GraphSAGEbased network to refine the graph features produced by the deep hybrid network.Extensive experiments on three public HSI datasets strongly demonstrate that the DHMG dramatically outperforms the state-ofthe-art models.展开更多
The inertial navigation system(INS),which is frequently used in emergency rescue operations and other situations,has the benefits of not relying on infrastructure,high positioning frequency,and strong real-time perfor...The inertial navigation system(INS),which is frequently used in emergency rescue operations and other situations,has the benefits of not relying on infrastructure,high positioning frequency,and strong real-time performance.However,the intricate and unpredictable pedestrian motion patterns lead the INS localization error to significantly diverge with time.This paper aims to enhance the accuracy of zero-velocity interval(ZVI)detection and reduce the heading and altitude drift of foot-mounted INS via deep learning and equation constraint of dual feet.Aiming at the observational noise problem of low-cost inertial sensors,we utilize a denoising autoencoder to automatically eliminate the inherent noise.Aiming at the problem that inaccurate detection of the ZVI detection results in obvious displacement error,we propose a sample-level ZVI detection algorithm based on the U-Net neural network,which effectively solves the problem of mislabeling caused by sliding windows.Aiming at the problem that Zero-Velocity Update(ZUPT)cannot suppress heading and altitude error,we propose a bipedal INS method based on the equation constraint and ellipsoid constraint,which uses foot-to-foot distance as a new observation to correct heading and altitude error.We conduct extensive and well-designed experiments to evaluate the performance of the proposed method.The experimental results indicate that the position error of our proposed method did not exceed 0.83% of the total traveled distance.展开更多
In a cloud-native era,the Kubernetes-based workflow engine enables workflow containerized execution through the inherent abilities of Kubernetes.However,when encountering continuous workflow requests and unexpected re...In a cloud-native era,the Kubernetes-based workflow engine enables workflow containerized execution through the inherent abilities of Kubernetes.However,when encountering continuous workflow requests and unexpected resource request spikes,the engine is limited to the current workflow load information for resource allocation,which lacks the agility and predictability of resource allocation,resulting in over and underprovisioning resources.This mechanism seriously hinders workflow execution efficiency and leads to high resource waste.To overcome these drawbacks,we propose an adaptive resource allocation scheme named adaptive resource allocation scheme(ARAS)for the Kubernetes-based workflow engines.Considering potential future workflow task requests within the current task pod’s lifecycle,the ARAS uses a resource scaling strategy to allocate resources in response to high-concurrency workflow scenarios.The ARAS offers resource discovery,resource evaluation,and allocation functionalities and serves as a key component for our tailored workflow engine(KubeAdaptor).By integrating the ARAS into KubeAdaptor for workflow containerized execution,we demonstrate the practical abilities of KubeAdaptor and the advantages of our ARAS.Compared with the baseline algorithm,experimental evaluation under three distinct workflow arrival patterns shows that ARAS gains time-saving of 9.8% to 40.92% in the average total duration of all workflows,time-saving of 26.4% to 79.86% in the average duration of individual workflow,and an increase of 1% to 16% in centrol processing unit(CPU)and memory resource usage rate.展开更多
In challenging situations,such as low illumination,rain,and background clutter,the stability of the thermal infrared(TIR)spectrum can help red,green,blue(RGB)visible spectrum to improve tracking performance.However,th...In challenging situations,such as low illumination,rain,and background clutter,the stability of the thermal infrared(TIR)spectrum can help red,green,blue(RGB)visible spectrum to improve tracking performance.However,the high-level image information and the modality-specific features have not been sufficiently studied.The proposed correlation filter uses the fused saliency content map to improve filter training and extracts different features of modalities.The fused content map is intro-duced into the spatial regularization term of correlation filter to highlight the training samples in the content region.Furthermore,the fused content map can avoid the incompleteness of the con-tent region caused by challenging situations.Additionally,differ-ent features are extracted according to the modality characteris-tics and are fused by the designed response-level fusion stra-tegy.The alternating direction method of multipliers(ADMM)algorithm is used to solve the tracker training efficiently.Experi-ments on the large-scale benchmark datasets show the effec-tiveness of the proposed tracker compared to the state-of-the-art traditional trackers and the deep learning based trackers.展开更多
基金Natural Science Foundation of Gansu Province(23JRRA866)Higher Education Innovation Fund of Gansu Provincial Department of Education(2025A-132)+1 种基金University-level Scientific Research and Innovation Project of Gansu University of Political Science and Law(GZF2024XQN16)Youth Foundation of Lanzhou Jiaotong University(2023023)。
文摘We show that the torsion module Tor_(j)^(R)(R/a,H_(a)^(i)(X))is in a Serre subcategory for the bounded below R-complex X.In addition,we prove the isomorphism Tor_(s-t)^(R)(R/a,X)≅Tor_(s)^(R)(R/a,H_(a)^(t)(X))in some case.As an application,the Betti number of a complex X in a prime ideal p can be computed by the Betti number of the local cohomology modules of X in p.
基金Projects(51975006, 51505006) supported by the National Natural Science Foundation of China。
文摘This experiment obtained different laser energy density(LED) by changing SLM molding process parameters.The surface morphology, surface quality, and microstructure of as-fabricated samples were studied. The effects of scanning speed, hatching space, and laser power on surface quality were analyzed, and the optimal LED range for surface quality was determined. The results show that pores and spherical particles appear on the sample’s surface when low LED is applied, while there are lamellar structures on the sides of the samples. Cracks appear on the sample’s surface,and the splash phenomenon increases when a high LED is taken. At the same time, a large amount of unmelted powder adhered to the side of the sample. The surface quality is the best when the LED is 150-170 J/mm^(3). The preferred hatch space is currently 0.05-0.09 mm, the laser power is 200-350 W, and the average surface roughness value is(15.1±3) μm.The average surface hardness reaches HV404±HV3, higher than the forging standard range of HV340-HV395.Increasing the LED within the experiment range can increase the surface hardness, yet an excessively high LED will not further increase the surface hardness. The microstructure is composed of needle-like α’-phases with a length of about 20μm, in a crisscross ‘N’ shape, when the LED is low. The β-phase grain boundary is not obvious, and the secondaryphase volume fraction is high;when the LED is high, the α’-phase of the microstructure is in the form of coarse slats, and the secondary-phase is composed of a small amount of secondary α’-phase, the tertiary α’-phase and the fourth α’-phase disappear, and the volume fraction of the secondary-phase becomes low.
基金supported by the National Natural Science Foundation of China(71571076)the National Key R&D Program for the 13th-Five-Year-Plan of China(2018YFF0300301).
文摘The multi-compartment electric vehicle routing problem(EVRP)with soft time window and multiple charging types(MCEVRP-STW&MCT)is studied,in which electric multi-compartment vehicles that are environmentally friendly but need to be recharged in course of transport process,are employed.A mathematical model for this optimization problem is established with the objective of minimizing the function composed of vehicle cost,distribution cost,time window penalty cost and charging service cost.To solve the problem,an estimation of the distribution algorithm based on Lévy flight(EDA-LF)is proposed to perform a local search at each iteration to prevent the algorithm from falling into local optimum.Experimental results demonstrate that the EDA-LF algorithm can find better solutions and has stronger robustness than the basic EDA algorithm.In addition,when comparing with existing algorithms,the result shows that the EDA-LF can often get better solutions in a relatively short time when solving medium and large-scale instances.Further experiments show that using electric multi-compartment vehicles to deliver incompatible products can produce better results than using traditional fuel vehicles.
基金Project(61702063)supported by the National Natural Science Foundation of China。
文摘With the warming up and continuous development of machine learning,especially deep learning,the research on visual question answering field has made significant progress,with important theoretical research significance and practical application value.Therefore,it is necessary to summarize the current research and provide some reference for researchers in this field.This article conducted a detailed and in-depth analysis and summarized of relevant research and typical methods of visual question answering field.First,relevant background knowledge about VQA(Visual Question Answering)was introduced.Secondly,the issues and challenges of visual question answering were discussed,and at the same time,some promising discussion on the particular methodologies was given.Thirdly,the key sub-problems affecting visual question answering were summarized and analyzed.Then,the current commonly used data sets and evaluation indicators were summarized.Next,in view of the popular algorithms and models in VQA research,comparison of the algorithms and models was summarized and listed.Finally,the future development trend and conclusion of visual question answering were prospected.
基金supported by the National Key R&D Program of China(2018AAA0101400)the Natural Science Foundation of Jiangsu Province of China(BK20202006)the National Natural Science Foundation of China(61921004,62173251).
文摘In this paper,the optimal control of non-linear switching system is investigated without knowing the system dynamics.First,the Hamilton-Jacobi-Bellman(HJB)equation is derived with the consideration of hybrid action space.Then,a novel data-based hybrid Q-learning(HQL)algorithm is proposed to find the optimal solution in an iterative manner.In addition,the theoretical analysis is provided to illustrate the convergence and optimality of the proposed algorithm.Finally,the algorithm is implemented with the actor-critic(AC)structure,and two linear-in-parameter neural networks are utilized to approximate the functions.Simulation results validate the effectiveness of the data-driven method.
文摘Network security problems bring many imperceptible threats to the integrity of data and the reliability of device services,so proposing a network intrusion detection model with high reliability is of great research significance for network security.Due to the strong generalization of invalid features during training process,it is more difficult for single autoencoder intrusion detection model to obtain effective results.A network intrusion detection model based on the Ensemble of Denoising Adversarial Autoencoder(EDAAE)was proposed,which had higher accuracy and reliability compared to the traditional anomaly detection model.Using the adversarial learning idea of Adversarial Autoencoder(AAE),the discriminator module was added to the original model,and the encoder part was used as the generator.The distribution of the hidden space of the data generated by the encoder matched with the distribution of the original data.The generalization of the model to the invalid features was also reduced to improve the detection accuracy.At the same time,the denoising autoencoder and integrated operation was introduced to prevent overfitting in the adversarial learning process.Experiments on the CICIDS2018 traffic dataset showed that the proposed intrusion detection model achieves an Accuracy of 95.23%,which out performs traditional self-encoders and other existing intrusion detection models methods in terms of overall performance.
基金This work was supported by the National Natural Science and Technology Innovation 2030 Major Project of Ministry of Science and Technology of China(2018AAA0101200)the National Natural Science Foundation of China(61502522,61502534)+4 种基金the Equipment Pre-Research Field Fund(JZX7Y20190253036101)the Equipment Pre-Research Ministry of Education Joint Fund(6141A02033703)Shaanxi Provincial Natural Science Foundation(2020JQ-493)the Military Science Project of the National Social Science Fund(WJ2019-SKJJ-C-092)the Theoretical Research Foundation of Armed Police Engineering University(WJY202148).
文摘It is difficult for the double suppression division algorithm of bee colony to solve the spatio-temporal coupling or have higher dimensional attributes and undertake sudden tasks.Using the idea of clustering,after clustering tasks according to spatio-temporal attributes,the clustered groups are linked into task sub-chains according to similarity.Then,based on the correlation between clusters,the child chains are connected to form a task chain.Therefore,the limitation is solved that the task chain in the bee colony algorithm can only be connected according to one dimension.When a sudden task occurs,a method of inserting a small number of tasks into the original task chain and a task chain reconstruction method are designed according to the relative relationship between the number of sudden tasks and the number of remaining tasks.Through the above improvements,the algorithm can be used to process tasks with spatio-temporal coupling and burst tasks.In order to reflect the efficiency and applicability of the algorithm,a task allocation model for the unmanned aerial vehicle(UAV)group is constructed,and a one-to-one correspondence between the improved bee colony double suppression division algorithm and each attribute in the UAV group is proposed.Task assignment has been constructed.The study uses the self-adjusting characteristics of the bee colony to achieve task allocation.Simulation verification and algorithm comparison show that the algorithm has stronger planning advantages and algorithm performance.
文摘With limited number of labeled samples,hyperspectral image(HSI)classification is a difficult Problem in current research.The graph neural network(GNN)has emerged as an approach to semi-supervised classification,and the application of GNN to hyperspectral images has attracted much attention.However,in the existing GNN-based methods a single graph neural network or graph filter is mainly used to extract HSI features,which does not take full advantage of various graph neural networks(graph filters).Moreover,the traditional GNNs have the problem of oversmoothing.To alleviate these shortcomings,we introduce a deep hybrid multi-graph neural network(DHMG),where two different graph filters,i.e.,the spectral filter and the autoregressive moving average(ARMA)filter,are utilized in two branches.The former can well extract the spectral features of the nodes,and the latter has a good suppression effect on graph noise.The network realizes information interaction between the two branches and takes good advantage of different graph filters.In addition,to address the problem of oversmoothing,a dense network is proposed,where the local graph features are preserved.The dense structure satisfies the needs of different classification targets presenting different features.Finally,we introduce a GraphSAGEbased network to refine the graph features produced by the deep hybrid network.Extensive experiments on three public HSI datasets strongly demonstrate that the DHMG dramatically outperforms the state-ofthe-art models.
基金supported in part by National Key Research and Development Program under Grant No.2020YFB1708800China Postdoctoral Science Foundation under Grant No.2021M700385+5 种基金Guang Dong Basic and Applied Basic Research Foundation under Grant No.2021A1515110577Guangdong Key Research and Development Program under Grant No.2020B0101130007Central Guidance on Local Science and Technology Development Fund of Shanxi Province under Grant No.YDZJSX2022B019Fundamental Research Funds for Central Universities under Grant No.FRF-MP-20-37Interdisciplinary Research Project for Young Teachers of USTB(Fundamental Research Funds for the Central Universities)under Grant No.FRF-IDRY-21-005National Natural Science Foundation of China under Grant No.62002026。
文摘The inertial navigation system(INS),which is frequently used in emergency rescue operations and other situations,has the benefits of not relying on infrastructure,high positioning frequency,and strong real-time performance.However,the intricate and unpredictable pedestrian motion patterns lead the INS localization error to significantly diverge with time.This paper aims to enhance the accuracy of zero-velocity interval(ZVI)detection and reduce the heading and altitude drift of foot-mounted INS via deep learning and equation constraint of dual feet.Aiming at the observational noise problem of low-cost inertial sensors,we utilize a denoising autoencoder to automatically eliminate the inherent noise.Aiming at the problem that inaccurate detection of the ZVI detection results in obvious displacement error,we propose a sample-level ZVI detection algorithm based on the U-Net neural network,which effectively solves the problem of mislabeling caused by sliding windows.Aiming at the problem that Zero-Velocity Update(ZUPT)cannot suppress heading and altitude error,we propose a bipedal INS method based on the equation constraint and ellipsoid constraint,which uses foot-to-foot distance as a new observation to correct heading and altitude error.We conduct extensive and well-designed experiments to evaluate the performance of the proposed method.The experimental results indicate that the position error of our proposed method did not exceed 0.83% of the total traveled distance.
基金supported by the National Natural Science Foundation of China(61873030,62002019).
文摘In a cloud-native era,the Kubernetes-based workflow engine enables workflow containerized execution through the inherent abilities of Kubernetes.However,when encountering continuous workflow requests and unexpected resource request spikes,the engine is limited to the current workflow load information for resource allocation,which lacks the agility and predictability of resource allocation,resulting in over and underprovisioning resources.This mechanism seriously hinders workflow execution efficiency and leads to high resource waste.To overcome these drawbacks,we propose an adaptive resource allocation scheme named adaptive resource allocation scheme(ARAS)for the Kubernetes-based workflow engines.Considering potential future workflow task requests within the current task pod’s lifecycle,the ARAS uses a resource scaling strategy to allocate resources in response to high-concurrency workflow scenarios.The ARAS offers resource discovery,resource evaluation,and allocation functionalities and serves as a key component for our tailored workflow engine(KubeAdaptor).By integrating the ARAS into KubeAdaptor for workflow containerized execution,we demonstrate the practical abilities of KubeAdaptor and the advantages of our ARAS.Compared with the baseline algorithm,experimental evaluation under three distinct workflow arrival patterns shows that ARAS gains time-saving of 9.8% to 40.92% in the average total duration of all workflows,time-saving of 26.4% to 79.86% in the average duration of individual workflow,and an increase of 1% to 16% in centrol processing unit(CPU)and memory resource usage rate.
基金supported by the National Natural Science Foundation of China(62073036,62076031)Beijing Natural Science Foundation(4242049).
文摘In challenging situations,such as low illumination,rain,and background clutter,the stability of the thermal infrared(TIR)spectrum can help red,green,blue(RGB)visible spectrum to improve tracking performance.However,the high-level image information and the modality-specific features have not been sufficiently studied.The proposed correlation filter uses the fused saliency content map to improve filter training and extracts different features of modalities.The fused content map is intro-duced into the spatial regularization term of correlation filter to highlight the training samples in the content region.Furthermore,the fused content map can avoid the incompleteness of the con-tent region caused by challenging situations.Additionally,differ-ent features are extracted according to the modality characteris-tics and are fused by the designed response-level fusion stra-tegy.The alternating direction method of multipliers(ADMM)algorithm is used to solve the tracker training efficiently.Experi-ments on the large-scale benchmark datasets show the effec-tiveness of the proposed tracker compared to the state-of-the-art traditional trackers and the deep learning based trackers.
