Two-Layer Attention Feature Pyramid Network for Small Object Detection

Effective small object detection is crucial in various applications including urban intelligent transportation and pedestrian detection.However,small objects are difficult to detect accurately because they contain less information.Many current methods,particularly those based on Feature Pyramid Network(FPN),address this challenge by leveraging multi-scale feature fusion.However,existing FPN-based methods often suffer from inadequate feature fusion due to varying resolutions across different layers,leading to suboptimal small object detection.To address this problem,we propose the Two-layerAttention Feature Pyramid Network(TA-FPN),featuring two key modules:the Two-layer Attention Module(TAM)and the Small Object Detail Enhancement Module(SODEM).TAM uses the attention module to make the network more focused on the semantic information of the object and fuse it to the lower layer,so that each layer contains similar semantic information,to alleviate the problem of small object information being submerged due to semantic gaps between different layers.At the same time,SODEM is introduced to strengthen the local features of the object,suppress background noise,enhance the information details of the small object,and fuse the enhanced features to other feature layers to ensure that each layer is rich in small object information,to improve small object detection accuracy.Our extensive experiments on challenging datasets such as Microsoft Common Objects inContext(MSCOCO)and Pattern Analysis Statistical Modelling and Computational Learning,Visual Object Classes(PASCAL VOC)demonstrate the validity of the proposedmethod.Experimental results show a significant improvement in small object detection accuracy compared to state-of-theart detectors.

3D Road Network Modeling and Road Structure Recognition in Internet of Vehicles

Internet of Vehicles (IoV) is a new system that enables individual vehicles to connect with nearby vehicles,people, transportation infrastructure, and networks, thereby realizing amore intelligent and efficient transportationsystem. The movement of vehicles and the three-dimensional (3D) nature of the road network cause the topologicalstructure of IoV to have the high space and time complexity.Network modeling and structure recognition for 3Droads can benefit the description of topological changes for IoV. This paper proposes a 3Dgeneral roadmodel basedon discrete points of roads obtained from GIS. First, the constraints imposed by 3D roads on moving vehicles areanalyzed. Then the effects of road curvature radius (Ra), longitudinal slope (Slo), and length (Len) on speed andacceleration are studied. Finally, a general 3D road network model based on road section features is established.This paper also presents intersection and road section recognition methods based on the structural features ofthe 3D road network model and the road features. Real GIS data from a specific region of Beijing is adopted tocreate the simulation scenario, and the simulation results validate the general 3D road network model and therecognitionmethod. Therefore, thiswork makes contributions to the field of intelligent transportation by providinga comprehensive approach tomodeling the 3Droad network and its topological changes in achieving efficient trafficflowand improved road safety.

Graph Convolutional Networks Embedding Textual Structure Information for Relation Extraction

Deep neural network-based relational extraction research has made significant progress in recent years,andit provides data support for many natural language processing downstream tasks such as building knowledgegraph,sentiment analysis and question-answering systems.However,previous studies ignored much unusedstructural information in sentences that could enhance the performance of the relation extraction task.Moreover,most existing dependency-based models utilize self-attention to distinguish the importance of context,whichhardly deals withmultiple-structure information.To efficiently leverage multiple structure information,this paperproposes a dynamic structure attention mechanism model based on textual structure information,which deeplyintegrates word embedding,named entity recognition labels,part of speech,dependency tree and dependency typeinto a graph convolutional network.Specifically,our model extracts text features of different structures from theinput sentence.Textual Structure information Graph Convolutional Networks employs the dynamic structureattention mechanism to learn multi-structure attention,effectively distinguishing important contextual features invarious structural information.In addition,multi-structure weights are carefully designed as amergingmechanismin the different structure attention to dynamically adjust the final attention.This paper combines these featuresand trains a graph convolutional network for relation extraction.We experiment on supervised relation extractiondatasets including SemEval 2010 Task 8,TACRED,TACREV,and Re-TACED,the result significantly outperformsthe previous.

Analysis of Urban Agglomeration Network Structure Based on Baidu Migration Data: A Case Study of the Guangdong-Hong Kong-Macao Greater Bay Urban Agglomeration

The inter-city linkage heat data provided by Baidu Migration is employed as a characterization of inter-city linkages in order to facilitate the study of the network linkage characteristics and hierarchical structure of urban agglomeration in the Greater Bay Area through the use of social network analysis method.This is the inaugural application of big data based on location services in the study of urban agglomeration network structure,which represents a novel research perspective on this topic.The study reveals that the density of network linkages in the Greater Bay Area urban agglomeration has reached 100%,indicating a mature network-like spatial structure.This structure has given rise to three distinct communities:Shenzhen-Dongguan-Huizhou,Guangzhou-Foshan-Zhaoqing,and Zhuhai-Zhongshan-Jiangmen.Additionally,cities within the Greater Bay Area urban agglomeration play different roles,suggesting that varying development strategies may be necessary to achieve staggered development.The study demonstrates that large datasets represented by LBS can offer novel insights and methodologies for the examination of urban agglomeration network structures,contingent on the appropriate mining and processing of the data.

Deep learning to estimate ocean subsurface salinity structure in the Indian Ocean using satellite observations

Accurately estimating the ocean subsurface salinity structure(OSSS)is crucial for understanding ocean dynamics and predicting climate variations.We present a convolutional neural network(CNN)model to estimate the OSSS in the Indian Ocean using satellite data and Argo observations.We evaluated the performance of the CNN model in terms of its vertical and spatial distribution,as well as seasonal variation of OSSS estimation.Results demonstrate that the CNN model accurately estimates the most significant salinity features in the Indian Ocean using sea surface data with no significant differences from Argo-derived OSSS.However,the estimation accuracy of the CNN model varies with depth,with the most challenging depth being approximately 70 m,corresponding to the halocline layer.Validations of the CNN model’s accuracy in estimating OSSS in the Indian Ocean are also conducted by comparing Argo observations and CNN model estimations along two selected sections and four selected boxes.The results show that the CNN model effectively captures the seasonal variability of salinity,demonstrating its high performance in salinity estimation using sea surface data.Our analysis reveals that sea surface salinity has the strongest correlation with OSSS in shallow layers,while sea surface height anomaly plays a more significant role in deeper layers.These preliminary results provide valuable insights into the feasibility of estimating OSSS using satellite observations and have implications for studying upper ocean dynamics using machine learning techniques.

Deformation,structure and potential hazard of a landslide based on InSAR in Banbar county,Xizang(Tibet)

The Tibetan Plateau is characterized by complex geological conditions and a relatively fragile ecological environment.In recent years,there has been continuous development and increased human activity in the Tibetan Plateau region,leading to a rising risk of landslides.The landslide in Banbar County,Xizang(Tibet),have been perturbed by ongoing disturbances from human engineering activities,making it susceptible to instability and displaying distinct features.In this study,small baseline subset synthetic aperture radar interferometry(SBAS-InSAR)technology is used to obtain the Line of Sight(LOS)deformation velocity field in the study area,and then the slope-orientation deformation field of the landslide is obtained according to the spatial geometric relationship between the satellite’s LOS direction and the landslide.Subsequently,the landslide thickness is inverted by applying the mass conservation criterion.The results show that the movement area of the landslide is about 6.57×10^(4)m^(2),and the landslide volume is about 1.45×10^(6)m^(3).The maximum estimated thickness and average thickness of the landslide are 39 m and 22 m,respectively.The thickness estimation results align with the findings from on-site investigation,indicating the applicability of this method to large-scale earth slides.The deformation rate of the landslide exhibits a notable correlation with temperature variations,with rainfall playing a supportive role in the deformation process and displaying a certain lag.Human activities exert the most substantial influence on the spatial heterogeneity of landslide deformation,leading to the direct impact of several prominent deformation areas due to human interventions.Simultaneously,utilizing the long short-term memory(LSTM)model to predict landslide displacement,and the forecast results demonstrate the effectiveness of the LSTM model in predicting landslides that are in a continuous development and movement phase.The landslide is still active,and based on the spatial heterogeneity of landslide deformation,new recommendations have been proposed for the future management of the landslide in order to mitigate potential hazards associated with landslide instability.

Influence of Al_(2)O_(3)/SiO_(2) Ratio on the Structure and Properties of Na^(+)/K^(+)Ion Exchange Na_(2)O-MgO-Al_(2)O_(3)-SiO_(2) Glasses

In this work,the structure,viscosity and ion-exchange process of Na_(2)O-MgO-Al_(2)O_(3)-SiO_(2) glasses with different Al_(2)O_(3)/SiO_(2) molar ratios were investigated.The results showed that,with increasing Al_(2)O_(3)/SiO_(2) ratio,the simple structural units Q_(1) and Q_(2) transformed into highly aggregated structural units Q_(3) and Q_(4),indicating the increase of polymerization degree of glass network.Meanwhile,the coefficient of thermal expansion decreased from 9.23×10^(-6)℃^(-1) to 8.88×10^(-6)℃^(-1).The characteristic temperatures such as melting,forming,softening and glass transition temperatures increased with the increase of Al_(2)O_(3)/SiO_(2) ratio,while the glasses working temperature range became narrow.The increasing Al_(2)O_(3)/SiO_(2) ratio and prolonging ion-exchange time enhanced the surface compressive stress(CS)and depth of stress layer(DOL).However,the increase of ion exchange temperature increased the DOL and decreased the CS affected by stress relaxation.There was a good linear relationship between stress relaxation and surface compressive stress.Chemical strengthening significantly improved the hardness of glasses,which reached the maximum value of(622.1±10)MPa for sample with Al_(2)O_(3)/SiO_(2) ratio of 0.27 after heat treated at 410℃for 2 h.

Effect of Fe_(2)O_(3)on the Structure,Physical Properties and Crystallization of CaO-Al_(2)O_(3)-SiO_(2)Glass

The calcium aluminosilicate-based glasses(CaO-Al_(2)O_(3)-SiO_(2),CAS)with different Fe_(2)O_(3)content(0.10wt%,0.50wt%,0.90wt%,and 1.30wt%)were prepared by traditional melt-quenching method.The glass network structure,thermal and mechanical properties,and crystallization behavior changes were investigated by nuclear magnetic resonance spectrometer,Fourier-transform infrared spectro-photometer,X-ray diffractometer,differential scanning calorimetry and field emission scanning electron microscope measurements.The change of Q^(n)in glass structures reveals the glass network connectivity decreases due to the increasing content of Fe_(2)O_(3)addition,resulting in the increasing of non-bridging number in glass structure.The glass densities slightly rise from 2.644 to 2.681 g/cm^(3),while Vickers’s hardness increases at first,from 6.469 to 6.901 GPa,then slightly drops to 6.745 GPa,with Fe_(2)O_(3)content increase.There is almost no thermal expansion coefficient change from different Fe_(2)O_(3)content.The glass transmittance in visible range gradually decreases with higher Fe_(2)O_(3)content,resulting from the strong absorption of Fe^(2+)and Fe^(3+)ions.The calculated activation energy from thermal analysis results first decreases from 282.70 to 231.18 kJ/mol,and then increases to 244.02 kJ/mol,with the Fe_(2)O_(3)content increasing from 0.10wt%to 1.30wt%.Meanwhile,the maximum Avrami constant of 2.33 means the CAS glasses exhibit two-dimensional crystallization.All of the CAS glass-ceramics samples contain main crystal phase of anorthite,the microstructure appears lamellar and columnar crystals.

Structural Characteristics and Evolution of a Weighted Sino-US Container Shipping Network

This study selected the Sino-US route data from the top 30 global container liner companies between December 1,2019,and December 29,2019,as the data source utilizing the complex network research methodology.It constructs a Sino-US container shipping network through voyage weighting and analyzes the essential structural characteristics to explore the network’s complex structural fea-tures.The network’s evolution is examined from three perspectives,namely,time,space,and event influence,aiming to comprehens-ively explore the network’s evolution mechanism.The results revealed that:1)the weighted Sino-US container shipping network exhib-its small-world and scale-free properties.Key hub ports in the United States include NEW YORK NY,SAVANNAH GA,LOS ANGELES CA,and OAKLAND CA,whereas SHANGHAI serving as the hub port in China.The geographical distribution of these hub ports is uneven.2)Concerning the evolution of the weighted Sino-US container shipping network,from a temporal perspective,the evolution of the regional structure of the entire Sino-US region and the Inland United States is in a stage of radiative expansion and de-velopment,with a need for further enhancement in competitiveness and development speed.The evolution of the regional structure of southern China and Europe is transitioning from the stage of radiative expansion and development to an advanced equilibrium stage.The shipping development in Northern China,the Western and Eastern United States,and Asia is undergoing significant changes but faces challenges of fierce competition and imbalances.From a spatial perspective,the rationality and effectiveness of the improved weighted Barrat-Barthelemy-Vespignani(BBV)model are confirmed through theoretical derivation.The applicability of the improved evolution model is verified by simulating the evolution of the weighted Sino-US container shipping network.From an event impact per-spective,the Corona Virus Disease 2019(COVID-19)pandemic has not fundamentally affected the spatial pattern of the weighted Sino-US container shipping network but has significantly impacted the network’s connectivity.The network lacks sufficient resilience and stability in emergency situations.3)Based on the analysis of the structural characteristics and evolution of the weighted Sino-US con-tainer shipping network,recommendations for network development are proposed from three aspects:emphasizing the development of hub ports,focusing on the balanced development of the network,and optimizing the layout of Chinese ports.

Unified deep learning model for predicting fundus fluorescein angiography image from fundus structure image

The prediction of fundus fluorescein angiography(FFA)images from fundus structural images is a cutting-edge research topic in ophthalmological image processing.Prediction comprises estimating FFA from fundus camera imaging,single-phase FFA from scanning laser ophthalmoscopy(SLO),and three-phase FFA also from SLO.Although many deep learning models are available,a single model can only perform one or two of these prediction tasks.To accomplish three prediction tasks using a unified method,we propose a unified deep learning model for predicting FFA images from fundus structure images using a supervised generative adversarial network.The three prediction tasks are processed as follows:data preparation,network training under FFA supervision,and FFA image prediction from fundus structure images on a test set.By comparing the FFA images predicted by our model,pix2pix,and CycleGAN,we demonstrate the remarkable progress achieved by our proposal.The high performance of our model is validated in terms of the peak signal-to-noise ratio,structural similarity index,and mean squared error.

Structural and functional connectivity of the whole brain and subnetworks in individuals with mild traumatic brain injury:predictors of patient prognosis

Patients with mild traumatic brain injury have a diverse clinical presentation,and the underlying pathophysiology remains poorly understood.Magnetic resonance imaging is a non-invasive technique that has been widely utilized to investigate neuro biological markers after mild traumatic brain injury.This approach has emerged as a promising tool for investigating the pathogenesis of mild traumatic brain injury.G raph theory is a quantitative method of analyzing complex networks that has been widely used to study changes in brain structure and function.However,most previous mild traumatic brain injury studies using graph theory have focused on specific populations,with limited exploration of simultaneous abnormalities in structural and functional connectivity.Given that mild traumatic brain injury is the most common type of traumatic brain injury encounte red in clinical practice,further investigation of the patient characteristics and evolution of structural and functional connectivity is critical.In the present study,we explored whether abnormal structural and functional connectivity in the acute phase could serve as indicators of longitudinal changes in imaging data and cognitive function in patients with mild traumatic brain injury.In this longitudinal study,we enrolled 46 patients with mild traumatic brain injury who were assessed within 2 wee ks of injury,as well as 36 healthy controls.Resting-state functional magnetic resonance imaging and diffusion-weighted imaging data were acquired for graph theoretical network analysis.In the acute phase,patients with mild traumatic brain injury demonstrated reduced structural connectivity in the dorsal attention network.More than 3 months of followup data revealed signs of recovery in structural and functional connectivity,as well as cognitive function,in 22 out of the 46 patients.Furthermore,better cognitive function was associated with more efficient networks.Finally,our data indicated that small-worldness in the acute stage could serve as a predictor of longitudinal changes in connectivity in patients with mild traumatic brain injury.These findings highlight the importance of integrating structural and functional connectivity in unde rstanding the occurrence and evolution of mild traumatic brain injury.Additionally,exploratory analysis based on subnetworks could serve a predictive function in the prognosis of patients with mild traumatic brain injury.

Modeling of multiphase flow in low permeability porous media:Effect of wettability and pore structure properties

Multiphase flow in low permeability porous media is involved in numerous energy and environmental applications.However,a complete description of this process is challenging due to the limited modeling scale and the effects of complex pore structures and wettability.To address this issue,based on the digital rock of low permeability sandstone,a direct numerical simulation is performed considering the interphase drag and boundary slip to clarify the microscopic water-oil displacement process.In addition,a dual-porosity pore network model(PNM)is constructed to obtain the water-oil relative permeability of the sample.The displacement efficiency as a recovery process is assessed under different wetting and pore structure properties.Results show that microscopic displacement mechanisms explain the corresponding macroscopic relative permeability.The injected water breaks through the outlet earlier with a large mass flow,while thick oil films exist in rough hydrophobic surfaces and poorly connected pores.The variation of water-oil relative permeability is significant,and residual oil saturation is high in the oil-wet system.The flooding is extensive,and the residual oil is trapped in complex pore networks for hydrophilic pore surfaces;thus,water relative permeability is lower in the water-wet system.While the displacement efficiency is the worst in mixed-wetting systems for poor water connectivity.Microporosity negatively correlates with invading oil volume fraction due to strong capillary resistance,and a large microporosity corresponds to low residual oil saturation.This work provides insights into the water-oil flow from different modeling perspectives and helps to optimize the development plan for enhanced recovery.

Input Structure Design for Structural Controllability of Complex Networks

This paper addresses the problem of the input design of large-scale complex networks.Two types of network components,redundant inaccessible strongly connected component(RISCC)and intermittent inaccessible strongly connected component(IISCC)are defined,and a subnetwork called a driver network is developed.Based on these,an efficient method is proposed to find the minimum number of controlled nodes to achieve structural complete controllability of a network,in the case that each input can act on multiple state nodes.The range of the number of input nodes to achieve minimal control,and the configuration method(the connection between the input nodes and the controlled nodes)are presented.All possible input solutions can be obtained by this method.Moreover,we give an example and some experiments on real-world networks to illustrate the effectiveness of the method.

An aerial ammunition ad hoc network collaborative localization algorithm based on relative ranging and velocity measurement in a highly-dynamic topographic structure

In the internet of battlefield things, ammunition is becoming networked and intelligent, which depends on location information. Therefore, this paper focuses on the self-organized network collaborative localization of munitions with an aerial three-dimensional(3D) highly-dynamic topographic structure under a satellite denied environment. As for aerial networked munitions, the measurement of munitions is objectively incomplete due to the degenerated and interrupted link of munitions. For this reason, a cluster-oriented collaborative localization method is put forward in this paper. Multidimensional scaling(MDS) was first integrated with a trilateration localization method(TLM) to construct a relative localization algorithm for determining the relative location of a mobile cluster network. The information related to relative velocity was then combined into a collaborative localization framework to devise a TLM-vMDS algorithm. Finally, an iterative refinement algorithm based on scaling by majorizing a complicated function(SMACOF) was employed to effectively eliminate the influence of incomplete link observation on localization accuracy. Compared with the currently available advanced algorithms, the proposed TLM-vMDS algorithm achieves higher localization accuracy and faster convergence for a cluster of extensively networked munitions, and also offers better numerical stability and robustness for highspeed motion models.

DHSEGATs:distance and hop-wise structures encoding enhanced graph attention networks

Numerous works prove that existing neighbor-averaging graph neural networks(GNNs)cannot efficiently catch structure features,and many works show that injecting structure,distance,position,or spatial features can significantly improve the performance of GNNs,however,injecting high-level structure and distance into GNNs is an intuitive but untouched idea.This work sheds light on this issue and proposes a scheme to enhance graph attention networks(GATs)by encoding distance and hop-wise structure statistics.Firstly,the hop-wise structure and distributional distance information are extracted based on several hop-wise ego-nets of every target node.Secondly,the derived structure information,distance information,and intrinsic features are encoded into the same vector space and then added together to get initial embedding vectors.Thirdly,the derived embedding vectors are fed into GATs,such as GAT and adaptive graph diffusion network(AGDN)to get the soft labels.Fourthly,the soft labels are fed into correct and smooth(C&S)to conduct label propagation and get final predictions.Experiments show that the distance and hop-wise structures encoding enhanced graph attention networks(DHSEGATs)achieve a competitive result.

Big Model Strategy for Bridge Structural Health Monitoring Based on Data-Driven, Adaptive Method and Convolutional Neural Network (CNN) Group

This study introduces an innovative“Big Model”strategy to enhance Bridge Structural Health Monitoring(SHM)using a Convolutional Neural Network(CNN),time-frequency analysis,and fine element analysis.Leveraging ensemble methods,collaborative learning,and distributed computing,the approach effectively manages the complexity and scale of large-scale bridge data.The CNN employs transfer learning,fine-tuning,and continuous monitoring to optimize models for adaptive and accurate structural health assessments,focusing on extracting meaningful features through time-frequency analysis.By integrating Finite Element Analysis,time-frequency analysis,and CNNs,the strategy provides a comprehensive understanding of bridge health.Utilizing diverse sensor data,sophisticated feature extraction,and advanced CNN architecture,the model is optimized through rigorous preprocessing and hyperparameter tuning.This approach significantly enhances the ability to make accurate predictions,monitor structural health,and support proactive maintenance practices,thereby ensuring the safety and longevity of critical infrastructure.

The Study on the Evolution of Urban Spatial Structure in Zhuhai City Based on Spatial Syntax

With the deepening of the Guangdong-Hong Kong-Macao Greater Bay Area strategy and the accelerated integration and development of the east and west sides of the Pearl River Estuary,Zhuhai’s hub position is becoming more and more prominent.The city of Zhuhai has a dense water network and is divided into two urban areas,the east and the west,under the influence of the Mordor Gate waterway.Based on the theory of spatial syntax,this paper carries out an analytical study on the urban spatial structure of Zhuhai,identifies the distribution characteristics of urban POIs,and provides theoretical support for the urban development of Zhuhai.

Network autoregression model with grouped factor structures

Network autoregression and factor model are effective methods for modeling network time series data.In this study,we propose a network autoregression model with a factor structure that incorporates a latent group structure to address nodal heterogeneity within the network.An iterative algorithm is employed to minimize a least-squares objective function,allowing for simultaneous estimation of both the parameters and the group structure.To determine the unknown number of groups and factors,a PIC criterion is introduced.Additionally,statistical inference of the estimated parameters is presented.To assess the validity of the proposed estimation and inference procedures,we conduct extensive numerical studies.We also demonstrate the utility of our model using a stock dataset obtained from the Chinese A-Share stock market.

Cooperative Caching Strategy Based on Two-Layer Caching Model for Remote Sensing Satellite Networks

In Information Centric Networking(ICN)where content is the object of exchange,in-network caching is a unique functional feature with the ability to handle data storage and distribution in remote sensing satellite networks.Setting up cache space at any node enables users to access data nearby,thus relieving the processing pressure on the servers.However,the existing caching strategies still suffer from the lack of global planning of cache contents and low utilization of cache resources due to the lack of fine-grained division of cache contents.To address the issues mentioned,a cooperative caching strategy(CSTL)for remote sensing satellite networks based on a two-layer caching model is proposed.The two-layer caching model is constructed by setting up separate cache spaces in the satellite network and the ground station.Probabilistic caching of popular contents in the region at the ground station to reduce the access delay of users.A content classification method based on hierarchical division is proposed in the satellite network,and differential probabilistic caching is employed for different levels of content.The cached content is also dynamically adjusted by analyzing the subsequent changes in the popularity of the cached content.In the two-layer caching model,ground stations and satellite networks collaboratively cache to achieve global planning of cache contents,rationalize the utilization of cache resources,and reduce the propagation delay of remote sensing data.Simulation results show that the CSTL strategy not only has a high cache hit ratio compared with other caching strategies but also effectively reduces user request delay and server load,which satisfies the timeliness requirement of remote sensing data transmission.

Sensor Network Structure Recognition Based on P-law

A sensor graph network is a sensor network model organized according to graph network structure.Structural unit and signal propagation of core nodes are the basic characteristics of sensor graph networks.In sensor networks,network structure recognition is the basis for accurate identification and effective prediction and control of node states.Aiming at the problems of difficult global structure identification and poor interpretability in complex sensor graph networks,based on the characteristics of sensor networks,a method is proposed to firstly unitize the graph network structure and then expand the unit based on the signal transmission path of the core node.This method which builds on unit patulousness and core node signal propagation(called p-law)can rapidly and effectively achieve the global structure identification of a sensor graph network.Different from the traditional graph network structure recognition algorithms such as modularity maximization and spectral clustering,the proposed method reveals the natural evolution process and law of graph network subgroup generation.Experimental results confirm the effectiveness,accuracy and rationality of the proposed method and suggest that our method can be a new approach for graph network global structure recognition.