Pneumonia is a serious disease that can be fatal,particularly among children and the elderly.The accuracy of pneumonia diagnosis can be improved by combining artificial-intelligence technology with X-ray imaging.This ...Pneumonia is a serious disease that can be fatal,particularly among children and the elderly.The accuracy of pneumonia diagnosis can be improved by combining artificial-intelligence technology with X-ray imaging.This study proposes X-ODFCANet,which addresses the issues of low accuracy and excessive parameters in existing deep-learningbased pneumonia-classification methods.This network incorporates a feature coordination attention module and an omni-dimensional dynamic convolution(ODConv)module,leveraging the residual module for feature extraction from X-ray images.The feature coordination attention module utilizes two one-dimensional feature encoding processes to aggregate feature information from different spatial directions.Additionally,the ODConv module extracts and fuses feature information in four dimensions:the spatial dimension of the convolution kernel,input and output channel quantities,and convolution kernel quantity.The experimental results demonstrate that the proposed method can effectively improve the accuracy of pneumonia classification,which is 3.77%higher than that of ResNet18.The model parameters are 4.45M,which was reduced by approximately 2.5 times.The code is available at https://github.com/limuni/X ODFCA NET.展开更多
The ability to accurately predict urban traffic flows is crucial for optimising city operations.Consequently,various methods for forecasting urban traffic have been developed,focusing on analysing historical data to u...The ability to accurately predict urban traffic flows is crucial for optimising city operations.Consequently,various methods for forecasting urban traffic have been developed,focusing on analysing historical data to understand complex mobility patterns.Deep learning techniques,such as graph neural networks(GNNs),are popular for their ability to capture spatio-temporal dependencies.However,these models often become overly complex due to the large number of hyper-parameters involved.In this study,we introduce Dynamic Multi-Graph Spatial-Temporal Graph Neural Ordinary Differential Equation Networks(DMST-GNODE),a framework based on ordinary differential equations(ODEs)that autonomously discovers effective spatial-temporal graph neural network(STGNN)architectures for traffic prediction tasks.The comparative analysis of DMST-GNODE and baseline models indicates that DMST-GNODE model demonstrates superior performance across multiple datasets,consistently achieving the lowest Root Mean Square Error(RMSE)and Mean Absolute Error(MAE)values,alongside the highest accuracy.On the BKK(Bangkok)dataset,it outperformed other models with an RMSE of 3.3165 and an accuracy of 0.9367 for a 20-min interval,maintaining this trend across 40 and 60 min.Similarly,on the PeMS08 dataset,DMST-GNODE achieved the best performance with an RMSE of 19.4863 and an accuracy of 0.9377 at 20 min,demonstrating its effectiveness over longer periods.The Los_Loop dataset results further emphasise this model’s advantage,with an RMSE of 3.3422 and an accuracy of 0.7643 at 20 min,consistently maintaining superiority across all time intervals.These numerical highlights indicate that DMST-GNODE not only outperforms baseline models but also achieves higher accuracy and lower errors across different time intervals and datasets.展开更多
Traffic flow prediction is a crucial element of intelligent transportation systems.However,accu-rate traffic flow prediction is quite challenging because of its highly nonlinear,complex,and dynam-ic characteristics.To...Traffic flow prediction is a crucial element of intelligent transportation systems.However,accu-rate traffic flow prediction is quite challenging because of its highly nonlinear,complex,and dynam-ic characteristics.To address the difficulties in simultaneously capturing local and global dynamic spatiotemporal correlations in traffic flow,as well as the high time complexity of existing models,a multi-head flow attention-based local-global dynamic hypergraph convolution(MFA-LGDHC)pre-diction model is proposed.which consists of multi-head flow attention(MHFA)mechanism,graph convolution network(GCN),and local-global dynamic hypergraph convolution(LGHC).MHFA is utilized to extract the time dependency of traffic flow and reduce the time complexity of the model.GCN is employed to catch the spatial dependency of traffic flow.LGHC utilizes down-sampling con-volution and isometric convolution to capture the local and global spatial dependencies of traffic flow.And dynamic hypergraph convolution is used to model the dynamic higher-order relationships of the traffic road network.Experimental results indicate that the MFA-LGDHC model outperforms current popular baseline models and exhibits good prediction performance.展开更多
Hyperspectral image(HSI)classification is crucial for numerous remote sensing applications.Traditional deep learning methods may miss pixel relationships and context,leading to inefficiencies.This paper introduces the...Hyperspectral image(HSI)classification is crucial for numerous remote sensing applications.Traditional deep learning methods may miss pixel relationships and context,leading to inefficiencies.This paper introduces the spectral band graph convolutional and attention-enhanced CNN joint network(SGCCN),a novel approach that harnesses the power of spectral band graph convolutions for capturing long-range relationships,utilizes local perception of attention-enhanced multi-level convolutions for local spatial feature and employs a dynamic attention mechanism to enhance feature extraction.The SGCCN integrates spectral and spatial features through a self-attention fusion network,significantly improving classification accuracy and efficiency.The proposed method outperforms existing techniques,demonstrating its effectiveness in handling the challenges associated with HSI data.展开更多
BACKGROUND The accurate classification of focal liver lesions(FLLs)is essential to properly guide treatment options and predict prognosis.Dynamic contrast-enhanced computed tomography(DCE-CT)is still the cornerstone i...BACKGROUND The accurate classification of focal liver lesions(FLLs)is essential to properly guide treatment options and predict prognosis.Dynamic contrast-enhanced computed tomography(DCE-CT)is still the cornerstone in the exact classification of FLLs due to its noninvasive nature,high scanning speed,and high-density resolution.Since their recent development,convolutional neural network-based deep learning techniques has been recognized to have high potential for image recognition tasks.AIM To develop and evaluate an automated multiphase convolutional dense network(MP-CDN)to classify FLLs on multiphase CT.METHODS A total of 517 FLLs scanned on a 320-detector CT scanner using a four-phase DCECT imaging protocol(including precontrast phase,arterial phase,portal venous phase,and delayed phase)from 2012 to 2017 were retrospectively enrolled.FLLs were classified into four categories:Category A,hepatocellular carcinoma(HCC);category B,liver metastases;category C,benign non-inflammatory FLLs including hemangiomas,focal nodular hyperplasias and adenomas;and category D,hepatic abscesses.Each category was split into a training set and test set in an approximate 8:2 ratio.An MP-CDN classifier with a sequential input of the fourphase CT images was developed to automatically classify FLLs.The classification performance of the model was evaluated on the test set;the accuracy and specificity were calculated from the confusion matrix,and the area under the receiver operating characteristic curve(AUC)was calculated from the SoftMax probability outputted from the last layer of the MP-CDN.RESULTS A total of 410 FLLs were used for training and 107 FLLs were used for testing.The mean classification accuracy of the test set was 81.3%(87/107).The accuracy/specificity of distinguishing each category from the others were 0.916/0.964,0.925/0.905,0.860/0.918,and 0.925/0.963 for HCC,metastases,benign non-inflammatory FLLs,and abscesses on the test set,respectively.The AUC(95%confidence interval)for differentiating each category from the others was 0.92(0.837-0.992),0.99(0.967-1.00),0.88(0.795-0.955)and 0.96(0.914-0.996)for HCC,metastases,benign non-inflammatory FLLs,and abscesses on the test set,respectively.CONCLUSION MP-CDN accurately classified FLLs detected on four-phase CT as HCC,metastases,benign non-inflammatory FLLs and hepatic abscesses and may assist radiologists in identifying the different types of FLLs.展开更多
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,...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.展开更多
In this work,a three dimensional(3D)convolutional neural network(CNN)model based on image slices of various normal and pathological vocal folds is proposed for accurate and efficient prediction of glottal flows.The 3D...In this work,a three dimensional(3D)convolutional neural network(CNN)model based on image slices of various normal and pathological vocal folds is proposed for accurate and efficient prediction of glottal flows.The 3D CNN model is composed of the feature extraction block and regression block.The feature extraction block is capable of learning low dimensional features from the high dimensional image data of the glottal shape,and the regression block is employed to flatten the output from the feature extraction block and obtain the desired glottal flow data.The input image data is the condensed set of 2D image slices captured in the axial plane of the 3D vocal folds,where these glottal shapes are synthesized based on the equations of normal vibration modes.The output flow data is the corresponding flow rate,averaged glottal pressure and nodal pressure distributions over the glottal surface.The 3D CNN model is built to establish the mapping between the input image data and output flow data.The ground-truth flow variables of each glottal shape in the training and test datasets are obtained by a high-fidelity sharp-interface immersed-boundary solver.The proposed model is trained to predict the concerned flow variables for glottal shapes in the test set.The present 3D CNN model is more efficient than traditional Computational Fluid Dynamics(CFD)models while the accuracy can still be retained,and more powerful than previous data-driven prediction models because more details of the glottal flow can be provided.The prediction performance of the trained 3D CNN model in accuracy and efficiency indicates that this model could be promising for future clinical applications.展开更多
Classic Graph Convolutional Networks (GCNs) often learn node representation holistically, which ignores the distinct impacts from different neighbors when aggregating their features to update a node’s representation....Classic Graph Convolutional Networks (GCNs) often learn node representation holistically, which ignores the distinct impacts from different neighbors when aggregating their features to update a node’s representation. Disentangled GCNs have been proposed to divide each node’s representation into several feature units. However, current disentangling methods do not try to figure out how many inherent factors the model should assign to help extract the best representation of each node. This paper then proposes D^(2)-GCN to provide dynamic disentanglement in GCNs and present the most appropriate factorization of each node’s mixed features. The convergence of the proposed method is proved both theoretically and experimentally. Experiments on real-world datasets show that D^(2)-GCN outperforms the baseline models concerning node classification results in both single- and multi-label tasks.展开更多
With the rapid advancement of virtual reality,dynamic gesture recognition technology has become an indispensable and critical technique for users to achieve human–computer interaction in virtual environments.The reco...With the rapid advancement of virtual reality,dynamic gesture recognition technology has become an indispensable and critical technique for users to achieve human–computer interaction in virtual environments.The recognition of dynamic gestures is a challenging task due to the high degree of freedom and the influence of individual differences and the change of gesture space.To solve the problem of low recognition accuracy of existing networks,an improved dynamic gesture recognition algorithm based on ResNeXt architecture is proposed.The algorithm employs three-dimensional convolution techniques to effectively capture the spatiotemporal features intrinsic to dynamic gestures.Additionally,to enhance the model’s focus and improve its accuracy in identifying dynamic gestures,a lightweight convolutional attention mechanism is introduced.This mechanism not only augments the model’s precision but also facilitates faster convergence during the training phase.In order to further optimize the performance of the model,a deep attention submodule is added to the convolutional attention mechanism module to strengthen the network’s capability in temporal feature extraction.Empirical evaluations on EgoGesture and NvGesture datasets show that the accuracy of the proposed model in dynamic gesture recognition reaches 95.03%and 86.21%,respectively.When operating in RGB mode,the accuracy reached 93.49%and 80.22%,respectively.These results underscore the effectiveness of the proposed algorithm in recognizing dynamic gestures with high accuracy,showcasing its potential for applications in advanced human–computer interaction systems.展开更多
A reliable transformer protection method is crucial for power systems. Aiming at improving the generalization performance and response speed of multi-feature fusion based transformer protection, this paper presents a ...A reliable transformer protection method is crucial for power systems. Aiming at improving the generalization performance and response speed of multi-feature fusion based transformer protection, this paper presents a dynamic differential current by fusing pre-disturbance and post-disturbance differential currents in real time then developing a dynamic differential current based transformer protection focusing on the feature changes of differential current. Generally, the image of differential current can comprehensively embody the feature changes resulting from any disturbance. In addition, a short window is sometimes sufficient to clearly reflect the internal fault because the differential current will instantly change when an internal fault occurs. Therefore, in order to identify the running states reliably in the shortest possible time, multiple images, including the differential current from a pre-disturbance one cycle to a post-disturbance different time, are combined by time order to define a dynamic differential current. After the protection method is started, this dynamic differential current serves as input for the deep learning algorithm to identify the running states in real time. Once the transformer is identified as a faulty one, a tripping signal is issued and the protection method stops. The dynamic model experiments show that the proposed protection method has a strong generalization ability and rapid response speed.展开更多
Modeling and analysis of complex social networks is an important topic in social computing.Graph convolutional networks(GCNs)are widely used for learning social network embeddings and social network analysis.However,r...Modeling and analysis of complex social networks is an important topic in social computing.Graph convolutional networks(GCNs)are widely used for learning social network embeddings and social network analysis.However,real-world complex social networks,such as Facebook and Math,exhibit significant global structural and dynamic characteristics that are not adequately captured by conventional GCN models.To address the above issues,this paper proposes a novel graph convolutional network considering global structural features and global temporal dependencies(GSTGCN).Specifically,we innovatively design a graph coarsening strategy based on the importance of social membership to construct a dynamic diffusion process of graphs.This dynamic diffusion process can be viewed as using higher-order subgraph embeddings to guide the generation of lower-order subgraph embeddings,and we model this process using gate recurrent unit(GRU)to extract comprehensive global structural features of the graph and the evolutionary processes embedded among subgraphs.Furthermore,we design a new evolutionary strategy that incorporates a temporal self-attention mechanism to enhance the extraction of global temporal dependencies of dynamic networks by GRU.GSTGCN outperforms current state-of-the-art network embedding methods in important social networks tasks such as link prediction and financial fraud identification.展开更多
Multicolor fiber lasers have emerged as a promising technology with significant applications in optical communications,laser ranging,and precision sensing.Beyond their practical utility,these systems serve as ideal pl...Multicolor fiber lasers have emerged as a promising technology with significant applications in optical communications,laser ranging,and precision sensing.Beyond their practical utility,these systems serve as ideal platforms for investigating fundamental soliton phenomena,including soliton collisions,explosions,and state transitions.However,the complex nonlinear dynamics inherent in these systems present substantial challenges for conventional numerical simulations.展开更多
Automatically extracting Drug-Drug Interactions (DDIs) from text is a crucial and challenging task, particularly when multiple medications are taken concurrently. In this study, we propose a novel approach, called Enh...Automatically extracting Drug-Drug Interactions (DDIs) from text is a crucial and challenging task, particularly when multiple medications are taken concurrently. In this study, we propose a novel approach, called Enhanced Attention-driven Dynamic Graph Convolutional Network (E-ADGCN), for DDI extraction. Our model combines the Attention-driven Dynamic Graph Convolutional Network (ADGCN) with a feature fusion method and multi-task learning framework. The ADGCN effectively utilizes entity information and dependency tree information from biomedical texts to extract DDIs. The feature fusion method integrates User-Generated Content (UGC) and molecular information with drug entity information from text through dynamic routing. By leveraging external resources, our approach maximizes the auxiliary effect and improves the accuracy of DDI extraction. We evaluate the E-ADGCN model on the extended DDIExtraction2013 dataset and achieve an F1-score of 81.45%. This research contributes to the advancement of automated methods for extracting valuable drug interaction information from textual sources, facilitating improved medication management and patient safety.展开更多
Impact dynamics of flexible solids is important in engineering practice. Obtaining its dynamic response is a challenging task and usually achieved by numerical methods. The objectives of the study are twofold. Firstly...Impact dynamics of flexible solids is important in engineering practice. Obtaining its dynamic response is a challenging task and usually achieved by numerical methods. The objectives of the study are twofold. Firstly, the discrete singular convolution （DSC） is used for the first time to analyze the impact dynamics. Secondly, the efficiency of various numerical methods for dynamic analysis is explored via an example of a flexible rod hit by a rigid ball. Three numerical methods, including the conventional finite element （FE） method, the DSC algorithm, and the spectral finite element （SFE） method, and one proposed modeling strategy, the improved spectral finite element （ISFE） method, are involved. Numerical results are compared with the known analytical solutions to show their efficiency. It is demonstrated that the proposed ISFE modeling strategy with a proper length of con- ventional FE yields the most accurate contact stress among the four investigated models. It is also found that the DSC algorithm is an alternative method for collision problems.展开更多
Due to self-occlusion and high degree of freedom,estimating 3D hand pose from a single RGB image is a great challenging problem.Graph convolutional networks(GCNs)use graphs to describe the physical connection relation...Due to self-occlusion and high degree of freedom,estimating 3D hand pose from a single RGB image is a great challenging problem.Graph convolutional networks(GCNs)use graphs to describe the physical connection relationships between hand joints and improve the accuracy of 3D hand pose regression.However,GCNs cannot effectively describe the relationships between non-adjacent hand joints.Recently,hypergraph convolutional networks(HGCNs)have received much attention as they can describe multi-dimensional relationships between nodes through hyperedges;therefore,this paper proposes a framework for 3D hand pose estimation based on HGCN,which can better extract correlated relationships between adjacent and non-adjacent hand joints.To overcome the shortcomings of predefined hypergraph structures,a kind of dynamic hypergraph convolutional network is proposed,in which hyperedges are constructed dynamically based on hand joint feature similarity.To better explore the local semantic relationships between nodes,a kind of semantic dynamic hypergraph convolution is proposed.The proposed method is evaluated on publicly available benchmark datasets.Qualitative and quantitative experimental results both show that the proposed HGCN and improved methods for 3D hand pose estimation are better than GCN,and achieve state-of-the-art performance compared with existing methods.展开更多
Reliable traffic flow prediction is crucial for mitigating urban congestion.This paper proposes Attentionbased spatiotemporal Interactive Dynamic Graph Convolutional Network(AIDGCN),a novel architecture integrating In...Reliable traffic flow prediction is crucial for mitigating urban congestion.This paper proposes Attentionbased spatiotemporal Interactive Dynamic Graph Convolutional Network(AIDGCN),a novel architecture integrating Interactive Dynamic Graph Convolution Network(IDGCN)with Temporal Multi-Head Trend-Aware Attention.Its core innovation lies in IDGCN,which uniquely splits sequences into symmetric intervals for interactive feature sharing via dynamic graphs,and a novel attention mechanism incorporating convolutional operations to capture essential local traffic trends—addressing a critical gap in standard attention for continuous data.For 15-and 60-min forecasting on METR-LA,AIDGCN achieves MAEs of 0.75%and 0.39%,and RMSEs of 1.32%and 0.14%,respectively.In the 60-min long-term forecasting of the PEMS-BAY dataset,the AIDGCN out-performs the MRA-BGCN method by 6.28%,4.93%,and 7.17%in terms of MAE,RMSE,and MAPE,respectively.Experimental results demonstrate the superiority of our pro-posed model over state-of-the-art methods.展开更多
Venanico-Filho et al. developed an elegant matrix formulation for dynamic analysis by frequency domain （FD）, but the convergence, causality and extended period need further refining. In the present paper, it was arg...Venanico-Filho et al. developed an elegant matrix formulation for dynamic analysis by frequency domain （FD）, but the convergence, causality and extended period need further refining. In the present paper, it was argued that： （1） under reasonable assumptions （approximating the frequency response function by the discrete Fourier transform of the discretized unitary impulse response function）, the matrix formulation by FD is equivalent to a circular convolution; （2） to avoid the wraparound interference, the excitation vector and impulse response must be padded with enough zeros; （3） provided that the zero padding requirement satisfied, the convergence and accuracy of direct time domain analysis, which is equivalent to that by FD, are guaranteed by the numerical integration scheme; （4） the imaginary part of the computational response approaching zero is due to the continuity of the impulse response functions.展开更多
Latent information is difficult to get from the text in speech synthesis.Studies show that features from speech can get more information to help text encoding.In the field of speech encoding,a lot of work has been con...Latent information is difficult to get from the text in speech synthesis.Studies show that features from speech can get more information to help text encoding.In the field of speech encoding,a lot of work has been conducted on two aspects.The first aspect is to encode speech frame by frame.The second aspect is to encode the whole speech to a vector.But the scale in these aspects is fixed.So,encoding speech with an adjustable scale for more latent information is worthy of investigation.But current alignment approaches only support frame-by-frame encoding and speech-to-vector encoding.It remains a challenge to propose a new alignment approach to support adjustable scale speech encoding.This paper presents the dynamic speech encoder with a new alignment approach in conjunction with frame-by-frame encoding and speech-to-vector encoding.The speech feature fromourmodel achieves three functions.First,the speech feature can reconstruct the origin speech while the length of the speech feature is equal to the text length.Second,our model can get text embedding fromspeech,and the encoded speech feature is similar to the text embedding result.Finally,it can transfer the style of synthesis speech and make it more similar to the given reference speech.展开更多
Low dynamic range(LDR)images captured by consumer cameras have a limited luminance range.As the conventional method for generating high dynamic range(HDR)images involves merging multiple-exposure LDR images of the sam...Low dynamic range(LDR)images captured by consumer cameras have a limited luminance range.As the conventional method for generating high dynamic range(HDR)images involves merging multiple-exposure LDR images of the same scene(assuming a stationary scene),we introduce a learning-based model for single-image HDR reconstruction.An input LDR image is sequentially segmented into the local region maps based on the cumulative histogram of the input brightness distribution.Using the local region maps,SParam-Net estimates the parameters of an inverse tone mapping function to generate a pseudo-HDR image.We process the segmented region maps as the input sequences on long short-term memory.Finally,a fast super-resolution convolutional neural network is used for HDR image reconstruction.The proposed method was trained and tested on datasets including HDR-Real,LDR-HDR-pair,and HDR-Eye.The experimental results revealed that HDR images can be generated more reliably than using contemporary end-to-end approaches.展开更多
Much attention has been given to the Internet of Things (IoT) by citizens, industries, governments, and universities for applications like smart buildings, environmental monitoring, health care and so on. With IoT, ne...Much attention has been given to the Internet of Things (IoT) by citizens, industries, governments, and universities for applications like smart buildings, environmental monitoring, health care and so on. With IoT, networkconnectivity is facilitated between smart devices from anyplace and anytime.IoT-based health monitoring systems are gaining popularity and acceptance forcontinuous monitoring and detect health abnormalities from the data collected.Electrocardiographic (ECG) signals are widely used for heart diseases detection.A novel method has been proposed in this work for ECG monitoring using IoTtechniques. In this work, a two-stage approach is employed. In the first stage, arouting protocol based on Dynamic Source Routing (DSR) and Routing byEnergy and Link quality (REL) for IoT healthcare platform is proposed for effi-cient data collection, and in the second stage, classification of ECG for Arrhythmia. Furthermore, this work has evaluated Support Vector Machine (SVM),Artificial Neural Network (ANN), and Convolution Neural Networks (CNNs)-based approach for ECG signals classification. Deep-ECG will use a deep CNNto extract critical features and then compare through evaluation of simple and fastdistance functions in order to obtain an efficient classification of heart abnormalities. For the identification of abnormal data, this work has proposed techniquesfor the classification of ECG data, which has been obtained from mobile watchusers. For experimental verification of the proposed methods, the Beth Israel Hospital (MIT/BIH) Arrhythmia and Massachusetts Institute of Technology (MIT)Database was used for evaluation. Results confirm the presented method’s superior performance with regards to the accuracy of classification. The CNN achievedan accuracy of 91.92% and has a higher accuracy of 4.98% for the SVM and2.68% for the ANN.展开更多
基金supported in part by the Key Research and Development Program of Shaanxi Province of China,No.2024GX-YBXM-149in part by the National Natural Science Foundation of China,No.62071381.
文摘Pneumonia is a serious disease that can be fatal,particularly among children and the elderly.The accuracy of pneumonia diagnosis can be improved by combining artificial-intelligence technology with X-ray imaging.This study proposes X-ODFCANet,which addresses the issues of low accuracy and excessive parameters in existing deep-learningbased pneumonia-classification methods.This network incorporates a feature coordination attention module and an omni-dimensional dynamic convolution(ODConv)module,leveraging the residual module for feature extraction from X-ray images.The feature coordination attention module utilizes two one-dimensional feature encoding processes to aggregate feature information from different spatial directions.Additionally,the ODConv module extracts and fuses feature information in four dimensions:the spatial dimension of the convolution kernel,input and output channel quantities,and convolution kernel quantity.The experimental results demonstrate that the proposed method can effectively improve the accuracy of pneumonia classification,which is 3.77%higher than that of ResNet18.The model parameters are 4.45M,which was reduced by approximately 2.5 times.The code is available at https://github.com/limuni/X ODFCA NET.
文摘The ability to accurately predict urban traffic flows is crucial for optimising city operations.Consequently,various methods for forecasting urban traffic have been developed,focusing on analysing historical data to understand complex mobility patterns.Deep learning techniques,such as graph neural networks(GNNs),are popular for their ability to capture spatio-temporal dependencies.However,these models often become overly complex due to the large number of hyper-parameters involved.In this study,we introduce Dynamic Multi-Graph Spatial-Temporal Graph Neural Ordinary Differential Equation Networks(DMST-GNODE),a framework based on ordinary differential equations(ODEs)that autonomously discovers effective spatial-temporal graph neural network(STGNN)architectures for traffic prediction tasks.The comparative analysis of DMST-GNODE and baseline models indicates that DMST-GNODE model demonstrates superior performance across multiple datasets,consistently achieving the lowest Root Mean Square Error(RMSE)and Mean Absolute Error(MAE)values,alongside the highest accuracy.On the BKK(Bangkok)dataset,it outperformed other models with an RMSE of 3.3165 and an accuracy of 0.9367 for a 20-min interval,maintaining this trend across 40 and 60 min.Similarly,on the PeMS08 dataset,DMST-GNODE achieved the best performance with an RMSE of 19.4863 and an accuracy of 0.9377 at 20 min,demonstrating its effectiveness over longer periods.The Los_Loop dataset results further emphasise this model’s advantage,with an RMSE of 3.3422 and an accuracy of 0.7643 at 20 min,consistently maintaining superiority across all time intervals.These numerical highlights indicate that DMST-GNODE not only outperforms baseline models but also achieves higher accuracy and lower errors across different time intervals and datasets.
基金Supported by the Key R&D Program of Gansu Province(No.23YFGA0063)the Key Talent Project of Gansu Province(No.2024RCXM57,2024RCXM22)the Major Science and Technology Special Program of Gansu Province(No.25ZYJA037).
文摘Traffic flow prediction is a crucial element of intelligent transportation systems.However,accu-rate traffic flow prediction is quite challenging because of its highly nonlinear,complex,and dynam-ic characteristics.To address the difficulties in simultaneously capturing local and global dynamic spatiotemporal correlations in traffic flow,as well as the high time complexity of existing models,a multi-head flow attention-based local-global dynamic hypergraph convolution(MFA-LGDHC)pre-diction model is proposed.which consists of multi-head flow attention(MHFA)mechanism,graph convolution network(GCN),and local-global dynamic hypergraph convolution(LGHC).MHFA is utilized to extract the time dependency of traffic flow and reduce the time complexity of the model.GCN is employed to catch the spatial dependency of traffic flow.LGHC utilizes down-sampling con-volution and isometric convolution to capture the local and global spatial dependencies of traffic flow.And dynamic hypergraph convolution is used to model the dynamic higher-order relationships of the traffic road network.Experimental results indicate that the MFA-LGDHC model outperforms current popular baseline models and exhibits good prediction performance.
基金supported in part by the National Natural Science Foundations of China(No.61801214)the Postgraduate Research Practice Innovation Program of NUAA(No.xcxjh20231504)。
文摘Hyperspectral image(HSI)classification is crucial for numerous remote sensing applications.Traditional deep learning methods may miss pixel relationships and context,leading to inefficiencies.This paper introduces the spectral band graph convolutional and attention-enhanced CNN joint network(SGCCN),a novel approach that harnesses the power of spectral band graph convolutions for capturing long-range relationships,utilizes local perception of attention-enhanced multi-level convolutions for local spatial feature and employs a dynamic attention mechanism to enhance feature extraction.The SGCCN integrates spectral and spatial features through a self-attention fusion network,significantly improving classification accuracy and efficiency.The proposed method outperforms existing techniques,demonstrating its effectiveness in handling the challenges associated with HSI data.
基金Supported by National Natural Science Foundation of China,No.91959118Science and Technology Program of Guangzhou,China,No.201704020016+1 种基金SKY Radiology Department International Medical Research Foundation of China,No.Z-2014-07-1912-15Clinical Research Foundation of the 3rd Affiliated Hospital of Sun Yat-Sen University,No.YHJH201901.
文摘BACKGROUND The accurate classification of focal liver lesions(FLLs)is essential to properly guide treatment options and predict prognosis.Dynamic contrast-enhanced computed tomography(DCE-CT)is still the cornerstone in the exact classification of FLLs due to its noninvasive nature,high scanning speed,and high-density resolution.Since their recent development,convolutional neural network-based deep learning techniques has been recognized to have high potential for image recognition tasks.AIM To develop and evaluate an automated multiphase convolutional dense network(MP-CDN)to classify FLLs on multiphase CT.METHODS A total of 517 FLLs scanned on a 320-detector CT scanner using a four-phase DCECT imaging protocol(including precontrast phase,arterial phase,portal venous phase,and delayed phase)from 2012 to 2017 were retrospectively enrolled.FLLs were classified into four categories:Category A,hepatocellular carcinoma(HCC);category B,liver metastases;category C,benign non-inflammatory FLLs including hemangiomas,focal nodular hyperplasias and adenomas;and category D,hepatic abscesses.Each category was split into a training set and test set in an approximate 8:2 ratio.An MP-CDN classifier with a sequential input of the fourphase CT images was developed to automatically classify FLLs.The classification performance of the model was evaluated on the test set;the accuracy and specificity were calculated from the confusion matrix,and the area under the receiver operating characteristic curve(AUC)was calculated from the SoftMax probability outputted from the last layer of the MP-CDN.RESULTS A total of 410 FLLs were used for training and 107 FLLs were used for testing.The mean classification accuracy of the test set was 81.3%(87/107).The accuracy/specificity of distinguishing each category from the others were 0.916/0.964,0.925/0.905,0.860/0.918,and 0.925/0.963 for HCC,metastases,benign non-inflammatory FLLs,and abscesses on the test set,respectively.The AUC(95%confidence interval)for differentiating each category from the others was 0.92(0.837-0.992),0.99(0.967-1.00),0.88(0.795-0.955)and 0.96(0.914-0.996)for HCC,metastases,benign non-inflammatory FLLs,and abscesses on the test set,respectively.CONCLUSION MP-CDN accurately classified FLLs detected on four-phase CT as HCC,metastases,benign non-inflammatory FLLs and hepatic abscesses and may assist radiologists in identifying the different types of FLLs.
文摘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.
基金supported by the Open Project of Key Laboratory of Computational Aerodynamics,AVIC Aerodynamics Research Institute(Grant No.YL2022XFX0409).
文摘In this work,a three dimensional(3D)convolutional neural network(CNN)model based on image slices of various normal and pathological vocal folds is proposed for accurate and efficient prediction of glottal flows.The 3D CNN model is composed of the feature extraction block and regression block.The feature extraction block is capable of learning low dimensional features from the high dimensional image data of the glottal shape,and the regression block is employed to flatten the output from the feature extraction block and obtain the desired glottal flow data.The input image data is the condensed set of 2D image slices captured in the axial plane of the 3D vocal folds,where these glottal shapes are synthesized based on the equations of normal vibration modes.The output flow data is the corresponding flow rate,averaged glottal pressure and nodal pressure distributions over the glottal surface.The 3D CNN model is built to establish the mapping between the input image data and output flow data.The ground-truth flow variables of each glottal shape in the training and test datasets are obtained by a high-fidelity sharp-interface immersed-boundary solver.The proposed model is trained to predict the concerned flow variables for glottal shapes in the test set.The present 3D CNN model is more efficient than traditional Computational Fluid Dynamics(CFD)models while the accuracy can still be retained,and more powerful than previous data-driven prediction models because more details of the glottal flow can be provided.The prediction performance of the trained 3D CNN model in accuracy and efficiency indicates that this model could be promising for future clinical applications.
基金supported by the National Natural Science Foundation of China(Grant Nos.62141214 and 62272171).
文摘Classic Graph Convolutional Networks (GCNs) often learn node representation holistically, which ignores the distinct impacts from different neighbors when aggregating their features to update a node’s representation. Disentangled GCNs have been proposed to divide each node’s representation into several feature units. However, current disentangling methods do not try to figure out how many inherent factors the model should assign to help extract the best representation of each node. This paper then proposes D^(2)-GCN to provide dynamic disentanglement in GCNs and present the most appropriate factorization of each node’s mixed features. The convergence of the proposed method is proved both theoretically and experimentally. Experiments on real-world datasets show that D^(2)-GCN outperforms the baseline models concerning node classification results in both single- and multi-label tasks.
文摘With the rapid advancement of virtual reality,dynamic gesture recognition technology has become an indispensable and critical technique for users to achieve human–computer interaction in virtual environments.The recognition of dynamic gestures is a challenging task due to the high degree of freedom and the influence of individual differences and the change of gesture space.To solve the problem of low recognition accuracy of existing networks,an improved dynamic gesture recognition algorithm based on ResNeXt architecture is proposed.The algorithm employs three-dimensional convolution techniques to effectively capture the spatiotemporal features intrinsic to dynamic gestures.Additionally,to enhance the model’s focus and improve its accuracy in identifying dynamic gestures,a lightweight convolutional attention mechanism is introduced.This mechanism not only augments the model’s precision but also facilitates faster convergence during the training phase.In order to further optimize the performance of the model,a deep attention submodule is added to the convolutional attention mechanism module to strengthen the network’s capability in temporal feature extraction.Empirical evaluations on EgoGesture and NvGesture datasets show that the accuracy of the proposed model in dynamic gesture recognition reaches 95.03%and 86.21%,respectively.When operating in RGB mode,the accuracy reached 93.49%and 80.22%,respectively.These results underscore the effectiveness of the proposed algorithm in recognizing dynamic gestures with high accuracy,showcasing its potential for applications in advanced human–computer interaction systems.
基金supported by the the National Natural Science Foundation of China(51877167)。
文摘A reliable transformer protection method is crucial for power systems. Aiming at improving the generalization performance and response speed of multi-feature fusion based transformer protection, this paper presents a dynamic differential current by fusing pre-disturbance and post-disturbance differential currents in real time then developing a dynamic differential current based transformer protection focusing on the feature changes of differential current. Generally, the image of differential current can comprehensively embody the feature changes resulting from any disturbance. In addition, a short window is sometimes sufficient to clearly reflect the internal fault because the differential current will instantly change when an internal fault occurs. Therefore, in order to identify the running states reliably in the shortest possible time, multiple images, including the differential current from a pre-disturbance one cycle to a post-disturbance different time, are combined by time order to define a dynamic differential current. After the protection method is started, this dynamic differential current serves as input for the deep learning algorithm to identify the running states in real time. Once the transformer is identified as a faulty one, a tripping signal is issued and the protection method stops. The dynamic model experiments show that the proposed protection method has a strong generalization ability and rapid response speed.
基金supported by the National Natural Science Foundation of China(Nos.62002063 and U21A20472)Natural Science Foundation of Fujian Province(Nos.2020J05112 and 2022J01118)+1 种基金National Key Research and Development Plan of China(No.2021YFB3600503)Major Science and Technology Project of Fujian Province(No.2021HZ022007).
文摘Modeling and analysis of complex social networks is an important topic in social computing.Graph convolutional networks(GCNs)are widely used for learning social network embeddings and social network analysis.However,real-world complex social networks,such as Facebook and Math,exhibit significant global structural and dynamic characteristics that are not adequately captured by conventional GCN models.To address the above issues,this paper proposes a novel graph convolutional network considering global structural features and global temporal dependencies(GSTGCN).Specifically,we innovatively design a graph coarsening strategy based on the importance of social membership to construct a dynamic diffusion process of graphs.This dynamic diffusion process can be viewed as using higher-order subgraph embeddings to guide the generation of lower-order subgraph embeddings,and we model this process using gate recurrent unit(GRU)to extract comprehensive global structural features of the graph and the evolutionary processes embedded among subgraphs.Furthermore,we design a new evolutionary strategy that incorporates a temporal self-attention mechanism to enhance the extraction of global temporal dependencies of dynamic networks by GRU.GSTGCN outperforms current state-of-the-art network embedding methods in important social networks tasks such as link prediction and financial fraud identification.
文摘Multicolor fiber lasers have emerged as a promising technology with significant applications in optical communications,laser ranging,and precision sensing.Beyond their practical utility,these systems serve as ideal platforms for investigating fundamental soliton phenomena,including soliton collisions,explosions,and state transitions.However,the complex nonlinear dynamics inherent in these systems present substantial challenges for conventional numerical simulations.
基金supported by the National Natural Science Foundation of China(No.62476025)the Shaanxi Provincial Department of Science and Technology Projects(No.2013K06-39).
文摘Automatically extracting Drug-Drug Interactions (DDIs) from text is a crucial and challenging task, particularly when multiple medications are taken concurrently. In this study, we propose a novel approach, called Enhanced Attention-driven Dynamic Graph Convolutional Network (E-ADGCN), for DDI extraction. Our model combines the Attention-driven Dynamic Graph Convolutional Network (ADGCN) with a feature fusion method and multi-task learning framework. The ADGCN effectively utilizes entity information and dependency tree information from biomedical texts to extract DDIs. The feature fusion method integrates User-Generated Content (UGC) and molecular information with drug entity information from text through dynamic routing. By leveraging external resources, our approach maximizes the auxiliary effect and improves the accuracy of DDI extraction. We evaluate the E-ADGCN model on the extended DDIExtraction2013 dataset and achieve an F1-score of 81.45%. This research contributes to the advancement of automated methods for extracting valuable drug interaction information from textual sources, facilitating improved medication management and patient safety.
基金Supported by the National Natural Science Foundation of China(50830201)the Priority Academic Program Development of Jiangsu Higher Education Institutions~~
文摘Impact dynamics of flexible solids is important in engineering practice. Obtaining its dynamic response is a challenging task and usually achieved by numerical methods. The objectives of the study are twofold. Firstly, the discrete singular convolution （DSC） is used for the first time to analyze the impact dynamics. Secondly, the efficiency of various numerical methods for dynamic analysis is explored via an example of a flexible rod hit by a rigid ball. Three numerical methods, including the conventional finite element （FE） method, the DSC algorithm, and the spectral finite element （SFE） method, and one proposed modeling strategy, the improved spectral finite element （ISFE） method, are involved. Numerical results are compared with the known analytical solutions to show their efficiency. It is demonstrated that the proposed ISFE modeling strategy with a proper length of con- ventional FE yields the most accurate contact stress among the four investigated models. It is also found that the DSC algorithm is an alternative method for collision problems.
基金the National Key Research and Development Program of China(No.2021ZD0111902)the National Natural Science Foundation of China(Nos.62172022 and U21B2038)。
文摘Due to self-occlusion and high degree of freedom,estimating 3D hand pose from a single RGB image is a great challenging problem.Graph convolutional networks(GCNs)use graphs to describe the physical connection relationships between hand joints and improve the accuracy of 3D hand pose regression.However,GCNs cannot effectively describe the relationships between non-adjacent hand joints.Recently,hypergraph convolutional networks(HGCNs)have received much attention as they can describe multi-dimensional relationships between nodes through hyperedges;therefore,this paper proposes a framework for 3D hand pose estimation based on HGCN,which can better extract correlated relationships between adjacent and non-adjacent hand joints.To overcome the shortcomings of predefined hypergraph structures,a kind of dynamic hypergraph convolutional network is proposed,in which hyperedges are constructed dynamically based on hand joint feature similarity.To better explore the local semantic relationships between nodes,a kind of semantic dynamic hypergraph convolution is proposed.The proposed method is evaluated on publicly available benchmark datasets.Qualitative and quantitative experimental results both show that the proposed HGCN and improved methods for 3D hand pose estimation are better than GCN,and achieve state-of-the-art performance compared with existing methods.
文摘Reliable traffic flow prediction is crucial for mitigating urban congestion.This paper proposes Attentionbased spatiotemporal Interactive Dynamic Graph Convolutional Network(AIDGCN),a novel architecture integrating Interactive Dynamic Graph Convolution Network(IDGCN)with Temporal Multi-Head Trend-Aware Attention.Its core innovation lies in IDGCN,which uniquely splits sequences into symmetric intervals for interactive feature sharing via dynamic graphs,and a novel attention mechanism incorporating convolutional operations to capture essential local traffic trends—addressing a critical gap in standard attention for continuous data.For 15-and 60-min forecasting on METR-LA,AIDGCN achieves MAEs of 0.75%and 0.39%,and RMSEs of 1.32%and 0.14%,respectively.In the 60-min long-term forecasting of the PEMS-BAY dataset,the AIDGCN out-performs the MRA-BGCN method by 6.28%,4.93%,and 7.17%in terms of MAE,RMSE,and MAPE,respectively.Experimental results demonstrate the superiority of our pro-posed model over state-of-the-art methods.
文摘Venanico-Filho et al. developed an elegant matrix formulation for dynamic analysis by frequency domain （FD）, but the convergence, causality and extended period need further refining. In the present paper, it was argued that： （1） under reasonable assumptions （approximating the frequency response function by the discrete Fourier transform of the discretized unitary impulse response function）, the matrix formulation by FD is equivalent to a circular convolution; （2） to avoid the wraparound interference, the excitation vector and impulse response must be padded with enough zeros; （3） provided that the zero padding requirement satisfied, the convergence and accuracy of direct time domain analysis, which is equivalent to that by FD, are guaranteed by the numerical integration scheme; （4） the imaginary part of the computational response approaching zero is due to the continuity of the impulse response functions.
基金supported by National Key R&D Program of China (2020AAA0107901).
文摘Latent information is difficult to get from the text in speech synthesis.Studies show that features from speech can get more information to help text encoding.In the field of speech encoding,a lot of work has been conducted on two aspects.The first aspect is to encode speech frame by frame.The second aspect is to encode the whole speech to a vector.But the scale in these aspects is fixed.So,encoding speech with an adjustable scale for more latent information is worthy of investigation.But current alignment approaches only support frame-by-frame encoding and speech-to-vector encoding.It remains a challenge to propose a new alignment approach to support adjustable scale speech encoding.This paper presents the dynamic speech encoder with a new alignment approach in conjunction with frame-by-frame encoding and speech-to-vector encoding.The speech feature fromourmodel achieves three functions.First,the speech feature can reconstruct the origin speech while the length of the speech feature is equal to the text length.Second,our model can get text embedding fromspeech,and the encoded speech feature is similar to the text embedding result.Finally,it can transfer the style of synthesis speech and make it more similar to the given reference speech.
基金This study was supported by the Basic Science Research Program through the National Research Foundation of Korea(NRF)funded by the Ministry of Education(NRF-2018R1D1A1B07049932).
文摘Low dynamic range(LDR)images captured by consumer cameras have a limited luminance range.As the conventional method for generating high dynamic range(HDR)images involves merging multiple-exposure LDR images of the same scene(assuming a stationary scene),we introduce a learning-based model for single-image HDR reconstruction.An input LDR image is sequentially segmented into the local region maps based on the cumulative histogram of the input brightness distribution.Using the local region maps,SParam-Net estimates the parameters of an inverse tone mapping function to generate a pseudo-HDR image.We process the segmented region maps as the input sequences on long short-term memory.Finally,a fast super-resolution convolutional neural network is used for HDR image reconstruction.The proposed method was trained and tested on datasets including HDR-Real,LDR-HDR-pair,and HDR-Eye.The experimental results revealed that HDR images can be generated more reliably than using contemporary end-to-end approaches.
文摘Much attention has been given to the Internet of Things (IoT) by citizens, industries, governments, and universities for applications like smart buildings, environmental monitoring, health care and so on. With IoT, networkconnectivity is facilitated between smart devices from anyplace and anytime.IoT-based health monitoring systems are gaining popularity and acceptance forcontinuous monitoring and detect health abnormalities from the data collected.Electrocardiographic (ECG) signals are widely used for heart diseases detection.A novel method has been proposed in this work for ECG monitoring using IoTtechniques. In this work, a two-stage approach is employed. In the first stage, arouting protocol based on Dynamic Source Routing (DSR) and Routing byEnergy and Link quality (REL) for IoT healthcare platform is proposed for effi-cient data collection, and in the second stage, classification of ECG for Arrhythmia. Furthermore, this work has evaluated Support Vector Machine (SVM),Artificial Neural Network (ANN), and Convolution Neural Networks (CNNs)-based approach for ECG signals classification. Deep-ECG will use a deep CNNto extract critical features and then compare through evaluation of simple and fastdistance functions in order to obtain an efficient classification of heart abnormalities. For the identification of abnormal data, this work has proposed techniquesfor the classification of ECG data, which has been obtained from mobile watchusers. For experimental verification of the proposed methods, the Beth Israel Hospital (MIT/BIH) Arrhythmia and Massachusetts Institute of Technology (MIT)Database was used for evaluation. Results confirm the presented method’s superior performance with regards to the accuracy of classification. The CNN achievedan accuracy of 91.92% and has a higher accuracy of 4.98% for the SVM and2.68% for the ANN.
