Medical image analysis based on deep learning has become an important technical requirement in the field of smart healthcare.In view of the difficulties in collaborative modeling of local details and global features i...Medical image analysis based on deep learning has become an important technical requirement in the field of smart healthcare.In view of the difficulties in collaborative modeling of local details and global features in multimodal image analysis of ophthalmology,as well as the existence of information redundancy in cross-modal data fusion,this paper proposes amultimodal fusion framework based on cross-modal collaboration and weighted attention mechanism.In terms of feature extraction,the framework collaboratively extracts local fine-grained features and global structural dependencies through a parallel dual-branch architecture,overcoming the limitations of traditional single-modality models in capturing either local or global information;in terms of fusion strategy,the framework innovatively designs a cross-modal dynamic fusion strategy,combining overlappingmulti-head self-attention modules with a bidirectional feature alignment mechanism,addressing the bottlenecks of low feature interaction efficiency and excessive attention fusion computations in traditional parallel fusion,and further introduces cross-domain local integration technology,which enhances the representation ability of the lesion area through pixel-level feature recalibration and optimizes the diagnostic robustness of complex cases.Experiments show that the framework exhibits excellent feature expression and generalization performance in cross-domain scenarios of ophthalmic medical images and natural images,providing a high-precision,low-redundancy fusion paradigm for multimodal medical image analysis,and promoting the upgrade of intelligent diagnosis and treatment fromsingle-modal static analysis to dynamic decision-making.展开更多
Recent advances in spatially resolved transcriptomics(SRT)have provided new opportunities for characterizing spatial structures of various tissues.Graph-based geometric deep learning has gained widespread adoption for...Recent advances in spatially resolved transcriptomics(SRT)have provided new opportunities for characterizing spatial structures of various tissues.Graph-based geometric deep learning has gained widespread adoption for spatial domain identification tasks.Currently,most methods define adjacency relation between cells or spots by their spatial distance in SRT data,which overlooks key biological interactions like gene expression similarities,and leads to inaccuracies in spatial domain identification.To tackle this challenge,we propose a novel method,SpaGRA(https://github.com/sunxue-yy/SpaGRA),for automatic multi-relationship construction based on graph augmentation.SpaGRA uses spatial distance as prior knowledge and dynamically adjusts edge weights with multi-head graph attention networks(GATs).This helps SpaGRA to uncover diverse node relationships and enhance message passing in geometric contrastive learning.Additionally,SpaGRA uses these multi-view relationships to construct negative samples,addressing sampling bias posed by random selection.Experimental results show that SpaGRA presents superior domain identification performance on multiple datasets generated from different protocols.Using SpaGRA,we analyze the functional regions in the mouse hypothalamus,identify key genes related to heart development in mouse embryos,and observe cancer-associated fibroblasts enveloping cancer cells in the latest Visium HD data.Overall,SpaGRA can effectively characterize spatial structures across diverse SRT datasets.展开更多
Brain tumor identification is a challenging task in neuro-oncology.The brain’s complex anatomy makes it a crucial part of the central nervous system.Accurate tumor classification is crucial for clinical diagnosis and...Brain tumor identification is a challenging task in neuro-oncology.The brain’s complex anatomy makes it a crucial part of the central nervous system.Accurate tumor classification is crucial for clinical diagnosis and treatment planning.This research presents a significant advancement in the multi-classification of brain tumors.This paper proposed a novel architecture that integrates Enhanced ResNeXt 101_32×8d,a Convolutional Neural Network(CNN)with a multi-head self-attention(MHSA)mechanism.This combination harnesses the strengths of the feature extraction,feature representation by CNN,and long-range dependencies by MHSA.Magnetic Resonance Imaging(MRI)datasets were employed to check the effectiveness of the proposed architecture.The first dataset(DS-1,Msoud)included four brain tumor classes,and the second dataset(DS-2)contained seven brain tumor classes.This methodology effectively distinguished various tumor classes,achieving high accuracies of 99.75% on DS-1 and 98.80% on DS-2.These impressive results indicate the superior performance and adaptability of our model for multiclass brain tumor classification.Evaluationmetrics such as accuracy,precision,recall,F1 score,and ROC(receiver operating characteristic)curve were utilized to comprehensively evaluate model validity.The performance results showed that the model is well-suited for clinical applications,with reduced errors and high accuracy.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
To fully make use of information from different representation subspaces,a multi-head attention-based long short-term memory(LSTM)model is proposed in this study for speech emotion recognition(SER).The proposed model ...To fully make use of information from different representation subspaces,a multi-head attention-based long short-term memory(LSTM)model is proposed in this study for speech emotion recognition(SER).The proposed model uses frame-level features and takes the temporal information of emotion speech as the input of the LSTM layer.Here,a multi-head time-dimension attention(MHTA)layer was employed to linearly project the output of the LSTM layer into different subspaces for the reduced-dimension context vectors.To provide relative vital information from other dimensions,the output of MHTA,the output of feature-dimension attention,and the last time-step output of LSTM were utilized to form multiple context vectors as the input of the fully connected layer.To improve the performance of multiple vectors,feature-dimension attention was employed for the all-time output of the first LSTM layer.The proposed model was evaluated on the eNTERFACE and GEMEP corpora,respectively.The results indicate that the proposed model outperforms LSTM by 14.6%and 10.5%for eNTERFACE and GEMEP,respectively,proving the effectiveness of the proposed model in SER tasks.展开更多
Due to the closed working environment of shield machines,the construction personnel cannot observe the construction geological environment,which seriously restricts the safety and efficiency of the tunneling process.I...Due to the closed working environment of shield machines,the construction personnel cannot observe the construction geological environment,which seriously restricts the safety and efficiency of the tunneling process.In this study,we present an enhanced multi-head self-attention convolution neural network(EMSACNN)with two-stage feature extraction for geological condition prediction of shield machine.Firstly,we select 30 important parameters according to statistical analysis method and the working principle of the shield machine.Then,we delete the non-working sample data,and combine 10 consecutive data as the input of the model.Thereafter,to deeply mine and extract essential and relevant features,we build a novel model combined with the particularity of the geological type recognition task,in which an enhanced multi-head self-attention block is utilized as the first feature extractor to fully extract the correlation of geological information of adjacent working face of tunnel,and two-dimensional CNN(2dCNN)is utilized as the second feature extractor.The performance and superiority of proposed EMSACNN are verified by the actual data collected by the shield machine used in the construction of a double-track tunnel in Guangzhou,China.The results show that EMSACNN achieves at least 96%accuracy on the test sets of the two tunnels,and all the evaluation indicators of EMSACNN are much better than those of classical AI model and the model that use only the second-stage feature extractor.Therefore,the proposed EMSACNN achieves high accuracy and strong generalization for geological information prediction of shield machine,which is of great guiding significance to engineering practice.展开更多
The deep learning advancements have greatly improved the performance of speech recognition systems,and most recent systems are based on the Recurrent Neural Network(RNN).Overall,the RNN works fine with the small seque...The deep learning advancements have greatly improved the performance of speech recognition systems,and most recent systems are based on the Recurrent Neural Network(RNN).Overall,the RNN works fine with the small sequence data,but suffers from the gradient vanishing problem in case of large sequence.The transformer networks have neutralized this issue and have shown state-of-the-art results on sequential or speech-related data.Generally,in speech recognition,the input audio is converted into an image using Mel-spectrogram to illustrate frequencies and intensities.The image is classified by the machine learning mechanism to generate a classification transcript.However,the audio frequency in the image has low resolution and causing inaccurate predictions.This paper presents a novel end-to-end binary view transformer-based architecture for speech recognition to cope with the frequency resolution problem.Firstly,the input audio signal is transformed into a 2D image using Mel-spectrogram.Secondly,the modified universal transformers utilize the multi-head attention to derive contextual information and derive different speech-related features.Moreover,a feedforward neural network is also deployed for classification.The proposed system has generated robust results on Google’s speech command dataset with an accuracy of 95.16%and with minimal loss.The binary-view transformer eradicates the eventuality of the over-fitting problem by deploying a multiview mechanism to diversify the input data,and multi-head attention captures multiple contexts from the data’s feature map.展开更多
Considering the stealthiness and persistence of Advanced Persistent Threats(APTs),system audit logs are leveraged in recent studies to construct system entity interaction provenance graphs to unveil threats in a host....Considering the stealthiness and persistence of Advanced Persistent Threats(APTs),system audit logs are leveraged in recent studies to construct system entity interaction provenance graphs to unveil threats in a host.Rule-based provenance graph APT detection approaches require elaborate rules and cannot detect unknown attacks,and existing learning-based approaches are limited by the lack of available APT attack samples or generally only perform graph-level anomaly detection,which requires lots of manual efforts to locate attack entities.This paper proposes an APT-exploited process detection approach called ThreatSniffer,which constructs the benign provenance graph from attack-free audit logs,fits normal system entity interactions and then detects APT-exploited processes by predicting the rationality of entity interactions.Firstly,ThreatSniffer understands system entities in terms of their file paths,interaction sequences,and the number distribution of interaction types and uses the multi-head self-attention mechanism to fuse these semantics.Then,based on the insight that APT-exploited processes interact with system entities they should not invoke,ThreatSniffer performs negative sampling on the benign provenance graph to generate non-existent edges,thus characterizing irrational entity interactions without requiring APT attack samples.At last,it employs a heterogeneous graph neural network as the interaction prediction model to aggregate the contextual information of entity interactions,and locate processes exploited by attackers,thereby achieving fine-grained APT detection.Evaluation results demonstrate that anomaly-based detection enables ThreatSniffer to identify all attack activities.Compared to the node-level APT detection method APT-KGL,ThreatSniffer achieves a 6.1%precision improvement because of its comprehensive understanding of entity semantics.展开更多
The integration of Unmanned Aerial Vehicles(UAVs)into Intelligent Transportation Systems(ITS)holds trans-formative potential for real-time traffic monitoring,a critical component of emerging smart city infrastructure....The integration of Unmanned Aerial Vehicles(UAVs)into Intelligent Transportation Systems(ITS)holds trans-formative potential for real-time traffic monitoring,a critical component of emerging smart city infrastructure.UAVs offer unique advantages over stationary traffic cameras,including greater flexibility in monitoring large and dynamic urban areas.However,detecting small,densely packed vehicles in UAV imagery remains a significant challenge due to occlusion,variations in lighting,and the complexity of urban landscapes.Conventional models often struggle with these issues,leading to inaccurate detections and reduced performance in practical applications.To address these challenges,this paper introduces CFEMNet,an advanced deep learning model specifically designed for high-precision vehicle detection in complex urban environments.CFEMNet is built on the High-Resolution Network(HRNet)architecture and integrates a Context-aware Feature Extraction Module(CFEM),which combines multi-scale feature learning with a novel Self-Attention and Convolution layer setup within a Multi-scale Feature Block(MFB).This combination allows CFEMNet to accurately capture fine-grained details across varying scales,crucial for detecting small or partially occluded vehicles.Furthermore,the model incorporates an Equivalent Feed-Forward Network(EFFN)Block to ensure robust extraction of both spatial and semantic features,enhancing its ability to distinguish vehicles from similar objects.To optimize computational efficiency,CFEMNet employs a local window adaptation of Multi-head Self-Attention(MSA),which reduces memory overhead without sacrificing detection accuracy.Extensive experimental evaluations on the UAVDT and VisDrone-DET2018 datasets confirm CFEMNet’s superior performance in vehicle detection compared to existing models.This new architecture establishes CFEMNet as a benchmark for UAV-enabled traffic management,offering enhanced precision,reduced computational demands,and scalability for deployment in smart city applications.The advancements presented in CFEMNet contribute significantly to the evolution of smart city technologies,providing a foundation for intelligent and responsive traffic management systems that can adapt to the dynamic demands of urban environments.展开更多
Nowadays,with the rapid development of industrial Internet technology,on the one hand,advanced industrial control systems(ICS)have improved industrial production efficiency.However,there are more and more cyber-attack...Nowadays,with the rapid development of industrial Internet technology,on the one hand,advanced industrial control systems(ICS)have improved industrial production efficiency.However,there are more and more cyber-attacks targeting industrial control systems.To ensure the security of industrial networks,intrusion detection systems have been widely used in industrial control systems,and deep neural networks have always been an effective method for identifying cyber attacks.Current intrusion detection methods still suffer from low accuracy and a high false alarm rate.Therefore,it is important to build a more efficient intrusion detection model.This paper proposes a hybrid deep learning intrusion detection method based on convolutional neural networks and bidirectional long short-term memory neural networks(CNN-BiLSTM).To address the issue of imbalanced data within the dataset and improve the model’s detection capabilities,the Synthetic Minority Over-sampling Technique-Edited Nearest Neighbors(SMOTE-ENN)algorithm is applied in the preprocessing phase.This algorithm is employed to generate synthetic instances for the minority class,simultaneously mitigating the impact of noise in the majority class.This approach aims to create a more equitable distribution of classes,thereby enhancing the model’s ability to effectively identify patterns in both minority and majority classes.In the experimental phase,the detection performance of the method is verified using two data sets.Experimental results show that the accuracy rate on the CICIDS-2017 data set reaches 97.7%.On the natural gas pipeline dataset collected by Lan Turnipseed from Mississippi State University in the United States,the accuracy rate also reaches 85.5%.展开更多
Visual question answering(VQA)requires a deep understanding of images and their corresponding textual questions to answer questions about images more accurately.However,existing models tend to ignore the implicit know...Visual question answering(VQA)requires a deep understanding of images and their corresponding textual questions to answer questions about images more accurately.However,existing models tend to ignore the implicit knowledge in the images and focus only on the visual information in the images,which limits the understanding depth of the image content.The images contain more than just visual objects,some images contain textual information about the scene,and slightly more complex images contain relationships between individual visual objects.Firstly,this paper proposes a model using image description for feature enhancement.This model encodes images and their descriptions separately based on the question-guided coattention mechanism.This mechanism increases the feature representation of the model,enhancing the model’s ability for reasoning.In addition,this paper improves the bottom-up attention model by obtaining two image region features.After obtaining the two visual features and the spatial position information corresponding to each feature,concatenating the two features as the final image feature can better represent an image.Finally,the obtained spatial position information is processed to enable the model to perceive the size and relative position of each object in the image.Our best single model delivers a 74.16%overall accuracy on the VQA 2.0 dataset,our model even outperforms some multi-modal pre-training models with fewer images and a shorter time.展开更多
Due to the rapid evolution of Advanced Persistent Threats(APTs)attacks,the emergence of new and rare attack samples,and even those never seen before,make it challenging for traditional rule-based detection methods to ...Due to the rapid evolution of Advanced Persistent Threats(APTs)attacks,the emergence of new and rare attack samples,and even those never seen before,make it challenging for traditional rule-based detection methods to extract universal rules for effective detection.With the progress in techniques such as transfer learning and meta-learning,few-shot network attack detection has progressed.However,challenges in few-shot network attack detection arise from the inability of time sequence flow features to adapt to the fixed length input requirement of deep learning,difficulties in capturing rich information from original flow in the case of insufficient samples,and the challenge of high-level abstract representation.To address these challenges,a few-shot network attack detection based on NFHP(Network Flow Holographic Picture)-RN(ResNet)is proposed.Specifically,leveraging inherent properties of images such as translation invariance,rotation invariance,scale invariance,and illumination invariance,network attack traffic features and contextual relationships are intuitively represented in NFHP.In addition,an improved RN network model is employed for high-level abstract feature extraction,ensuring that the extracted high-level abstract features maintain the detailed characteristics of the original traffic behavior,regardless of changes in background traffic.Finally,a meta-learning model based on the self-attention mechanism is constructed,achieving the detection of novel APT few-shot network attacks through the empirical generalization of high-level abstract feature representations of known-class network attack behaviors.Experimental results demonstrate that the proposed method can learn high-level abstract features of network attacks across different traffic detail granularities.Comparedwith state-of-the-artmethods,it achieves favorable accuracy,precision,recall,and F1 scores for the identification of unknown-class network attacks through cross-validation onmultiple datasets.展开更多
基金funded by the Ongoing Research Funding Program(ORF-2025-102),King Saud University,Riyadh,Saudi Arabiaby the Science and Technology Research Programof Chongqing Municipal Education Commission(Grant No.KJQN202400813)by the Graduate Research Innovation Project(Grant Nos.yjscxx2025-269-193 and CYS25618).
文摘Medical image analysis based on deep learning has become an important technical requirement in the field of smart healthcare.In view of the difficulties in collaborative modeling of local details and global features in multimodal image analysis of ophthalmology,as well as the existence of information redundancy in cross-modal data fusion,this paper proposes amultimodal fusion framework based on cross-modal collaboration and weighted attention mechanism.In terms of feature extraction,the framework collaboratively extracts local fine-grained features and global structural dependencies through a parallel dual-branch architecture,overcoming the limitations of traditional single-modality models in capturing either local or global information;in terms of fusion strategy,the framework innovatively designs a cross-modal dynamic fusion strategy,combining overlappingmulti-head self-attention modules with a bidirectional feature alignment mechanism,addressing the bottlenecks of low feature interaction efficiency and excessive attention fusion computations in traditional parallel fusion,and further introduces cross-domain local integration technology,which enhances the representation ability of the lesion area through pixel-level feature recalibration and optimizes the diagnostic robustness of complex cases.Experiments show that the framework exhibits excellent feature expression and generalization performance in cross-domain scenarios of ophthalmic medical images and natural images,providing a high-precision,low-redundancy fusion paradigm for multimodal medical image analysis,and promoting the upgrade of intelligent diagnosis and treatment fromsingle-modal static analysis to dynamic decision-making.
基金supported by the National Natural Science Foundation of China(Nos.62303271,U1806202,62103397)the Natural Science Foundation of Shandong Province(ZR2023QF081)Funding for open access charge:the National Natural Science Foundation of China(Nos.62303271,U1806202).
文摘Recent advances in spatially resolved transcriptomics(SRT)have provided new opportunities for characterizing spatial structures of various tissues.Graph-based geometric deep learning has gained widespread adoption for spatial domain identification tasks.Currently,most methods define adjacency relation between cells or spots by their spatial distance in SRT data,which overlooks key biological interactions like gene expression similarities,and leads to inaccuracies in spatial domain identification.To tackle this challenge,we propose a novel method,SpaGRA(https://github.com/sunxue-yy/SpaGRA),for automatic multi-relationship construction based on graph augmentation.SpaGRA uses spatial distance as prior knowledge and dynamically adjusts edge weights with multi-head graph attention networks(GATs).This helps SpaGRA to uncover diverse node relationships and enhance message passing in geometric contrastive learning.Additionally,SpaGRA uses these multi-view relationships to construct negative samples,addressing sampling bias posed by random selection.Experimental results show that SpaGRA presents superior domain identification performance on multiple datasets generated from different protocols.Using SpaGRA,we analyze the functional regions in the mouse hypothalamus,identify key genes related to heart development in mouse embryos,and observe cancer-associated fibroblasts enveloping cancer cells in the latest Visium HD data.Overall,SpaGRA can effectively characterize spatial structures across diverse SRT datasets.
文摘Brain tumor identification is a challenging task in neuro-oncology.The brain’s complex anatomy makes it a crucial part of the central nervous system.Accurate tumor classification is crucial for clinical diagnosis and treatment planning.This research presents a significant advancement in the multi-classification of brain tumors.This paper proposed a novel architecture that integrates Enhanced ResNeXt 101_32×8d,a Convolutional Neural Network(CNN)with a multi-head self-attention(MHSA)mechanism.This combination harnesses the strengths of the feature extraction,feature representation by CNN,and long-range dependencies by MHSA.Magnetic Resonance Imaging(MRI)datasets were employed to check the effectiveness of the proposed architecture.The first dataset(DS-1,Msoud)included four brain tumor classes,and the second dataset(DS-2)contained seven brain tumor classes.This methodology effectively distinguished various tumor classes,achieving high accuracies of 99.75% on DS-1 and 98.80% on DS-2.These impressive results indicate the superior performance and adaptability of our model for multiclass brain tumor classification.Evaluationmetrics such as accuracy,precision,recall,F1 score,and ROC(receiver operating characteristic)curve were utilized to comprehensively evaluate model validity.The performance results showed that the model is well-suited for clinical applications,with reduced errors and high accuracy.
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.
基金The National Natural Science Foundation of China(No.61571106,61633013,61673108,81871444).
文摘To fully make use of information from different representation subspaces,a multi-head attention-based long short-term memory(LSTM)model is proposed in this study for speech emotion recognition(SER).The proposed model uses frame-level features and takes the temporal information of emotion speech as the input of the LSTM layer.Here,a multi-head time-dimension attention(MHTA)layer was employed to linearly project the output of the LSTM layer into different subspaces for the reduced-dimension context vectors.To provide relative vital information from other dimensions,the output of MHTA,the output of feature-dimension attention,and the last time-step output of LSTM were utilized to form multiple context vectors as the input of the fully connected layer.To improve the performance of multiple vectors,feature-dimension attention was employed for the all-time output of the first LSTM layer.The proposed model was evaluated on the eNTERFACE and GEMEP corpora,respectively.The results indicate that the proposed model outperforms LSTM by 14.6%and 10.5%for eNTERFACE and GEMEP,respectively,proving the effectiveness of the proposed model in SER tasks.
基金supported by the National Key R&D Program of China(Grant No.2019YFB1705203)Shanghai Municipal Science and Technology Major Project(2021SHZDZX0102).
文摘Due to the closed working environment of shield machines,the construction personnel cannot observe the construction geological environment,which seriously restricts the safety and efficiency of the tunneling process.In this study,we present an enhanced multi-head self-attention convolution neural network(EMSACNN)with two-stage feature extraction for geological condition prediction of shield machine.Firstly,we select 30 important parameters according to statistical analysis method and the working principle of the shield machine.Then,we delete the non-working sample data,and combine 10 consecutive data as the input of the model.Thereafter,to deeply mine and extract essential and relevant features,we build a novel model combined with the particularity of the geological type recognition task,in which an enhanced multi-head self-attention block is utilized as the first feature extractor to fully extract the correlation of geological information of adjacent working face of tunnel,and two-dimensional CNN(2dCNN)is utilized as the second feature extractor.The performance and superiority of proposed EMSACNN are verified by the actual data collected by the shield machine used in the construction of a double-track tunnel in Guangzhou,China.The results show that EMSACNN achieves at least 96%accuracy on the test sets of the two tunnels,and all the evaluation indicators of EMSACNN are much better than those of classical AI model and the model that use only the second-stage feature extractor.Therefore,the proposed EMSACNN achieves high accuracy and strong generalization for geological information prediction of shield machine,which is of great guiding significance to engineering practice.
基金This research was supported by Suranaree University of Technology,Thailand,Grant Number:BRO7-709-62-12-03.
文摘The deep learning advancements have greatly improved the performance of speech recognition systems,and most recent systems are based on the Recurrent Neural Network(RNN).Overall,the RNN works fine with the small sequence data,but suffers from the gradient vanishing problem in case of large sequence.The transformer networks have neutralized this issue and have shown state-of-the-art results on sequential or speech-related data.Generally,in speech recognition,the input audio is converted into an image using Mel-spectrogram to illustrate frequencies and intensities.The image is classified by the machine learning mechanism to generate a classification transcript.However,the audio frequency in the image has low resolution and causing inaccurate predictions.This paper presents a novel end-to-end binary view transformer-based architecture for speech recognition to cope with the frequency resolution problem.Firstly,the input audio signal is transformed into a 2D image using Mel-spectrogram.Secondly,the modified universal transformers utilize the multi-head attention to derive contextual information and derive different speech-related features.Moreover,a feedforward neural network is also deployed for classification.The proposed system has generated robust results on Google’s speech command dataset with an accuracy of 95.16%and with minimal loss.The binary-view transformer eradicates the eventuality of the over-fitting problem by deploying a multiview mechanism to diversify the input data,and multi-head attention captures multiple contexts from the data’s feature map.
基金This work was supported by the National Natural Science Foundation of China(Nos.U19A2081,62202320)the Fundamental Research Funds for the Central Universities(Nos.2022SCU12116,2023SCU12129,2023SCU12126)+1 种基金the Science and Engineering Connotation Development Project of Sichuan University(No.2020SCUNG129)the Key Laboratory of Data Protection and Intelligent Management(Sichuan University),Ministry of Education.
文摘Considering the stealthiness and persistence of Advanced Persistent Threats(APTs),system audit logs are leveraged in recent studies to construct system entity interaction provenance graphs to unveil threats in a host.Rule-based provenance graph APT detection approaches require elaborate rules and cannot detect unknown attacks,and existing learning-based approaches are limited by the lack of available APT attack samples or generally only perform graph-level anomaly detection,which requires lots of manual efforts to locate attack entities.This paper proposes an APT-exploited process detection approach called ThreatSniffer,which constructs the benign provenance graph from attack-free audit logs,fits normal system entity interactions and then detects APT-exploited processes by predicting the rationality of entity interactions.Firstly,ThreatSniffer understands system entities in terms of their file paths,interaction sequences,and the number distribution of interaction types and uses the multi-head self-attention mechanism to fuse these semantics.Then,based on the insight that APT-exploited processes interact with system entities they should not invoke,ThreatSniffer performs negative sampling on the benign provenance graph to generate non-existent edges,thus characterizing irrational entity interactions without requiring APT attack samples.At last,it employs a heterogeneous graph neural network as the interaction prediction model to aggregate the contextual information of entity interactions,and locate processes exploited by attackers,thereby achieving fine-grained APT detection.Evaluation results demonstrate that anomaly-based detection enables ThreatSniffer to identify all attack activities.Compared to the node-level APT detection method APT-KGL,ThreatSniffer achieves a 6.1%precision improvement because of its comprehensive understanding of entity semantics.
基金funded by the Deanship of Scientific Research at Northern Border University,Arar,Saudi Arabia through research group No.(RG-NBU-2022-1234).
文摘The integration of Unmanned Aerial Vehicles(UAVs)into Intelligent Transportation Systems(ITS)holds trans-formative potential for real-time traffic monitoring,a critical component of emerging smart city infrastructure.UAVs offer unique advantages over stationary traffic cameras,including greater flexibility in monitoring large and dynamic urban areas.However,detecting small,densely packed vehicles in UAV imagery remains a significant challenge due to occlusion,variations in lighting,and the complexity of urban landscapes.Conventional models often struggle with these issues,leading to inaccurate detections and reduced performance in practical applications.To address these challenges,this paper introduces CFEMNet,an advanced deep learning model specifically designed for high-precision vehicle detection in complex urban environments.CFEMNet is built on the High-Resolution Network(HRNet)architecture and integrates a Context-aware Feature Extraction Module(CFEM),which combines multi-scale feature learning with a novel Self-Attention and Convolution layer setup within a Multi-scale Feature Block(MFB).This combination allows CFEMNet to accurately capture fine-grained details across varying scales,crucial for detecting small or partially occluded vehicles.Furthermore,the model incorporates an Equivalent Feed-Forward Network(EFFN)Block to ensure robust extraction of both spatial and semantic features,enhancing its ability to distinguish vehicles from similar objects.To optimize computational efficiency,CFEMNet employs a local window adaptation of Multi-head Self-Attention(MSA),which reduces memory overhead without sacrificing detection accuracy.Extensive experimental evaluations on the UAVDT and VisDrone-DET2018 datasets confirm CFEMNet’s superior performance in vehicle detection compared to existing models.This new architecture establishes CFEMNet as a benchmark for UAV-enabled traffic management,offering enhanced precision,reduced computational demands,and scalability for deployment in smart city applications.The advancements presented in CFEMNet contribute significantly to the evolution of smart city technologies,providing a foundation for intelligent and responsive traffic management systems that can adapt to the dynamic demands of urban environments.
基金support from the Liaoning Province Nature Fund Project(No.2022-MS-291)the Scientific Research Project of Liaoning Province Education Department(LJKMZ20220781,LJKMZ20220783,LJKQZ20222457,JYTMS20231488).
文摘Nowadays,with the rapid development of industrial Internet technology,on the one hand,advanced industrial control systems(ICS)have improved industrial production efficiency.However,there are more and more cyber-attacks targeting industrial control systems.To ensure the security of industrial networks,intrusion detection systems have been widely used in industrial control systems,and deep neural networks have always been an effective method for identifying cyber attacks.Current intrusion detection methods still suffer from low accuracy and a high false alarm rate.Therefore,it is important to build a more efficient intrusion detection model.This paper proposes a hybrid deep learning intrusion detection method based on convolutional neural networks and bidirectional long short-term memory neural networks(CNN-BiLSTM).To address the issue of imbalanced data within the dataset and improve the model’s detection capabilities,the Synthetic Minority Over-sampling Technique-Edited Nearest Neighbors(SMOTE-ENN)algorithm is applied in the preprocessing phase.This algorithm is employed to generate synthetic instances for the minority class,simultaneously mitigating the impact of noise in the majority class.This approach aims to create a more equitable distribution of classes,thereby enhancing the model’s ability to effectively identify patterns in both minority and majority classes.In the experimental phase,the detection performance of the method is verified using two data sets.Experimental results show that the accuracy rate on the CICIDS-2017 data set reaches 97.7%.On the natural gas pipeline dataset collected by Lan Turnipseed from Mississippi State University in the United States,the accuracy rate also reaches 85.5%.
基金supported in part by the National Natural Science Foundation of China under Grant U1911401.
文摘Visual question answering(VQA)requires a deep understanding of images and their corresponding textual questions to answer questions about images more accurately.However,existing models tend to ignore the implicit knowledge in the images and focus only on the visual information in the images,which limits the understanding depth of the image content.The images contain more than just visual objects,some images contain textual information about the scene,and slightly more complex images contain relationships between individual visual objects.Firstly,this paper proposes a model using image description for feature enhancement.This model encodes images and their descriptions separately based on the question-guided coattention mechanism.This mechanism increases the feature representation of the model,enhancing the model’s ability for reasoning.In addition,this paper improves the bottom-up attention model by obtaining two image region features.After obtaining the two visual features and the spatial position information corresponding to each feature,concatenating the two features as the final image feature can better represent an image.Finally,the obtained spatial position information is processed to enable the model to perceive the size and relative position of each object in the image.Our best single model delivers a 74.16%overall accuracy on the VQA 2.0 dataset,our model even outperforms some multi-modal pre-training models with fewer images and a shorter time.
基金supported by the National Natural Science Foundation of China(Nos.U19A208162202320)+2 种基金the Fundamental Research Funds for the Central Universities(No.SCU2023D008)the Science and Engineering Connotation Development Project of Sichuan University(No.2020SCUNG129)the Key Laboratory of Data Protection and Intelligent Management(Sichuan University),Ministry of Education.
文摘Due to the rapid evolution of Advanced Persistent Threats(APTs)attacks,the emergence of new and rare attack samples,and even those never seen before,make it challenging for traditional rule-based detection methods to extract universal rules for effective detection.With the progress in techniques such as transfer learning and meta-learning,few-shot network attack detection has progressed.However,challenges in few-shot network attack detection arise from the inability of time sequence flow features to adapt to the fixed length input requirement of deep learning,difficulties in capturing rich information from original flow in the case of insufficient samples,and the challenge of high-level abstract representation.To address these challenges,a few-shot network attack detection based on NFHP(Network Flow Holographic Picture)-RN(ResNet)is proposed.Specifically,leveraging inherent properties of images such as translation invariance,rotation invariance,scale invariance,and illumination invariance,network attack traffic features and contextual relationships are intuitively represented in NFHP.In addition,an improved RN network model is employed for high-level abstract feature extraction,ensuring that the extracted high-level abstract features maintain the detailed characteristics of the original traffic behavior,regardless of changes in background traffic.Finally,a meta-learning model based on the self-attention mechanism is constructed,achieving the detection of novel APT few-shot network attacks through the empirical generalization of high-level abstract feature representations of known-class network attack behaviors.Experimental results demonstrate that the proposed method can learn high-level abstract features of network attacks across different traffic detail granularities.Comparedwith state-of-the-artmethods,it achieves favorable accuracy,precision,recall,and F1 scores for the identification of unknown-class network attacks through cross-validation onmultiple datasets.
