Background Digital twins are virtual representations of devices and processes that capture the physical properties of the environment and operational algorithms/techniques in the context of medical devices and tech-no...Background Digital twins are virtual representations of devices and processes that capture the physical properties of the environment and operational algorithms/techniques in the context of medical devices and tech-nologies.Digital twins may allow healthcare organizations to determine methods of improving medical processes,enhancing patient experience,lowering operating expenses,and extending the value of care.During the present COVID-19 pandemic,various medical devices,such as X-rays and CT scan machines and processes,are constantly being used to collect and analyze medical images.When collecting and processing an extensive volume of data in the form of images,machines and processes sometimes suffer from system failures,creating critical issues for hospitals and patients.Methods To address this,we introduce a digital-twin-based smart healthcare system in-tegrated with medical devices to collect information regarding the current health condition,configuration,and maintenance history of the device/machine/system.Furthermore,medical images,that is,X-rays,are analyzed by using a deep-learning model to detect the infection of COVID-19.The designed system is based on the cascade recurrent convolution neural network(RCNN)architecture.In this architecture,the detector stages are deeper and more sequentially selective against small and close false positives.This architecture is a multi-stage extension of the RCNN model and sequentially trained using the output of one stage for training the other.At each stage,the bounding boxes are adjusted to locate a suitable value of the nearest false positives during the training of the different stages.In this manner,the arrangement of detectors is adjusted to increase the intersection over union,overcoming the problem of overfitting.We train the model by using X-ray images as the model was previously trained on another dataset.Results The developed system achieves good accuracy during the detection phase of COVID-19.The experimental outcomes reveal the efficiency of the detection architecture,which yields a mean average precision rate of 0.94.展开更多
Collaborative Robotics is one of the high-interest research topics in the area of academia and industry.It has been progressively utilized in numerous applications,particularly in intelligent surveillance systems.It a...Collaborative Robotics is one of the high-interest research topics in the area of academia and industry.It has been progressively utilized in numerous applications,particularly in intelligent surveillance systems.It allows the deployment of smart cameras or optical sensors with computer vision techniques,which may serve in several object detection and tracking tasks.These tasks have been considered challenging and high-level perceptual problems,frequently dominated by relative information about the environment,where main concerns such as occlusion,illumination,background,object deformation,and object class variations are commonplace.In order to show the importance of top view surveillance,a collaborative robotics framework has been presented.It can assist in the detection and tracking of multiple objects in top view surveillance.The framework consists of a smart robotic camera embedded with the visual processing unit.The existing pre-trained deep learning models named SSD and YOLO has been adopted for object detection and localization.The detection models are further combined with different tracking algorithms,including GOTURN,MEDIANFLOW,TLD,KCF,MIL,and BOOSTING.These algorithms,along with detection models,help to track and predict the trajectories of detected objects.The pre-trained models are employed;therefore,the generalization performance is also investigated through testing the models on various sequences of top view data set.The detection models achieved maximum True Detection Rate 93%to 90%with a maximum 0.6%False Detection Rate.The tracking results of different algorithms are nearly identical,with tracking accuracy ranging from 90%to 94%.Furthermore,a discussion has been carried out on output results along with future guidelines.展开更多
The maturity of 5G technology has enabled crowd-sensing services to collect multimedia data over wireless network,so it has promoted the applications of crowd-sensing services in different fields,but also brings more ...The maturity of 5G technology has enabled crowd-sensing services to collect multimedia data over wireless network,so it has promoted the applications of crowd-sensing services in different fields,but also brings more privacy security challenges,the most commom which is privacy leakage.As a privacy protection technology combining data integrity check and identity anonymity,ring signature is widely used in the field of privacy protection.However,introducing signature technology leads to additional signature verification overhead.In the scenario of crowd-sensing,the existing signature schemes have low efficiency in multi-signature verification.Therefore,it is necessary to design an efficient multi-signature verification scheme while ensuring security.In this paper,a batch-verifiable signature scheme is proposed based on the crowd-sensing background,which supports the sensing platform to verify the uploaded multiple signature data efficiently,so as to overcoming the defects of the traditional signature scheme in multi-signature verification.In our proposal,a method for linking homologous data was presented,which was valuable for incentive mechanism and data analysis.Simulation results showed that the proposed scheme has good performance in terms of security and efficiency in crowd-sensing applications with a large number of users and data.展开更多
Electricity theft is one of the major issues in developing countries which is affecting their economy badly.Especially with the introduction of emerging technologies,this issue became more complicated.Though many new ...Electricity theft is one of the major issues in developing countries which is affecting their economy badly.Especially with the introduction of emerging technologies,this issue became more complicated.Though many new energy theft detection(ETD)techniques have been proposed by utilising different data mining(DM)techniques,state&network(S&N)based techniques,and game theory(GT)techniques.Here,a detailed survey is presented where many state-of-the-art ETD techniques are studied and analysed for their strengths and limitations.Three levels of taxonomy are presented to classify state-of-the-art ETD techniques.Different types and ways of energy theft and their consequences are studied and summarised and different parameters to benchmark the performance of proposed techniques are extracted from literature.The challenges of different ETD techniques and their mitigation are suggested for future work.It is observed that the literature on ETD lacks knowledge management techniques that can be more effective,not only for ETD but also for theft tracking.This can help in the prevention of energy theft,in the future,as well as for ETD.展开更多
The inability of traditional privacy-preserving models to protect multiple datasets based on sensitive attributes has prompted researchers to propose models such as SLOMS,SLAMSA,(p,k)-Angelization,and(p,l)-Angelizatio...The inability of traditional privacy-preserving models to protect multiple datasets based on sensitive attributes has prompted researchers to propose models such as SLOMS,SLAMSA,(p,k)-Angelization,and(p,l)-Angelization,but these were found to be insufficient in terms of robust privacy and performance.(p,l)-Angelization was successful against different privacy disclosures,but it was not efficient.To the best of our knowledge,no robust privacy model based on fuzzy logic has been proposed to protect the privacy of sensitive attributes with multiple records.In this paper,we suggest an improved version of(p,l)-Angelization based on a hybrid AI approach and privacy-preserving approach like Generalization.Fuzz-classification(p,l)-Angel uses artificial intelligence based fuzzy logic for classification,a high-dimensional segmentation technique for segmenting quasi-identifiers and multiple sensitive attributes.We demonstrate the feasibility of the proposed solution by modelling and analyzing privacy violations using High-Level Petri Nets.The results of the experiment demonstrate that the proposed approach produces better results in terms of efficiency and utility.展开更多
Smart home devices are vulnerable to a variety of attacks.The matter gets more complicated when a number of devices collaborate to launch a colluding attack(e.g.,Distributed-Denial-of-Service(DDoS))in a network(e.g.,S...Smart home devices are vulnerable to a variety of attacks.The matter gets more complicated when a number of devices collaborate to launch a colluding attack(e.g.,Distributed-Denial-of-Service(DDoS))in a network(e.g.,Smart home).To handle these attacks,most studies have hitherto proposed authentication protocols that cannot necessarily be implemented in devices,especially during Device-to-Device(D2D)interactions.Tapping into the potential of Ethereum blockchain and smart contracts,this work proposes a lightweight authentication mechanism that enables safe D2D interactions in a smart home.The Ethereum blockchain enables the implementation of a decentralized prototype as well as a peer-to-peer distributed ledger system.The work also uses a single server queuing system model and the authentication mechanism to curtail DDoS attacks by controlling the number of service requests in the system.The simulation was conducted twenty times,each with varying number of devices chosen at random(ranging from 1 to 30).Each requester device sends an arbitrary request with a unique resource requirement at a time.This is done to measure the system's consistency across a variety of device capabilities.The experimental results show that the proposed protocol not only prevents colluding attacks,but also outperforms the benchmark protocols in terms of computational cost,message processing,and response times.展开更多
Quantifying the number of individuals in images or videos to estimate crowd density is a challenging yet crucial task with significant implications for fields such as urban planning and public safety.Crowd counting ha...Quantifying the number of individuals in images or videos to estimate crowd density is a challenging yet crucial task with significant implications for fields such as urban planning and public safety.Crowd counting has attracted considerable attention in the field of computer vision,leading to the development of numerous advanced models and methodologies.These approaches vary in terms of supervision techniques,network architectures,and model complexity.Currently,most crowd counting methods rely on fully supervised learning,which has proven to be effective.However,this approach presents challenges in real-world scenarios,where labeled data and ground-truth annotations are often scarce.As a result,there is an increasing need to explore unsupervised and semi-supervised methods to effectively address crowd counting tasks in practical applications.This paper offers a comprehensive review of crowd counting models,with a particular focus on semi-supervised and unsupervised approaches based on their supervision paradigms.We summarize and critically analyze the key methods in these two categories,highlighting their strengths and limitations.Furthermore,we provide a comparative analysis of prominent crowd counting methods using widely adopted benchmark datasets.We believe that this survey will offer valuable insights and guide future advancements in crowd counting technology.展开更多
The rapid convergence of Information and Communication Technologies(ICT),driven by advancements in 5G/6G networks,cloud computing,Artificial Intelligence(AI),and the Internet of Things(IoT),is reshaping modern digital...The rapid convergence of Information and Communication Technologies(ICT),driven by advancements in 5G/6G networks,cloud computing,Artificial Intelligence(AI),and the Internet of Things(IoT),is reshaping modern digital ecosystems.As massive,distributed data streams are generated across edge devices and network layers,there is a growing need for intelligent,privacy-preserving AI solutions that can operate efficiently at the network edge.Federated Learning(FL)enables decentralized model training without transferring sensitive data,addressing key challenges around privacy,bandwidth,and latency.Despite its benefits in enhancing efficiency,real-time analytics,and regulatory compliance,FL adoption faces challenges,including communication overhead,heterogeneity,security vulnerabilities,and limited edge resources.While recent studies have addressed these issues individually,the literature lacks a unified,cross-domain perspective that reflects the architectural complexity and application diversity of Convergence ICT.This systematic review offers a comprehensive,cross-domain examination of FL within converged ICT infrastructures.The central research question guiding this review is:How can FL be effectively integrated into Convergence ICT environments,and what are the main challenges in implementing FL in such environments,along with possible solutions?We begin with a foundational overview of FL concepts and classifications,followed by a detailed taxonomy of FL architectures,learning strategies,and privacy-preserving mechanisms.Through in-depth case studies,we analyse FL’s application across diverse verticals,including smart cities,healthcare,industrial automation,and autonomous systems.We further identify critical challenges—such as system and data heterogeneity,limited edge resources,and security vulnerabilities—and review state-of-the-art mitigation strategies,including edge-aware optimization,secure aggregation,and adaptive model updates.In addition,we explore emerging directions in FL research,such as energy-efficient learning,federated reinforcement learning,and integration with blockchain,quantum computing,and self-adaptive networks.This review not only synthesizes current literature but also proposes a forward-looking road map to support scalable,secure,and sustainable FL deployment in future ICT ecosystems.展开更多
With the rapid progress in data-driven approaches,artificial intelligence,and big data analytics technologies,utilizing electroencephalogram(EEG)signals for emotion analysis in the field of the Internet of Medical Thi...With the rapid progress in data-driven approaches,artificial intelligence,and big data analytics technologies,utilizing electroencephalogram(EEG)signals for emotion analysis in the field of the Internet of Medical Things can effectively assist in the diagnosis of specific diseases.While existing emotion analysis methods focus on the utilization of effective deep models for data-driven and big data analytics technology,they often struggle to extract long-range dependencies and accurately model local relationships within multi-channel EEG signals.In addition,the subjective scores of the subjects may not match the predefined emotional labels.To overcome these limitations,this paper proposes a new data-driven dynamic graph-embedded Transformer network(DGETN)that has emerged in different tasks of graph data mining for emotion analysis of EEG signals in the scene of IoMT.Firstly,we extract the frequency features differential entropy(DE)and use the linear dynamic system(LDS)method to alleviate the redundancy and noise information.Secondly,to effectively explore the long-range information and local modeling ability,a novel feature extraction module is designed by embedding the dynamic graph convolution operations in the Transformer encoder for mining the discriminant features of data.Moreover,the graph convolution operations can effectively exploit the spatial information between different channels.At last,we introduce the minimum category confusion(MCC)loss to alleviate the fuzziness of classification.We take two commonly used EEG sentiment analysis datasets as a study.The DGETN has achieved state-of-the-art accuracies of 99.38%on the SEED dataset,and accuracies of 99.24%and 98.85%for valence and arousal prediction on the DEAP dataset,respectively.展开更多
Shield tunnel lining is prone to water leakage,which may further bring about corrosion and structural damage to the walls,potentially leading to dangerous accidents.To avoid tedious and inefficient manual inspection,m...Shield tunnel lining is prone to water leakage,which may further bring about corrosion and structural damage to the walls,potentially leading to dangerous accidents.To avoid tedious and inefficient manual inspection,many projects use artificial intelligence(Al)to detect cracks and water leakage.A novel method for water leakage inspection in shield tunnel lining that utilizes deep learning is introduced in this paper.Our proposal includes a ConvNeXt-S backbone,deconvolutional-feature pyramid network(D-FPN),spatial attention module(SPAM).and a detection head.It can extract representative features of leaking areas to aid inspection processes.To further improve the model's robustness,we innovatively use an inversed low-light enhancement method to convert normally illuminated images to low light ones and introduce them into the training samples.Validation experiments are performed,achieving the average precision(AP)score of 56.8%,which outperforms previous work by a margin of 5.7%.Visualization illustrations also support our method's practical effectiveness.展开更多
文摘Background Digital twins are virtual representations of devices and processes that capture the physical properties of the environment and operational algorithms/techniques in the context of medical devices and tech-nologies.Digital twins may allow healthcare organizations to determine methods of improving medical processes,enhancing patient experience,lowering operating expenses,and extending the value of care.During the present COVID-19 pandemic,various medical devices,such as X-rays and CT scan machines and processes,are constantly being used to collect and analyze medical images.When collecting and processing an extensive volume of data in the form of images,machines and processes sometimes suffer from system failures,creating critical issues for hospitals and patients.Methods To address this,we introduce a digital-twin-based smart healthcare system in-tegrated with medical devices to collect information regarding the current health condition,configuration,and maintenance history of the device/machine/system.Furthermore,medical images,that is,X-rays,are analyzed by using a deep-learning model to detect the infection of COVID-19.The designed system is based on the cascade recurrent convolution neural network(RCNN)architecture.In this architecture,the detector stages are deeper and more sequentially selective against small and close false positives.This architecture is a multi-stage extension of the RCNN model and sequentially trained using the output of one stage for training the other.At each stage,the bounding boxes are adjusted to locate a suitable value of the nearest false positives during the training of the different stages.In this manner,the arrangement of detectors is adjusted to increase the intersection over union,overcoming the problem of overfitting.We train the model by using X-ray images as the model was previously trained on another dataset.Results The developed system achieves good accuracy during the detection phase of COVID-19.The experimental outcomes reveal the efficiency of the detection architecture,which yields a mean average precision rate of 0.94.
基金the Framework of International Cooperation Program managed by the National Research Foundation of Korea(2019K1A3A1A8011295711).
文摘Collaborative Robotics is one of the high-interest research topics in the area of academia and industry.It has been progressively utilized in numerous applications,particularly in intelligent surveillance systems.It allows the deployment of smart cameras or optical sensors with computer vision techniques,which may serve in several object detection and tracking tasks.These tasks have been considered challenging and high-level perceptual problems,frequently dominated by relative information about the environment,where main concerns such as occlusion,illumination,background,object deformation,and object class variations are commonplace.In order to show the importance of top view surveillance,a collaborative robotics framework has been presented.It can assist in the detection and tracking of multiple objects in top view surveillance.The framework consists of a smart robotic camera embedded with the visual processing unit.The existing pre-trained deep learning models named SSD and YOLO has been adopted for object detection and localization.The detection models are further combined with different tracking algorithms,including GOTURN,MEDIANFLOW,TLD,KCF,MIL,and BOOSTING.These algorithms,along with detection models,help to track and predict the trajectories of detected objects.The pre-trained models are employed;therefore,the generalization performance is also investigated through testing the models on various sequences of top view data set.The detection models achieved maximum True Detection Rate 93%to 90%with a maximum 0.6%False Detection Rate.The tracking results of different algorithms are nearly identical,with tracking accuracy ranging from 90%to 94%.Furthermore,a discussion has been carried out on output results along with future guidelines.
基金supported by National Natural Science Foundation of China under Grant No.61972360Shandong Provincial Natural Science Foundation of China under Grant Nos.ZR2020MF148,ZR2020QF108.
文摘The maturity of 5G technology has enabled crowd-sensing services to collect multimedia data over wireless network,so it has promoted the applications of crowd-sensing services in different fields,but also brings more privacy security challenges,the most commom which is privacy leakage.As a privacy protection technology combining data integrity check and identity anonymity,ring signature is widely used in the field of privacy protection.However,introducing signature technology leads to additional signature verification overhead.In the scenario of crowd-sensing,the existing signature schemes have low efficiency in multi-signature verification.Therefore,it is necessary to design an efficient multi-signature verification scheme while ensuring security.In this paper,a batch-verifiable signature scheme is proposed based on the crowd-sensing background,which supports the sensing platform to verify the uploaded multiple signature data efficiently,so as to overcoming the defects of the traditional signature scheme in multi-signature verification.In our proposal,a method for linking homologous data was presented,which was valuable for incentive mechanism and data analysis.Simulation results showed that the proposed scheme has good performance in terms of security and efficiency in crowd-sensing applications with a large number of users and data.
基金supported by the European Union’s Horizon 2020 Research and Innovation Programme under the Marie Sk?odowska-Curie Grant Agreement(801522)Science Foundation Ireland and co-funded by the European Regional Development Fund through the ADAPT Centre for Digital Content Technology(13/RC/2106_P2)。
文摘Electricity theft is one of the major issues in developing countries which is affecting their economy badly.Especially with the introduction of emerging technologies,this issue became more complicated.Though many new energy theft detection(ETD)techniques have been proposed by utilising different data mining(DM)techniques,state&network(S&N)based techniques,and game theory(GT)techniques.Here,a detailed survey is presented where many state-of-the-art ETD techniques are studied and analysed for their strengths and limitations.Three levels of taxonomy are presented to classify state-of-the-art ETD techniques.Different types and ways of energy theft and their consequences are studied and summarised and different parameters to benchmark the performance of proposed techniques are extracted from literature.The challenges of different ETD techniques and their mitigation are suggested for future work.It is observed that the literature on ETD lacks knowledge management techniques that can be more effective,not only for ETD but also for theft tracking.This can help in the prevention of energy theft,in the future,as well as for ETD.
文摘The inability of traditional privacy-preserving models to protect multiple datasets based on sensitive attributes has prompted researchers to propose models such as SLOMS,SLAMSA,(p,k)-Angelization,and(p,l)-Angelization,but these were found to be insufficient in terms of robust privacy and performance.(p,l)-Angelization was successful against different privacy disclosures,but it was not efficient.To the best of our knowledge,no robust privacy model based on fuzzy logic has been proposed to protect the privacy of sensitive attributes with multiple records.In this paper,we suggest an improved version of(p,l)-Angelization based on a hybrid AI approach and privacy-preserving approach like Generalization.Fuzz-classification(p,l)-Angel uses artificial intelligence based fuzzy logic for classification,a high-dimensional segmentation technique for segmenting quasi-identifiers and multiple sensitive attributes.We demonstrate the feasibility of the proposed solution by modelling and analyzing privacy violations using High-Level Petri Nets.The results of the experiment demonstrate that the proposed approach produces better results in terms of efficiency and utility.
文摘Smart home devices are vulnerable to a variety of attacks.The matter gets more complicated when a number of devices collaborate to launch a colluding attack(e.g.,Distributed-Denial-of-Service(DDoS))in a network(e.g.,Smart home).To handle these attacks,most studies have hitherto proposed authentication protocols that cannot necessarily be implemented in devices,especially during Device-to-Device(D2D)interactions.Tapping into the potential of Ethereum blockchain and smart contracts,this work proposes a lightweight authentication mechanism that enables safe D2D interactions in a smart home.The Ethereum blockchain enables the implementation of a decentralized prototype as well as a peer-to-peer distributed ledger system.The work also uses a single server queuing system model and the authentication mechanism to curtail DDoS attacks by controlling the number of service requests in the system.The simulation was conducted twenty times,each with varying number of devices chosen at random(ranging from 1 to 30).Each requester device sends an arbitrary request with a unique resource requirement at a time.This is done to measure the system's consistency across a variety of device capabilities.The experimental results show that the proposed protocol not only prevents colluding attacks,but also outperforms the benchmark protocols in terms of computational cost,message processing,and response times.
基金supported by Research Project Support Program for Excellence Institute(2022,ESL)in Incheon National University.
文摘Quantifying the number of individuals in images or videos to estimate crowd density is a challenging yet crucial task with significant implications for fields such as urban planning and public safety.Crowd counting has attracted considerable attention in the field of computer vision,leading to the development of numerous advanced models and methodologies.These approaches vary in terms of supervision techniques,network architectures,and model complexity.Currently,most crowd counting methods rely on fully supervised learning,which has proven to be effective.However,this approach presents challenges in real-world scenarios,where labeled data and ground-truth annotations are often scarce.As a result,there is an increasing need to explore unsupervised and semi-supervised methods to effectively address crowd counting tasks in practical applications.This paper offers a comprehensive review of crowd counting models,with a particular focus on semi-supervised and unsupervised approaches based on their supervision paradigms.We summarize and critically analyze the key methods in these two categories,highlighting their strengths and limitations.Furthermore,we provide a comparative analysis of prominent crowd counting methods using widely adopted benchmark datasets.We believe that this survey will offer valuable insights and guide future advancements in crowd counting technology.
文摘The rapid convergence of Information and Communication Technologies(ICT),driven by advancements in 5G/6G networks,cloud computing,Artificial Intelligence(AI),and the Internet of Things(IoT),is reshaping modern digital ecosystems.As massive,distributed data streams are generated across edge devices and network layers,there is a growing need for intelligent,privacy-preserving AI solutions that can operate efficiently at the network edge.Federated Learning(FL)enables decentralized model training without transferring sensitive data,addressing key challenges around privacy,bandwidth,and latency.Despite its benefits in enhancing efficiency,real-time analytics,and regulatory compliance,FL adoption faces challenges,including communication overhead,heterogeneity,security vulnerabilities,and limited edge resources.While recent studies have addressed these issues individually,the literature lacks a unified,cross-domain perspective that reflects the architectural complexity and application diversity of Convergence ICT.This systematic review offers a comprehensive,cross-domain examination of FL within converged ICT infrastructures.The central research question guiding this review is:How can FL be effectively integrated into Convergence ICT environments,and what are the main challenges in implementing FL in such environments,along with possible solutions?We begin with a foundational overview of FL concepts and classifications,followed by a detailed taxonomy of FL architectures,learning strategies,and privacy-preserving mechanisms.Through in-depth case studies,we analyse FL’s application across diverse verticals,including smart cities,healthcare,industrial automation,and autonomous systems.We further identify critical challenges—such as system and data heterogeneity,limited edge resources,and security vulnerabilities—and review state-of-the-art mitigation strategies,including edge-aware optimization,secure aggregation,and adaptive model updates.In addition,we explore emerging directions in FL research,such as energy-efficient learning,federated reinforcement learning,and integration with blockchain,quantum computing,and self-adaptive networks.This review not only synthesizes current literature but also proposes a forward-looking road map to support scalable,secure,and sustainable FL deployment in future ICT ecosystems.
文摘With the rapid progress in data-driven approaches,artificial intelligence,and big data analytics technologies,utilizing electroencephalogram(EEG)signals for emotion analysis in the field of the Internet of Medical Things can effectively assist in the diagnosis of specific diseases.While existing emotion analysis methods focus on the utilization of effective deep models for data-driven and big data analytics technology,they often struggle to extract long-range dependencies and accurately model local relationships within multi-channel EEG signals.In addition,the subjective scores of the subjects may not match the predefined emotional labels.To overcome these limitations,this paper proposes a new data-driven dynamic graph-embedded Transformer network(DGETN)that has emerged in different tasks of graph data mining for emotion analysis of EEG signals in the scene of IoMT.Firstly,we extract the frequency features differential entropy(DE)and use the linear dynamic system(LDS)method to alleviate the redundancy and noise information.Secondly,to effectively explore the long-range information and local modeling ability,a novel feature extraction module is designed by embedding the dynamic graph convolution operations in the Transformer encoder for mining the discriminant features of data.Moreover,the graph convolution operations can effectively exploit the spatial information between different channels.At last,we introduce the minimum category confusion(MCC)loss to alleviate the fuzziness of classification.We take two commonly used EEG sentiment analysis datasets as a study.The DGETN has achieved state-of-the-art accuracies of 99.38%on the SEED dataset,and accuracies of 99.24%and 98.85%for valence and arousal prediction on the DEAP dataset,respectively.
基金This work is funded by the National Natural Science Foundation of China(Grant Nos.62171114 and 52222810)the Fundamental Research Funds for the Central Universities(No.DUT22RC(3)099).
文摘Shield tunnel lining is prone to water leakage,which may further bring about corrosion and structural damage to the walls,potentially leading to dangerous accidents.To avoid tedious and inefficient manual inspection,many projects use artificial intelligence(Al)to detect cracks and water leakage.A novel method for water leakage inspection in shield tunnel lining that utilizes deep learning is introduced in this paper.Our proposal includes a ConvNeXt-S backbone,deconvolutional-feature pyramid network(D-FPN),spatial attention module(SPAM).and a detection head.It can extract representative features of leaking areas to aid inspection processes.To further improve the model's robustness,we innovatively use an inversed low-light enhancement method to convert normally illuminated images to low light ones and introduce them into the training samples.Validation experiments are performed,achieving the average precision(AP)score of 56.8%,which outperforms previous work by a margin of 5.7%.Visualization illustrations also support our method's practical effectiveness.
