Image captioning,a pivotal research area at the intersection of image understanding,artificial intelligence,and linguistics,aims to generate natural language descriptions for images.This paper proposes an efficient im...Image captioning,a pivotal research area at the intersection of image understanding,artificial intelligence,and linguistics,aims to generate natural language descriptions for images.This paper proposes an efficient image captioning model named Mob-IMWTC,which integrates improved wavelet convolution(IMWTC)with an enhanced MobileNet V3 architecture.The enhanced MobileNet V3 integrates a transformer encoder as its encoding module and a transformer decoder as its decoding module.This innovative neural network significantly reduces the memory space required and model training time,while maintaining a high level of accuracy in generating image descriptions.IMWTC facilitates large receptive fields without significantly increasing the number of parameters or computational overhead.The improvedMobileNet V3 model has its classifier removed,and simultaneously,it employs IMWTC layers to replace the original convolutional layers.This makes Mob-IMWTC exceptionally well-suited for deployment on lowresource devices.Experimental results,based on objective evaluation metrics such as BLEU,ROUGE,CIDEr,METEOR,and SPICE,demonstrate that Mob-IMWTC outperforms state-of-the-art models,including three CNN architectures(CNN-LSTM,CNN-Att-LSTM,CNN-Tran),two mainstream methods(LCM-Captioner,ClipCap),and our previous work(Mob-Tran).Subjective evaluations further validate the model’s superiority in terms of grammaticality,adequacy,logic,readability,and humanness.Mob-IMWTC offers a lightweight yet effective solution for image captioning,making it suitable for deployment on resource-constrained devices.展开更多
Future 6G communications will open up opportunities for innovative applications,including Cyber-Physical Systems,edge computing,supporting Industry 5.0,and digital agriculture.While automation is creating efficiencies...Future 6G communications will open up opportunities for innovative applications,including Cyber-Physical Systems,edge computing,supporting Industry 5.0,and digital agriculture.While automation is creating efficiencies,it can also create new cyber threats,such as vulnerabilities in trust and malicious node injection.Denialof-Service(DoS)attacks can stop many forms of operations by overwhelming networks and systems with data noise.Current anomaly detection methods require extensive software changes and only detect static threats.Data collection is important for being accurate,but it is often a slow,tedious,and sometimes inefficient process.This paper proposes a new wavelet transformassisted Bayesian deep learning based probabilistic(WT-BDLP)approach tomitigate malicious data injection attacks in 6G edge networks.The proposed approach combines outlier detection based on a Bayesian learning conditional variational autoencoder(Bay-LCVariAE)and traffic pattern analysis based on continuous wavelet transform(CWT).The Bay-LCVariAE framework allows for probabilistic modelling of generative features to facilitate capturing how features of interest change over time,spatially,and for recognition of anomalies.Similarly,CWT allows emphasizing the multi-resolution spectral analysis and permits temporally relevant frequency pattern recognition.Experimental testing showed that the flexibility of the Bayesian probabilistic framework offers a vast improvement in anomaly detection accuracy over existing methods,with a maximum accuracy of 98.21%recognizing anomalies.展开更多
Deep learning-based methods have shown great potential in intelligent bearing fault diagnosis.However,most existing approaches suffer from the scarcity of labeled data,which often results in insufficient robustness un...Deep learning-based methods have shown great potential in intelligent bearing fault diagnosis.However,most existing approaches suffer from the scarcity of labeled data,which often results in insufficient robustness under complex working conditions and a general lack of interpretability.To address these challenges,we propose a physics-informed multimodal fault diagnosis framework based on few-shot learning,which integrates a 2D timefrequency image encoder and a 1Dvibration signal encoder.Specifically,we embed prior knowledge ofmulti-resolution analysis from signal processing into the model by designing a Laplace Wavelet Convolution(LWC)module,which enhances interpretability since wavelet coefficients naturally correspond to specific frequency and temporal structures.To further balance the guidance of physical priors with the flexibility of learnable representations,we introduce a parametric multi-kernel wavelet that employs channel-wise dynamic attention to adaptively select relevant wavelet bases,thereby improving the feature expressiveness.Moreover,we develop a Mahalanobis-Prototype Joint Metric,which constructs more accurate and distribution-consistent decision boundaries under few-shot conditions.Comprehensive experiments on the Case Western Reserve University(CWRU)and Paderborn University(PU)bearing datasets demonstrate the superior effectiveness,robustness,and interpretability of the proposed approach compared with state-of-the-art baselines.展开更多
Roads inevitably have defects during use,which not only seriously affect their service life but also pose a hidden danger to traffic safety.Existing algorithms for detecting road defects are unsatisfactory in terms of...Roads inevitably have defects during use,which not only seriously affect their service life but also pose a hidden danger to traffic safety.Existing algorithms for detecting road defects are unsatisfactory in terms of accuracy and generalization,so this paper proposes an algorithm based on YOLOv11.The method embeds wavelet transform convolution(WTConv)into the backbone’s C3k2 module to enhance low-frequency feature extraction while avoiding parameter bloat.Secondly,a novel multi-scale fusion diffusion network(MFDN)architecture is designed for the neck to strengthen cross-scale feature interactions,boosting detection precision.In terms of model optimization,the traditional downsampling method is discarded,and the innovative Adown(adaptive downsampling)technique is adopted,which streamlines the parameter scales while effectively mitigating the information loss problem during downsampling.Finally,in this paper,we propose Wise-PIDIoU by combining WiseIoU and MPDIoU to minimize the negative impact of low-quality anchor frames and enhance the detection capability of the model.The experimental results indicate that the proposed algorithm achieves an average detection accuracy of 86.5%for mAP@50 on the RDD2022 dataset,which is 2%higher than the original algorithm while ensuring that the amount of computation is basically unchanged.The number of parameters is reduced by 17%,and the F1 score is improved by 3%,showing better detection performance than other algorithms when facing different types of defects.The excellent performance on embedded devices proves that the algorithm also has favorable application prospects in practical inspection.展开更多
To address challenges in wind turbine blade defect detection models,primarily due to insufficient feature extraction capabilities and the difficulty of deploying models on drone-type edge devices,this study proposes a...To address challenges in wind turbine blade defect detection models,primarily due to insufficient feature extraction capabilities and the difficulty of deploying models on drone-type edge devices,this study proposes a wind turbine blade defect detection model,WtCS-YOLO11,that incorporates multiscale feature extraction and an attention mechanism.Firstly,the cross-stage partial with two kernels and a wavelet convolution module(C3k2_WTConv)is proposed by introducing wavelet convolution into the module.The cross-stage partial with two kernels(C3k2)module in the necking network is replaced with the C3k2_WTConv module to increase the model’s receptive field,enable multiscale feature extraction,and reduce computational parameter usage.Second,the convolutional block attention module(CBAM)is proposed and applied to the neck network,integrating channel and spatial attention,allowing the model to focus on essential features and enhance its ability to detect large targets.In addition,the model employs shape-aware intersection over union(Shape-IoU),which focuses on the shape and scale of bounding boxes,and combines the normalized Wasserstein distance(NWD)to calculate bounding box similarity,thereby improving the accuracy of bounding-box regression.In this study,a dataset for wind turbine blade defect detection was constructed covering six defect categories.The experimental results showed that the precision(P),recall(R),and mean average precision at the intersection over union threshold of 0.5(mAP50)for the WtCS-YOLO11 model were 84.4%,86.9%,and 89.7%,respectively.Compared to the baseline You Only Look Once 11(YOLO11)model,P,R,and mAP50 improved by 5.9%,2.5%,and 2.4%,respectively,with virtually no increase in computational complexity or parameter count.WtCS-YOLO11 improved the precision measurement accuracy.Its model size and computational complexity are suitable for deployment on edge devices,and it achieves high inference speed,meeting the application requirements for real-time wind turbine blade defect detection.展开更多
基金funded by National Social Science Fund of China,grant number 23BYY197.
文摘Image captioning,a pivotal research area at the intersection of image understanding,artificial intelligence,and linguistics,aims to generate natural language descriptions for images.This paper proposes an efficient image captioning model named Mob-IMWTC,which integrates improved wavelet convolution(IMWTC)with an enhanced MobileNet V3 architecture.The enhanced MobileNet V3 integrates a transformer encoder as its encoding module and a transformer decoder as its decoding module.This innovative neural network significantly reduces the memory space required and model training time,while maintaining a high level of accuracy in generating image descriptions.IMWTC facilitates large receptive fields without significantly increasing the number of parameters or computational overhead.The improvedMobileNet V3 model has its classifier removed,and simultaneously,it employs IMWTC layers to replace the original convolutional layers.This makes Mob-IMWTC exceptionally well-suited for deployment on lowresource devices.Experimental results,based on objective evaluation metrics such as BLEU,ROUGE,CIDEr,METEOR,and SPICE,demonstrate that Mob-IMWTC outperforms state-of-the-art models,including three CNN architectures(CNN-LSTM,CNN-Att-LSTM,CNN-Tran),two mainstream methods(LCM-Captioner,ClipCap),and our previous work(Mob-Tran).Subjective evaluations further validate the model’s superiority in terms of grammaticality,adequacy,logic,readability,and humanness.Mob-IMWTC offers a lightweight yet effective solution for image captioning,making it suitable for deployment on resource-constrained devices.
文摘Future 6G communications will open up opportunities for innovative applications,including Cyber-Physical Systems,edge computing,supporting Industry 5.0,and digital agriculture.While automation is creating efficiencies,it can also create new cyber threats,such as vulnerabilities in trust and malicious node injection.Denialof-Service(DoS)attacks can stop many forms of operations by overwhelming networks and systems with data noise.Current anomaly detection methods require extensive software changes and only detect static threats.Data collection is important for being accurate,but it is often a slow,tedious,and sometimes inefficient process.This paper proposes a new wavelet transformassisted Bayesian deep learning based probabilistic(WT-BDLP)approach tomitigate malicious data injection attacks in 6G edge networks.The proposed approach combines outlier detection based on a Bayesian learning conditional variational autoencoder(Bay-LCVariAE)and traffic pattern analysis based on continuous wavelet transform(CWT).The Bay-LCVariAE framework allows for probabilistic modelling of generative features to facilitate capturing how features of interest change over time,spatially,and for recognition of anomalies.Similarly,CWT allows emphasizing the multi-resolution spectral analysis and permits temporally relevant frequency pattern recognition.Experimental testing showed that the flexibility of the Bayesian probabilistic framework offers a vast improvement in anomaly detection accuracy over existing methods,with a maximum accuracy of 98.21%recognizing anomalies.
文摘Deep learning-based methods have shown great potential in intelligent bearing fault diagnosis.However,most existing approaches suffer from the scarcity of labeled data,which often results in insufficient robustness under complex working conditions and a general lack of interpretability.To address these challenges,we propose a physics-informed multimodal fault diagnosis framework based on few-shot learning,which integrates a 2D timefrequency image encoder and a 1Dvibration signal encoder.Specifically,we embed prior knowledge ofmulti-resolution analysis from signal processing into the model by designing a Laplace Wavelet Convolution(LWC)module,which enhances interpretability since wavelet coefficients naturally correspond to specific frequency and temporal structures.To further balance the guidance of physical priors with the flexibility of learnable representations,we introduce a parametric multi-kernel wavelet that employs channel-wise dynamic attention to adaptively select relevant wavelet bases,thereby improving the feature expressiveness.Moreover,we develop a Mahalanobis-Prototype Joint Metric,which constructs more accurate and distribution-consistent decision boundaries under few-shot conditions.Comprehensive experiments on the Case Western Reserve University(CWRU)and Paderborn University(PU)bearing datasets demonstrate the superior effectiveness,robustness,and interpretability of the proposed approach compared with state-of-the-art baselines.
文摘Roads inevitably have defects during use,which not only seriously affect their service life but also pose a hidden danger to traffic safety.Existing algorithms for detecting road defects are unsatisfactory in terms of accuracy and generalization,so this paper proposes an algorithm based on YOLOv11.The method embeds wavelet transform convolution(WTConv)into the backbone’s C3k2 module to enhance low-frequency feature extraction while avoiding parameter bloat.Secondly,a novel multi-scale fusion diffusion network(MFDN)architecture is designed for the neck to strengthen cross-scale feature interactions,boosting detection precision.In terms of model optimization,the traditional downsampling method is discarded,and the innovative Adown(adaptive downsampling)technique is adopted,which streamlines the parameter scales while effectively mitigating the information loss problem during downsampling.Finally,in this paper,we propose Wise-PIDIoU by combining WiseIoU and MPDIoU to minimize the negative impact of low-quality anchor frames and enhance the detection capability of the model.The experimental results indicate that the proposed algorithm achieves an average detection accuracy of 86.5%for mAP@50 on the RDD2022 dataset,which is 2%higher than the original algorithm while ensuring that the amount of computation is basically unchanged.The number of parameters is reduced by 17%,and the F1 score is improved by 3%,showing better detection performance than other algorithms when facing different types of defects.The excellent performance on embedded devices proves that the algorithm also has favorable application prospects in practical inspection.
基金supported in part by the Science and Technology Research Project of Henan Province under grants 242102240040 and 222102210087in part by the Training Plan for Young Backbone Teachers at Undergraduate Universities in Henan Province under grant 2024GGJS155.
文摘To address challenges in wind turbine blade defect detection models,primarily due to insufficient feature extraction capabilities and the difficulty of deploying models on drone-type edge devices,this study proposes a wind turbine blade defect detection model,WtCS-YOLO11,that incorporates multiscale feature extraction and an attention mechanism.Firstly,the cross-stage partial with two kernels and a wavelet convolution module(C3k2_WTConv)is proposed by introducing wavelet convolution into the module.The cross-stage partial with two kernels(C3k2)module in the necking network is replaced with the C3k2_WTConv module to increase the model’s receptive field,enable multiscale feature extraction,and reduce computational parameter usage.Second,the convolutional block attention module(CBAM)is proposed and applied to the neck network,integrating channel and spatial attention,allowing the model to focus on essential features and enhance its ability to detect large targets.In addition,the model employs shape-aware intersection over union(Shape-IoU),which focuses on the shape and scale of bounding boxes,and combines the normalized Wasserstein distance(NWD)to calculate bounding box similarity,thereby improving the accuracy of bounding-box regression.In this study,a dataset for wind turbine blade defect detection was constructed covering six defect categories.The experimental results showed that the precision(P),recall(R),and mean average precision at the intersection over union threshold of 0.5(mAP50)for the WtCS-YOLO11 model were 84.4%,86.9%,and 89.7%,respectively.Compared to the baseline You Only Look Once 11(YOLO11)model,P,R,and mAP50 improved by 5.9%,2.5%,and 2.4%,respectively,with virtually no increase in computational complexity or parameter count.WtCS-YOLO11 improved the precision measurement accuracy.Its model size and computational complexity are suitable for deployment on edge devices,and it achieves high inference speed,meeting the application requirements for real-time wind turbine blade defect detection.
