Convolutional neural networks(CNNs)have shown remarkable success across numerous tasks such as image classification,yet the theoretical understanding of their convergence remains underdeveloped compared to their empir...Convolutional neural networks(CNNs)have shown remarkable success across numerous tasks such as image classification,yet the theoretical understanding of their convergence remains underdeveloped compared to their empirical achievements.In this paper,the first filter learning framework with convergence-guaranteed learning laws for end-to-end learning of deep CNNs is proposed.Novel update laws with convergence analysis are formulated based on the mathematical representation of each layer in convolutional neural networks.The proposed learning laws enable concurrent updates of weights across all layers of the deep convolutional neural network and the analysis shows that the training errors converge to certain bounds which are dependent on the approximation errors.Case studies are conducted on benchmark datasets and the results show that the proposed concurrent filter learning framework guarantees the convergence and offers more consistent and reliable results during training with a trade-off in performance compared to stochastic gradient descent methods.This framework represents a significant step towards enhancing the reliability and effectiveness of deep convolutional neural network by developing a theoretical analysis which allows practical implementation of the learning laws with automatic tuning of the learning rate to guarantee the convergence during training.展开更多
Due to the strong background noise and the acquisition system noise,the useful characteristics are often difficult to be detected.To solve this problem,sparse coding captures a concise representation of the high-level...Due to the strong background noise and the acquisition system noise,the useful characteristics are often difficult to be detected.To solve this problem,sparse coding captures a concise representation of the high-level features in the signal using the underlying structure of the signal.Recently,an Online Convolutional Sparse Coding(OCSC)denoising algorithm has been proposed.However,it does not consider the structural characteristics of the signal,the sparsity of each iteration is not enough.Therefore,a threshold shrinkage algorithm considering neighborhood sparsity is proposed,and a training strategy from loose to tight is developed to further improve the denoising performance of the algorithm,called Variable Threshold Neighborhood Online Convolution Sparse Coding(VTNOCSC).By embedding the structural sparse threshold shrinkage operator into the process of solving the sparse coefficient and gradually approaching the optimal noise separation point in the training,the signal denoising performance of the algorithm is greatly improved.VTNOCSC is used to process the actual bearing fault signal,the noise interference is successfully reduced and the interest features are more evident.Compared with other existing methods,VTNOCSC has better denoising performance.展开更多
The issue related to the risk of identity impersonation, where one person can be replaced by another in online exam surveillance systems, poses challenges. This study focuses on the effectiveness of detecting attempts...The issue related to the risk of identity impersonation, where one person can be replaced by another in online exam surveillance systems, poses challenges. This study focuses on the effectiveness of detecting attempts of identity impersonation through face substitution during online exams, with the aim of ensuring the integrity of assessments. The goal is to develop facial recognition algorithms capable of precisely detecting these impersonations, training them on a tailored database rather than biased generic data. An original database of student faces has been created. An algorithm leveraging advanced deep learning techniques such as depthwise separable convolution has been developed and evaluated on this database. We achieved very high levels of precision, reaching an accuracy rate of 98% in face detection and recognition.展开更多
In order to address the challenges posed by complex background interference,high miss-detection rates of micro-scale defects,and limited model deployment efficiency in photovoltaic(PV)module defect detection,this pape...In order to address the challenges posed by complex background interference,high miss-detection rates of micro-scale defects,and limited model deployment efficiency in photovoltaic(PV)module defect detection,this paper proposes an efficient detection framework based on an improved YOLOv11 architecture.First,a Re-parameterized Convolution(RepConv)module is integrated into the backbone to enhance the model’s sensitivity to fine-grained defects—such as micro-cracks and hot spots—while maintaining high inference efficiency.Second,a Multi-Scale Feature Fusion Convolutional Block Attention Mechanism(MSFF-CBAM)is designed to guide the network toward critical defect regions by jointly modeling channel-wise and spatial attention.This mechanism effectively strengthens the specificity and robustness of feature representations.Third,a lightweight Dynamic Sampling Module(DySample)is employed to replace conventional upsampling operations,thereby improving the localization accuracy of small-scale defect targets.Experimental evaluations conducted on the PVEL-AD dataset demonstrate that the proposed RMDYOLOv11 model surpasses the baseline YOLOv11 in terms of mean Average Precision(mAP)@0.5,Precision,and Recall,achieving respective improvements of 4.70%,1.51%,and 5.50%.The model also exhibits notable advantages in inference speed and model compactness.Further validation on the ELPV dataset confirms the model’s generalization capability,showing respective performance gains of 1.99%,2.28%,and 1.45%across the same metrics.Overall,the enhanced model significantly improves the accuracy of micro-defect identification on PV module surfaces,effectively reducing both false negatives and false positives.This advancement provides a robust and reliable technical foundation for automated PV module defect detection.展开更多
The injection molding process underpins modern mass manufacturing,yet surface defects like sink marks and flash cause quality issues,material waste and production delays.Traditional manual inspection is labor-intensiv...The injection molding process underpins modern mass manufacturing,yet surface defects like sink marks and flash cause quality issues,material waste and production delays.Traditional manual inspection is labor-intensive,costly and inconsistent,unfit for automated lines.This paper presents an online machine vision detection system for such defects,integrating high-resolution cameras and LED lighting to capture in-line images.Its pipeline includes preprocessing,hybrid feature extraction with traditional analysis and a CNN model,and real-time analysis via PLC for defect flagging and ejection.Trials on polymer components show 99.2%recognition accuracy,0.5%false positive rate and 180 parts/min processing speed,meeting cycle demands and boosting smart manufacturing quality control with lower operational costs.展开更多
针对基于知识图谱的推荐方法在教育领域应用主要集中在丰富课程特征表示上,对用户特征提取相对较少问题,提出一种基于堆叠LSTM(stacked long short-term memory)和知识图卷积网络MOOC课程推荐算法。通过课程内容信息和用户行为记录构建...针对基于知识图谱的推荐方法在教育领域应用主要集中在丰富课程特征表示上,对用户特征提取相对较少问题,提出一种基于堆叠LSTM(stacked long short-term memory)和知识图卷积网络MOOC课程推荐算法。通过课程内容信息和用户行为记录构建知识图谱,提供课程间语义关系。利用堆叠LSTM模型动态捕捉用户兴趣变化。堆叠LSTM通过多层隐藏单元对用户历史行为进行建模,提取更复杂的时间依赖特征,生成用户特征向量。这些向量与KGCN模型生成的用户特征向量进行加权融合,增强用户特征表示。结合图卷积网络(GCN)进一步探索课程之间潜在联系,预测用户对课程的评分。实验表明,该算法在AUC(area under curve)和F1指标上分别提高了2.32%和2.48%。该算法准确捕捉用户兴趣的动态变化,提升推荐性能。展开更多
基金supported by the Ministry of Education(MOE)Singapore,Academic Research Fund(AcRF)Tier 1(RG65/22)。
文摘Convolutional neural networks(CNNs)have shown remarkable success across numerous tasks such as image classification,yet the theoretical understanding of their convergence remains underdeveloped compared to their empirical achievements.In this paper,the first filter learning framework with convergence-guaranteed learning laws for end-to-end learning of deep CNNs is proposed.Novel update laws with convergence analysis are formulated based on the mathematical representation of each layer in convolutional neural networks.The proposed learning laws enable concurrent updates of weights across all layers of the deep convolutional neural network and the analysis shows that the training errors converge to certain bounds which are dependent on the approximation errors.Case studies are conducted on benchmark datasets and the results show that the proposed concurrent filter learning framework guarantees the convergence and offers more consistent and reliable results during training with a trade-off in performance compared to stochastic gradient descent methods.This framework represents a significant step towards enhancing the reliability and effectiveness of deep convolutional neural network by developing a theoretical analysis which allows practical implementation of the learning laws with automatic tuning of the learning rate to guarantee the convergence during training.
基金supported by the National Key Research and Development Program of China(No.2018YFB2003300)National Science and Technology Major Project,China(No.2017-IV-0008-0045)National Natural Science Foundation of China(No.51675262).
文摘Due to the strong background noise and the acquisition system noise,the useful characteristics are often difficult to be detected.To solve this problem,sparse coding captures a concise representation of the high-level features in the signal using the underlying structure of the signal.Recently,an Online Convolutional Sparse Coding(OCSC)denoising algorithm has been proposed.However,it does not consider the structural characteristics of the signal,the sparsity of each iteration is not enough.Therefore,a threshold shrinkage algorithm considering neighborhood sparsity is proposed,and a training strategy from loose to tight is developed to further improve the denoising performance of the algorithm,called Variable Threshold Neighborhood Online Convolution Sparse Coding(VTNOCSC).By embedding the structural sparse threshold shrinkage operator into the process of solving the sparse coefficient and gradually approaching the optimal noise separation point in the training,the signal denoising performance of the algorithm is greatly improved.VTNOCSC is used to process the actual bearing fault signal,the noise interference is successfully reduced and the interest features are more evident.Compared with other existing methods,VTNOCSC has better denoising performance.
文摘The issue related to the risk of identity impersonation, where one person can be replaced by another in online exam surveillance systems, poses challenges. This study focuses on the effectiveness of detecting attempts of identity impersonation through face substitution during online exams, with the aim of ensuring the integrity of assessments. The goal is to develop facial recognition algorithms capable of precisely detecting these impersonations, training them on a tailored database rather than biased generic data. An original database of student faces has been created. An algorithm leveraging advanced deep learning techniques such as depthwise separable convolution has been developed and evaluated on this database. We achieved very high levels of precision, reaching an accuracy rate of 98% in face detection and recognition.
基金supported by the Gansu Provincial Department of Education Industry Support Plan Project(2025CYZC-018).
文摘In order to address the challenges posed by complex background interference,high miss-detection rates of micro-scale defects,and limited model deployment efficiency in photovoltaic(PV)module defect detection,this paper proposes an efficient detection framework based on an improved YOLOv11 architecture.First,a Re-parameterized Convolution(RepConv)module is integrated into the backbone to enhance the model’s sensitivity to fine-grained defects—such as micro-cracks and hot spots—while maintaining high inference efficiency.Second,a Multi-Scale Feature Fusion Convolutional Block Attention Mechanism(MSFF-CBAM)is designed to guide the network toward critical defect regions by jointly modeling channel-wise and spatial attention.This mechanism effectively strengthens the specificity and robustness of feature representations.Third,a lightweight Dynamic Sampling Module(DySample)is employed to replace conventional upsampling operations,thereby improving the localization accuracy of small-scale defect targets.Experimental evaluations conducted on the PVEL-AD dataset demonstrate that the proposed RMDYOLOv11 model surpasses the baseline YOLOv11 in terms of mean Average Precision(mAP)@0.5,Precision,and Recall,achieving respective improvements of 4.70%,1.51%,and 5.50%.The model also exhibits notable advantages in inference speed and model compactness.Further validation on the ELPV dataset confirms the model’s generalization capability,showing respective performance gains of 1.99%,2.28%,and 1.45%across the same metrics.Overall,the enhanced model significantly improves the accuracy of micro-defect identification on PV module surfaces,effectively reducing both false negatives and false positives.This advancement provides a robust and reliable technical foundation for automated PV module defect detection.
文摘The injection molding process underpins modern mass manufacturing,yet surface defects like sink marks and flash cause quality issues,material waste and production delays.Traditional manual inspection is labor-intensive,costly and inconsistent,unfit for automated lines.This paper presents an online machine vision detection system for such defects,integrating high-resolution cameras and LED lighting to capture in-line images.Its pipeline includes preprocessing,hybrid feature extraction with traditional analysis and a CNN model,and real-time analysis via PLC for defect flagging and ejection.Trials on polymer components show 99.2%recognition accuracy,0.5%false positive rate and 180 parts/min processing speed,meeting cycle demands and boosting smart manufacturing quality control with lower operational costs.
文摘针对基于知识图谱的推荐方法在教育领域应用主要集中在丰富课程特征表示上,对用户特征提取相对较少问题,提出一种基于堆叠LSTM(stacked long short-term memory)和知识图卷积网络MOOC课程推荐算法。通过课程内容信息和用户行为记录构建知识图谱,提供课程间语义关系。利用堆叠LSTM模型动态捕捉用户兴趣变化。堆叠LSTM通过多层隐藏单元对用户历史行为进行建模,提取更复杂的时间依赖特征,生成用户特征向量。这些向量与KGCN模型生成的用户特征向量进行加权融合,增强用户特征表示。结合图卷积网络(GCN)进一步探索课程之间潜在联系,预测用户对课程的评分。实验表明,该算法在AUC(area under curve)和F1指标上分别提高了2.32%和2.48%。该算法准确捕捉用户兴趣的动态变化,提升推荐性能。
