An improved cycle-consistent generative adversarial network(CycleGAN) method for defect data augmentation based on feature fusion and self attention residual module is proposed to address the insufficiency of defect s...An improved cycle-consistent generative adversarial network(CycleGAN) method for defect data augmentation based on feature fusion and self attention residual module is proposed to address the insufficiency of defect sample data for light guide plate(LGP) in production,as well as the problem of minor defects.Two optimizations are made to the generator of CycleGAN:fusion of low resolution features obtained from partial up-sampling and down-sampling with high-resolution features,combination of self attention mechanism with residual network structure to replace the original residual module.Qualitative and quantitative experiments were conducted to compare different data augmentation methods,and the results show that the defect images of the LGP generated by the improved network were more realistic,and the accuracy of the you only look once version 5(YOLOv5) detection network for the LGP was improved by 5.6%,proving the effectiveness and accuracy of the proposed method.展开更多
Surface quality has been one of the key factors influencing the ongoing improvement of the quality of steel. Therefore,it is urgent to provide methods for efficient supervision of surface defects. This paper first exp...Surface quality has been one of the key factors influencing the ongoing improvement of the quality of steel. Therefore,it is urgent to provide methods for efficient supervision of surface defects. This paper first expressed the main problems existing in defect management and then focused on constructing a data platform of surface defect management using a multidimensional database. Finally, some onqine applications of the platform at Baosteel were demonstrated. Results show that the constructed multidimensional database provides more structured defect data, and thus it is suitable for swift and multi-angle analysis of the defect data.展开更多
提出一种(γ,l-p,k)-匿名模型,模型引入了信息熵作为计算缺损数据的属性距离,通过敏感属性度高低分为不同的敏感级别,并设置相应的权重ω值,同时让等价类元组的不同敏感级别个数满足阈值γ。接着根据模型设计了加权信息熵匿名算法(Weigh...提出一种(γ,l-p,k)-匿名模型,模型引入了信息熵作为计算缺损数据的属性距离,通过敏感属性度高低分为不同的敏感级别,并设置相应的权重ω值,同时让等价类元组的不同敏感级别个数满足阈值γ。接着根据模型设计了加权信息熵匿名算法(Weighted Information Entropy Anonymous Algorithm based on Defect-Sensitive Attributes,WISA^(*))对缺损型数据集进行匿名化。实验结果表明,该算法不仅可以减少等价类信息损失,同时提高了敏感属性的多样性,从而降低了数据隐私泄露风险且复杂度较低。展开更多
电力设备运维过程中积累了大量缺陷图像与文本数据,这些数据对指导电力设备故障诊断及维护决策具有重要意义。针对现有电力设备缺陷分类任务中数据形式单一、融合层次浅、数据质量差等问题,该文提出了一种基于改进注意力机制和对比学习...电力设备运维过程中积累了大量缺陷图像与文本数据,这些数据对指导电力设备故障诊断及维护决策具有重要意义。针对现有电力设备缺陷分类任务中数据形式单一、融合层次浅、数据质量差等问题,该文提出了一种基于改进注意力机制和对比学习的图文融合分类方法(image-text fusion classification method based on improved attention mechanism and contrastive learning,IAC-ITFusion)。首先,该方法设计了一种双循环跨模态注意力机制(dual-cycle cross-modal attention,DCCA),用于捕捉图文数据映射关系的同时整合特征信息。其次,基于对比学习的思想,提出了一种注意力引导损失函数,用于调控DCCA机制的学习方向,使其聚焦于正确的特征信息,实现图文数据特征的有效融合。最后,针对电力线、变电站设备缺陷图文融合分类任务进行实验验证,结果显示所提方法准确率分别达到98.48%和98.57%,证明了该方法在电力设备缺陷图文融合分类任务上的有效性,对于推动电力设备运维智能化发展具有重要意义。展开更多
基金supported by the Jiangsu Province IUR Cooperation Project (No.BY2021258)the Wuxi Science and Technology Development Fund Project (No.G20212028)。
文摘An improved cycle-consistent generative adversarial network(CycleGAN) method for defect data augmentation based on feature fusion and self attention residual module is proposed to address the insufficiency of defect sample data for light guide plate(LGP) in production,as well as the problem of minor defects.Two optimizations are made to the generator of CycleGAN:fusion of low resolution features obtained from partial up-sampling and down-sampling with high-resolution features,combination of self attention mechanism with residual network structure to replace the original residual module.Qualitative and quantitative experiments were conducted to compare different data augmentation methods,and the results show that the defect images of the LGP generated by the improved network were more realistic,and the accuracy of the you only look once version 5(YOLOv5) detection network for the LGP was improved by 5.6%,proving the effectiveness and accuracy of the proposed method.
文摘Surface quality has been one of the key factors influencing the ongoing improvement of the quality of steel. Therefore,it is urgent to provide methods for efficient supervision of surface defects. This paper first expressed the main problems existing in defect management and then focused on constructing a data platform of surface defect management using a multidimensional database. Finally, some onqine applications of the platform at Baosteel were demonstrated. Results show that the constructed multidimensional database provides more structured defect data, and thus it is suitable for swift and multi-angle analysis of the defect data.
文摘提出一种(γ,l-p,k)-匿名模型,模型引入了信息熵作为计算缺损数据的属性距离,通过敏感属性度高低分为不同的敏感级别,并设置相应的权重ω值,同时让等价类元组的不同敏感级别个数满足阈值γ。接着根据模型设计了加权信息熵匿名算法(Weighted Information Entropy Anonymous Algorithm based on Defect-Sensitive Attributes,WISA^(*))对缺损型数据集进行匿名化。实验结果表明,该算法不仅可以减少等价类信息损失,同时提高了敏感属性的多样性,从而降低了数据隐私泄露风险且复杂度较低。
文摘电力设备运维过程中积累了大量缺陷图像与文本数据,这些数据对指导电力设备故障诊断及维护决策具有重要意义。针对现有电力设备缺陷分类任务中数据形式单一、融合层次浅、数据质量差等问题,该文提出了一种基于改进注意力机制和对比学习的图文融合分类方法(image-text fusion classification method based on improved attention mechanism and contrastive learning,IAC-ITFusion)。首先,该方法设计了一种双循环跨模态注意力机制(dual-cycle cross-modal attention,DCCA),用于捕捉图文数据映射关系的同时整合特征信息。其次,基于对比学习的思想,提出了一种注意力引导损失函数,用于调控DCCA机制的学习方向,使其聚焦于正确的特征信息,实现图文数据特征的有效融合。最后,针对电力线、变电站设备缺陷图文融合分类任务进行实验验证,结果显示所提方法准确率分别达到98.48%和98.57%,证明了该方法在电力设备缺陷图文融合分类任务上的有效性,对于推动电力设备运维智能化发展具有重要意义。
