Zone plates play an important role in X-ray and optics fileds.In this work,we proposed a generalized Fibonacci zone plates(GFiZP)is proposed,which enables the manipulation of focusing properties for X-ray and light.Co...Zone plates play an important role in X-ray and optics fileds.In this work,we proposed a generalized Fibonacci zone plates(GFiZP)is proposed,which enables the manipulation of focusing properties for X-ray and light.Compared to conventional Fibonacci zone plates(FiZPs)and Cantor-type fractal zone plates(FraZPs),this device offers additional degrees of freedom for engineering focusing performance.In comparison with FraZPs,GFiZPs are constructed with inflation rule instead of extraction rule,thus avoiding sparse structures at high recursion orders,the sparse structures would lead to issues in fabrication.Notably,numerical simulations demonstrate that,similar to FraZPs,GFiZPs retain two critical functionalities:multi-focal capability and self-similar intensity distribution.Furthermore,we conducted a detailed investigation into the axial irradiance profiles of GFiZPs with varying structural parameters.Given these favorable properties,this novel zone plate design holds great potential for expanded applications in areas such as optical manufacturing and optical manipulation of microparticles.展开更多
对抗样本攻击是识别网络面临的主要安全威胁之一。针对对抗样本检测过程中由分类边界模糊导致识别准确率低及需大量对抗样本参与训练导致模型收敛速率慢等问题,本文提出一种联合图像重构技术和图像生成技术实现的对抗样本检测方法。首先...对抗样本攻击是识别网络面临的主要安全威胁之一。针对对抗样本检测过程中由分类边界模糊导致识别准确率低及需大量对抗样本参与训练导致模型收敛速率慢等问题,本文提出一种联合图像重构技术和图像生成技术实现的对抗样本检测方法。首先,设计由卷积层和Swin-Transformer联合实现的图像重构网络,还原图像的语义信息并消除对抗性扰动;然后,利用条件生成式对抗网络依据标签信息生成对应类别图像;最后,将重构图像和生成图像送至卷积识别网络,依据分类结果一致性判断是否为对抗样本。该方法将对抗样本检测问题转化为图像分类问题,无需对抗样本参与训练,无需先验地了解攻击者的攻击类型和被攻击模型的结构和参数即可直接检测对抗样本。在VGG-16、Res Net-18、Goog Le Net分类网络和MNIST、GTSRB数据集上的实验结果表明,该检测方法相较于其他经典检测方法,平均识别准确率提升了4.75%~22.86%,F1分数提升了3.40%~13.64%,证明了其优越性。展开更多
文摘Zone plates play an important role in X-ray and optics fileds.In this work,we proposed a generalized Fibonacci zone plates(GFiZP)is proposed,which enables the manipulation of focusing properties for X-ray and light.Compared to conventional Fibonacci zone plates(FiZPs)and Cantor-type fractal zone plates(FraZPs),this device offers additional degrees of freedom for engineering focusing performance.In comparison with FraZPs,GFiZPs are constructed with inflation rule instead of extraction rule,thus avoiding sparse structures at high recursion orders,the sparse structures would lead to issues in fabrication.Notably,numerical simulations demonstrate that,similar to FraZPs,GFiZPs retain two critical functionalities:multi-focal capability and self-similar intensity distribution.Furthermore,we conducted a detailed investigation into the axial irradiance profiles of GFiZPs with varying structural parameters.Given these favorable properties,this novel zone plate design holds great potential for expanded applications in areas such as optical manufacturing and optical manipulation of microparticles.
文摘对抗样本攻击是识别网络面临的主要安全威胁之一。针对对抗样本检测过程中由分类边界模糊导致识别准确率低及需大量对抗样本参与训练导致模型收敛速率慢等问题,本文提出一种联合图像重构技术和图像生成技术实现的对抗样本检测方法。首先,设计由卷积层和Swin-Transformer联合实现的图像重构网络,还原图像的语义信息并消除对抗性扰动;然后,利用条件生成式对抗网络依据标签信息生成对应类别图像;最后,将重构图像和生成图像送至卷积识别网络,依据分类结果一致性判断是否为对抗样本。该方法将对抗样本检测问题转化为图像分类问题,无需对抗样本参与训练,无需先验地了解攻击者的攻击类型和被攻击模型的结构和参数即可直接检测对抗样本。在VGG-16、Res Net-18、Goog Le Net分类网络和MNIST、GTSRB数据集上的实验结果表明,该检测方法相较于其他经典检测方法,平均识别准确率提升了4.75%~22.86%,F1分数提升了3.40%~13.64%,证明了其优越性。
