针对天基短波红外图像中弱小目标易被云层、地表杂波淹没,且在低信杂比条件下检测困难的问题,提出一种融合安德森加速的自正则化加权稀疏模型(Self-Regularized Weighted Sparse,SRWS)与相对局部对比度(Relative Local Contrast Measure...针对天基短波红外图像中弱小目标易被云层、地表杂波淹没,且在低信杂比条件下检测困难的问题,提出一种融合安德森加速的自正则化加权稀疏模型(Self-Regularized Weighted Sparse,SRWS)与相对局部对比度(Relative Local Contrast Measure,RLCM)的改进检测方法。通过引入安德森加速机制,显著降低了背景估计的计算复杂度,利用背景残差图和RLCM实现了多尺度目标检测性。实验结果表明,本文算法在复杂背景下仍保持优异性能,接收者操作曲线下面积(Area Under Curve,AUC)最高达0.950,最低不低于0.842;信杂比增益(Signal-to-Clutter Ratio Gain,SCRG)显著优于红外图像块(Infrared Patch Image,IPI)、局部对比度法(Local Contrast Measure,LCM)等传统方法。本研究有效提升了天基短波红外弱小目标的检测精度与稳定性,为复杂背景下的遥感目标检测提供了可靠的解决方案。展开更多
Reflection and transmission(R/T)responses characterize the propagation and energy distribution of incident and reflected waves on both sides of an interface which is crucial for imaging,amplitude variation with offset...Reflection and transmission(R/T)responses characterize the propagation and energy distribution of incident and reflected waves on both sides of an interface which is crucial for imaging,amplitude variation with offset(AVO),and seismic inversion techniques.Subsurface media are typically characterized by anisotropy which can have a significant impact on the R/T response,even at small incident angles.Currently,anisotropic media problems including reflection,transmission,and inversion are generally discussed under a weak anisotropy assumption.However,this assumption is no longer valid in cases of large angles where anisotropy enhancement exacerbates the error of the conventional R/T coefficient approximation.An R/T coefficient approximation method for strong VTI media was proposed based on the assumption of weak-contrast of the media.In contrast to the conventional approach,which simplifies the phase velocity and polarization in an anisotropic background,the phase velocity and polarization at the weak-contrast interface of the elasticity and anisotropy parameters were approximated using a combination of the anisotropic background and perturbation terms.Specifically,a first-order approximation of the R/T coefficients for the VTI media characterized by elastic and anisotropic parameters was derived using Cramer's law to invert the anisotropic background matrix,avoiding the assumption of weak anisotropy.Subsequently,the exact solution of the Zoeppritz equations was used to correct the isotropic part,improving the accuracy of the R/T coefficients at interfaces with high-velocity contrast.Modeling tests on four classes of typical interfaces showed that the derived equations can be degraded to the Aki approximation in isotropic media,while exhibiting high accuracy in strong VTI media.Uncertainty analyses showed that a linear approximation that facilitates seismic inversion can be obtained by taking the S-to P-velocity ratio and anisotropy parameters in the coefficient terms a priori.展开更多
近年来,小样本图异常检测在各个领域中引起了广泛的研究兴趣,其旨在在少量有标记训练节点(支持集)的引导下去检测出大量无标记测试节点(查询集)中的异常行为。然而,现有的小样本图异常检测算法通常假设其可以从具有大量有标记节点的训...近年来,小样本图异常检测在各个领域中引起了广泛的研究兴趣,其旨在在少量有标记训练节点(支持集)的引导下去检测出大量无标记测试节点(查询集)中的异常行为。然而,现有的小样本图异常检测算法通常假设其可以从具有大量有标记节点的训练任务(元训练任务)中学习,从而有效地推广到具有少量标记节点的测试任务(元测试任务),这一假设并不符合真实世界的应用条件。在实际应用中,用于小样本图异常检测训练的元训练任务通常只包含极其有限的有标记节点,其标签占比通常不超过0.1%,甚至更低。由于元训练和元测试任务之间存在的巨大任务差异,现有的小样本图异常检测算法很容易出现模型的过拟合问题。除此之外,现有的小样本图异常检测算法仅利用节点间的一阶邻域(局部结构信息)来学习节点的低维特征嵌入,反而忽略了节点间的长距离依赖关系(全局结构信息),进而导致学习到的低维特征嵌入的不准确性和失真问题。针对上述挑战,本文提出了极其弱监督场景下的小样本图异常检测算法——EWSFSGAD。具体来说,该方法首先提出了一个简单且有效的图神经网络框架——GLN(Global and Local Network),其能够同时有效地利用节点间的全局和局部结构信息,并进一步引入注意力机制实现节点间的信息交互,从而更加有效地学习节点鲁棒的低维特征嵌入;该方法还引入了图对比学习中的自监督重建损失,使得节点原始视图与其增强视图之间低维特征嵌入的互信息尽可能一致,为EWS-FSGAD模型的优化提供更多有效的自监督信息,进而提升模型的泛化性;为了提升模型在真实场景中小样本图异常检测任务的快速适应性,该方法引入跨网络元学习训练机制,从多个辅助网络学习可迁移元知识,为模型提供良好的参数初始化,从而能够通过在仅有很少甚至一个标记节点的目标网络上进行微调并有效泛化。在三个真实世界的数据集(Flickr、PubMed、Yelp)上的大量实验结果表明,本文所提方法的性能明显优于现有的图异常检测算法。特别是在PubMed数据集上,AUC-PR提升了28.8%~35.4%。这些实验结果强有力地证明了在极其有限标记的元训练任务引导下,本文所提方法能够更好地学习到异常节点本质特征,从而提升小样本图异常检测任务的有效性。展开更多
文摘针对天基短波红外图像中弱小目标易被云层、地表杂波淹没,且在低信杂比条件下检测困难的问题,提出一种融合安德森加速的自正则化加权稀疏模型(Self-Regularized Weighted Sparse,SRWS)与相对局部对比度(Relative Local Contrast Measure,RLCM)的改进检测方法。通过引入安德森加速机制,显著降低了背景估计的计算复杂度,利用背景残差图和RLCM实现了多尺度目标检测性。实验结果表明,本文算法在复杂背景下仍保持优异性能,接收者操作曲线下面积(Area Under Curve,AUC)最高达0.950,最低不低于0.842;信杂比增益(Signal-to-Clutter Ratio Gain,SCRG)显著优于红外图像块(Infrared Patch Image,IPI)、局部对比度法(Local Contrast Measure,LCM)等传统方法。本研究有效提升了天基短波红外弱小目标的检测精度与稳定性,为复杂背景下的遥感目标检测提供了可靠的解决方案。
基金supported by the National Natural Science Foundation of China(Grant Nos.42030103,42274157,41974119)the Marine S&T Fund of Shandong Province for Pilot National Laboratory for Marine Science and Technology(Grant No.2021QNLM020001-6)the Natural Science Foundation of Shangdong Province(Grant No.ZR2022MD092)。
文摘Reflection and transmission(R/T)responses characterize the propagation and energy distribution of incident and reflected waves on both sides of an interface which is crucial for imaging,amplitude variation with offset(AVO),and seismic inversion techniques.Subsurface media are typically characterized by anisotropy which can have a significant impact on the R/T response,even at small incident angles.Currently,anisotropic media problems including reflection,transmission,and inversion are generally discussed under a weak anisotropy assumption.However,this assumption is no longer valid in cases of large angles where anisotropy enhancement exacerbates the error of the conventional R/T coefficient approximation.An R/T coefficient approximation method for strong VTI media was proposed based on the assumption of weak-contrast of the media.In contrast to the conventional approach,which simplifies the phase velocity and polarization in an anisotropic background,the phase velocity and polarization at the weak-contrast interface of the elasticity and anisotropy parameters were approximated using a combination of the anisotropic background and perturbation terms.Specifically,a first-order approximation of the R/T coefficients for the VTI media characterized by elastic and anisotropic parameters was derived using Cramer's law to invert the anisotropic background matrix,avoiding the assumption of weak anisotropy.Subsequently,the exact solution of the Zoeppritz equations was used to correct the isotropic part,improving the accuracy of the R/T coefficients at interfaces with high-velocity contrast.Modeling tests on four classes of typical interfaces showed that the derived equations can be degraded to the Aki approximation in isotropic media,while exhibiting high accuracy in strong VTI media.Uncertainty analyses showed that a linear approximation that facilitates seismic inversion can be obtained by taking the S-to P-velocity ratio and anisotropy parameters in the coefficient terms a priori.
文摘近年来,小样本图异常检测在各个领域中引起了广泛的研究兴趣,其旨在在少量有标记训练节点(支持集)的引导下去检测出大量无标记测试节点(查询集)中的异常行为。然而,现有的小样本图异常检测算法通常假设其可以从具有大量有标记节点的训练任务(元训练任务)中学习,从而有效地推广到具有少量标记节点的测试任务(元测试任务),这一假设并不符合真实世界的应用条件。在实际应用中,用于小样本图异常检测训练的元训练任务通常只包含极其有限的有标记节点,其标签占比通常不超过0.1%,甚至更低。由于元训练和元测试任务之间存在的巨大任务差异,现有的小样本图异常检测算法很容易出现模型的过拟合问题。除此之外,现有的小样本图异常检测算法仅利用节点间的一阶邻域(局部结构信息)来学习节点的低维特征嵌入,反而忽略了节点间的长距离依赖关系(全局结构信息),进而导致学习到的低维特征嵌入的不准确性和失真问题。针对上述挑战,本文提出了极其弱监督场景下的小样本图异常检测算法——EWSFSGAD。具体来说,该方法首先提出了一个简单且有效的图神经网络框架——GLN(Global and Local Network),其能够同时有效地利用节点间的全局和局部结构信息,并进一步引入注意力机制实现节点间的信息交互,从而更加有效地学习节点鲁棒的低维特征嵌入;该方法还引入了图对比学习中的自监督重建损失,使得节点原始视图与其增强视图之间低维特征嵌入的互信息尽可能一致,为EWS-FSGAD模型的优化提供更多有效的自监督信息,进而提升模型的泛化性;为了提升模型在真实场景中小样本图异常检测任务的快速适应性,该方法引入跨网络元学习训练机制,从多个辅助网络学习可迁移元知识,为模型提供良好的参数初始化,从而能够通过在仅有很少甚至一个标记节点的目标网络上进行微调并有效泛化。在三个真实世界的数据集(Flickr、PubMed、Yelp)上的大量实验结果表明,本文所提方法的性能明显优于现有的图异常检测算法。特别是在PubMed数据集上,AUC-PR提升了28.8%~35.4%。这些实验结果强有力地证明了在极其有限标记的元训练任务引导下,本文所提方法能够更好地学习到异常节点本质特征,从而提升小样本图异常检测任务的有效性。
