为了全面了解太阳活动规律,需要连续观测时长覆盖多个太阳活动周的、高质量数据.太阳磁图是研究太阳活动的重要数据,连续、长时间和高空间分辨率的太阳磁图能够提供更精细的太阳磁场演化信息,有助于更准确地预报太阳活动和空间天气事件...为了全面了解太阳活动规律,需要连续观测时长覆盖多个太阳活动周的、高质量数据.太阳磁图是研究太阳活动的重要数据,连续、长时间和高空间分辨率的太阳磁图能够提供更精细的太阳磁场演化信息,有助于更准确地预报太阳活动和空间天气事件.因此,提出一种基于深度学习的超分率算法,对MDI(Michelson Doppler Imager)磁图进行超分,取得与日震和磁成像仪(Helioseismic and Magnetic Imager,HMI)磁图一致的分辨率,从而能够获得持续时长将近两个太阳活动周的高质量太阳磁图数据库.为了引导网络学习磁图中有效的特征信息,将注意力机制引入到网络中,学习注意力权重图.此外,采用了不确定性损失作为模型训练的损失函数,该方法能够对于带有磁场变化的纹理和边缘分配更大的权重,同时不增加网络参数和计算量.实验证明,提出的算法显著提高了超分太阳磁图的质量,在峰值信噪比(Peak Signal-to-Noise Ratio,PSNR:33.3168)、结构相似性(Structure Similarity Index Measure,SSIM:0.8754)、相关性(Correlation Coefficient,CC:0.9323)和均方根误差(Root Mean Square Error,RMSE:21.8544)等指标上取得了最优的结果.展开更多
Measurement uncertainty plays an important role in laser tracking measurement analyses. In the present work, the guides to the expression of uncertainty in measurement(GUM) uncertainty framework(GUF) and its supplemen...Measurement uncertainty plays an important role in laser tracking measurement analyses. In the present work, the guides to the expression of uncertainty in measurement(GUM) uncertainty framework(GUF) and its supplement, the Monte Carlo method, were used to estimate the uncertainty of task-specific laser tracker measurements. First, the sources of error in laser tracker measurement were analyzed in detail, including instruments, measuring network fusion, measurement strategies, measurement process factors(such as the operator), measurement environment, and task-specific data processing. Second, the GUM and Monte Carlo methods and their application to laser tracker measurement were presented. Finally, a case study involving the uncertainty estimation of a cylindricity measurement process using the GUF and Monte Carlo methods was illustrated. The expanded uncertainty results(at 95% confidence levels) obtained with the Monte Carlo method are 0.069 mm(least-squares criterion) and 0.062 mm(minimum zone criterion), respectively, while with the GUM uncertainty framework, none but the result of least-squares criterion can be got, which is 0.071 mm. Thus, the GUM uncertainty framework slightly underestimates the overall uncertainty by 10%. The results demonstrate that the two methods have different characteristics in task-specific uncertainty evaluations of laser tracker measurements. The results indicate that the Monte Carlo method is a practical tool for applying the principle of propagation of distributions and does not depend on the assumptions and limitations required by the law of propagation of uncertainties(GUF). These features of the Monte Carlo method reduce the risk of an unreliable measurement of uncertainty estimation, particularly in cases of complicated measurement models, without the need to evaluate partial derivatives. In addition, the impact of sampling strategy and evaluation method on the uncertainty of the measurement results can also be taken into account with Monte Carlo method, which plays a guiding role in measurement planning.展开更多
文摘为了全面了解太阳活动规律,需要连续观测时长覆盖多个太阳活动周的、高质量数据.太阳磁图是研究太阳活动的重要数据,连续、长时间和高空间分辨率的太阳磁图能够提供更精细的太阳磁场演化信息,有助于更准确地预报太阳活动和空间天气事件.因此,提出一种基于深度学习的超分率算法,对MDI(Michelson Doppler Imager)磁图进行超分,取得与日震和磁成像仪(Helioseismic and Magnetic Imager,HMI)磁图一致的分辨率,从而能够获得持续时长将近两个太阳活动周的高质量太阳磁图数据库.为了引导网络学习磁图中有效的特征信息,将注意力机制引入到网络中,学习注意力权重图.此外,采用了不确定性损失作为模型训练的损失函数,该方法能够对于带有磁场变化的纹理和边缘分配更大的权重,同时不增加网络参数和计算量.实验证明,提出的算法显著提高了超分太阳磁图的质量,在峰值信噪比(Peak Signal-to-Noise Ratio,PSNR:33.3168)、结构相似性(Structure Similarity Index Measure,SSIM:0.8754)、相关性(Correlation Coefficient,CC:0.9323)和均方根误差(Root Mean Square Error,RMSE:21.8544)等指标上取得了最优的结果.
基金Project(51318010402)supported by General Armament Department Pre-Research Program of China
文摘Measurement uncertainty plays an important role in laser tracking measurement analyses. In the present work, the guides to the expression of uncertainty in measurement(GUM) uncertainty framework(GUF) and its supplement, the Monte Carlo method, were used to estimate the uncertainty of task-specific laser tracker measurements. First, the sources of error in laser tracker measurement were analyzed in detail, including instruments, measuring network fusion, measurement strategies, measurement process factors(such as the operator), measurement environment, and task-specific data processing. Second, the GUM and Monte Carlo methods and their application to laser tracker measurement were presented. Finally, a case study involving the uncertainty estimation of a cylindricity measurement process using the GUF and Monte Carlo methods was illustrated. The expanded uncertainty results(at 95% confidence levels) obtained with the Monte Carlo method are 0.069 mm(least-squares criterion) and 0.062 mm(minimum zone criterion), respectively, while with the GUM uncertainty framework, none but the result of least-squares criterion can be got, which is 0.071 mm. Thus, the GUM uncertainty framework slightly underestimates the overall uncertainty by 10%. The results demonstrate that the two methods have different characteristics in task-specific uncertainty evaluations of laser tracker measurements. The results indicate that the Monte Carlo method is a practical tool for applying the principle of propagation of distributions and does not depend on the assumptions and limitations required by the law of propagation of uncertainties(GUF). These features of the Monte Carlo method reduce the risk of an unreliable measurement of uncertainty estimation, particularly in cases of complicated measurement models, without the need to evaluate partial derivatives. In addition, the impact of sampling strategy and evaluation method on the uncertainty of the measurement results can also be taken into account with Monte Carlo method, which plays a guiding role in measurement planning.
