Outlier detection for particle image velocimetry data using a locally estimated noise variance
Outlier detection for particle image velocimetry data using a locally estimated noise variance
复制标题
使用局部估计的噪声方差对粒子图像测速数据进行异常值检测
DOI：
10.1088/1361-6501/aa5431
复制
发表时间：
2017-01
影响因子：
2.4
通讯作者：
ZhouPing Yin
中科院分区：
文献类型：
--
作者：
Yong Lee;Hua Yang;ZhouPing Yin
这项工作描述了一种自适应空间变量阈值异常检测算法，用于原始网格粒子图像测速数据，使用局部估计的噪声方差。该方法是一个迭代过程，每次迭代由参考向量场重建步骤和离群点检测步骤组成。我们使用加权自适应平滑方法构建参考向量场（Garcia 2010 Comput）。Stat. Data Anal. 54 1167-78)，权重是使用改进的离群值检测器在离群值检测步骤中确定的(Ma et al. 2014 IEEE Trans。图像处理。23 1706-21)。对迭代的最终权重的艰难决定可能会产生字段的异常标签。技术上的贡献是首次在迭代框架中将空间变量阈值动机嵌入到具有局部估计噪声方差的改进离群值检测器中。结果表明，在涡旋、湍流等复杂流动中，空间变量阈值优于单一空间常数阈值。采用模拟散点或聚类异常值的合成细胞涡流来评估我们提出的方法的性能，并与流行的验证方法进行比较。该方法也适用于实际的PIV湍流测量。实验结果表明，该方法在异常值检测不足数和异常值检测过数方面具有一定的竞争力。此外，该离群点检测方法计算效率高，自适应能力强，不需要自定义参数，在补充资料中也提供了相应的实现方法。
This work describes an adaptive spatial variable threshold outlier detection algorithm for raw gridded particle image velocimetry data using a locally estimated noise variance. This method is an iterative procedure, and each iteration is composed of a reference vector field reconstruction step and an outlier detection step. We construct the reference vector field using a weighted adaptive smoothing method (Garcia 2010 Comput. Stat. Data Anal. 54 1167–78), and the weights are determined in the outlier detection step using a modified outlier detector (Ma et al 2014 IEEE Trans. Image Process. 23 1706–21). A hard decision on the final weights of the iteration can produce outlier labels of the field. The technical contribution is that the spatial variable threshold motivation is embedded in the modified outlier detector with a locally estimated noise variance in an iterative framework for the first time. It turns out that a spatial variable threshold is preferable to a single spatial constant threshold in complicated flows such as vortex flows or turbulent flows. Synthetic cellular vortical flows with simulated scattered or clustered outliers are adopted to evaluate the performance of our proposed method in comparison with popular validation approaches. This method also turns out to be beneficial in a real PIV measurement of turbulent flow. The experimental results demonstrated that the proposed method yields the competitive performance in terms of outlier under-detection count and over-detection count. In addition, the outlier detection method is computational efficient and adaptive, requires no user-defined parameters, and corresponding implementations are also provided in supplementary materials.
登录
查看更多内容
用于检测 PIV 数据中的虚假向量的细胞神经网络
影响因子：
2.4
作者：
D.F.Liang;C.B.Jiang;Y.L.Li
通讯作者：
Y.L.Li
DOI：
10.1007/978-3-642-83787-6_4
发表时间：
1989
期刊：
--
影响因子：
--
作者：
L. Lourenço
通讯作者：
L. Lourenço
影响因子：
2.4
作者：
Yong Lee;Hua Yang;Zhouping Yin
通讯作者：
Yong Lee;Hua Yang;Zhouping Yin
DOI：
10.1109/sibgrapi.2009.20
发表时间：
2009-10
期刊：
2009 XXII Brazilian Symposium on Computer Graphics and Image Processing
影响因子：
--
作者：
Marcos Lage;Rener Castro;Fabiano Petronetto;A. Bordignon;G. Tavares;T. Lewiner;H. Lopes
通讯作者：
Marcos Lage;Rener Castro;Fabiano Petronetto;A. Bordignon;G. Tavares;T. Lewiner;H. Lopes
DOI：
10.1016/j.csda.2011.12.001
发表时间：
2012-06
期刊：
Comput. Stat. Data Anal.
影响因子：
--
作者：
L. L. Tarnec-L.;Damien Garcia
通讯作者：
L. L. Tarnec-L.;Damien Garcia