Improvement in universal PIV outlier detection by means of coherence adaptivity
Improvement in universal PIV outlier detection by means of coherence adaptivity
复制标题
DOI：
--
复制
发表时间：
2015-09
期刊：
影响因子：
--
通讯作者：
A. Masullo;R. Theunissen
中科院分区：
文献类型：
--
作者：
A. Masullo;R. Theunissen
本文提出了一种新的速度场验证技术。现有的矢量验证方法将字段的每个矢量与恒定数量的邻居进行比较，因此无法在离群聚类附近检测到错误矢量。此外，由于速度场中的强梯度，在模糊的情况下，它们易于过度检测正确的向量作为离群值。为此，提出了一种新的验证算法，以增强离群聚类的鲁棒性并减少过度检测。这一目标是追求通过一个连贯性自适应可变邻域：与现有的方法相比，新算法自动扩大的邻居的数量，以始终确保一个可靠的比较仔细的矢量。相干性被定义为具有其最近邻的抛物线回归的向量的残差，并且在应用于现有方法时也允许对离群值集群的鲁棒性的强烈增强。除此之外，该算法还提供了一个基于距离的高斯加权系统，并通过幅度和方向而不是矢量分量来执行矢量的比较。为了进一步提高检测，一个新的运营商来评估一个样本的数据的中位数也建议和自动评估的背景误差有助于自适应性。这种新技术的评估进行了几个Monte Carlo模拟：合法的速度场被污染的假向量收集在不同的大小和幅度的组。为了强调在这种新技术中引入的众多创新的重要性，还提出了一个修改版本的相干自适应可变邻域的先前验证技术作为比较。PIV实验的应用程序也显示了验证算法在真实的图像分析的影响。
A novel technique for the validation of velocity fields is proposed in this paper. Existing methodologies for vector validation compare each vector of the field with a constant number of neighbours and are as such, unable to detect false vectors in the vicinity of outlier clusters. Moreover, they are apt to over-detect correct vectors as outliers in case of ambiguity due to strong gradients in the velocity field. For this reason, a new validation algorithm has been proposed to enhance robustness in case of outlier clusters and reduce over-detection. This goal is pursued through a coherence-adaptive variable neighbourhood: in contrast with existing methodologies, the novel algorithm automatically enlarges the number of neighbours for a scrutinized vector in order to always ensure a reliable comparison. Coherence is defined as the residual of a vector with a parabolic regression of its closest neighbours and allows a strong enhancement in robustness against outlier clusters also when applied to existing methodologies. In addition to this, the proposed algorithm is provided with a distance-based Gaussian weighting system and the comparison of vectors is performed by means of magnitude and direction instead of vector components. To further improve detection, a new operator to evaluate the median of a sample of data is also suggested and an automatic evaluation of the background error contributes to the adaptivity. Assessment of this novel technique is performed with several Monte Carlo simulations: legitimate velocity fields are contaminated with false vectors collected in groups of different sizes and magnitudes. In order to stress the importance of the numerous innovations introduced in this novel technique, a modified version of previous validation techniques with the coherence-adaptive variable neighbourhood is also proposed as a comparison. An application to a PIV experiment also shows the influence of the validation algorithm in a real image analysis.