基于核密度估计的贝叶斯模型冰雹检测方法

A Bayesian Model Based Hail Detection Method Using Kernel Density Estimation

针对冰雹检测方法中大量带标签数据不易获取的问题,本文提出一种基于核密度估计(Kernel Density Estimation,KDE)的贝叶斯模型冰雹检测方法,利用大量无标签数据和少量带标签数据,根据概率对冰雹进行检测。该方法首先根据大量无标签数据的密集程度将其初步分成若干集合,然后每个集合采用高斯核密度函数以及最佳固定带宽值,计算得到每个集合的类条件概率,其次利用少量带标签数据以及降水粒子的融化层信息建立类先验概率,最后根据每个集合的最大后验概率匹配降水粒子类型,实现冰雹与其他降水粒子类型的检测结果。通过实测数据对模型的验证结果表明,该方法能够较好的实现冰雹检测。

In view of the problem that it is difficult to obtain a large amount of labeled data when using hail detection methods,this paper proposes a bayesian model based hail detection method using kernel density estimation(KDE).It utilizes a large amount of unlabeled data and a small amount of labeled data to detect the hail based on probability.Firstly,the proposed method divides a large number of unlabeled data into several sets initially according to their density.Then,the Gaussian kernel density function and the optimal fixed bandwidth value are used to calculate the class-conditional probability of each set.Secondly,a small amount of labeled data and the melting layer information of precipitation particles are used to establish class prior probabilities.Finally,the hail is detected from other types of precipitation particles by matching precipitation particle types according to the maximum posterior probability of each set.The results of a test using measured data show that the proposed method has good hail detection performance.