摘要
本文提出了一种基于决策树支持向量机-贝叶斯网络协同训练的双偏振气象雷达降水粒子分类方法。该方法首先使用有标签的数据初步训练得到贝叶斯网络和决策树支持向量机两种分类器。然后取出部分无标签的数据,通过两个分类器得到类别的预测结果,计算该部分预测结果的置信度,将预测结果满足阈值的无标签数据放入到有标签数据集中。使用更新后的有标签数据集重新训练两个分类器。重复该过程直到所有无标签训练数据都被放入到有标签数据集中,这时完成训练得到协同训练分类器,最后使用协同训练分类器对测试数据集进行降水粒子分类。实验表明,该方法使用一部分有标签数据结合一部分无标签数据,有效地提高了分类器性能,实现了降水粒子的准确分类。
This paper proposes a hydrometeor classification method for dual-polarization weather radar based on decision tree support vector machine(DTSVMs)and Bayesian network(BNT)collaborative training.The method first uses labeled data for preliminary training to obtain two classifiers,namely BNT and DTSVM.Then part of the unlabeled data is taken out,the category prediction results are obtained using the two classifiers,the confidence of the prediction results is calculated,and the unlabeled data with prediction result meeting a threshold is put into the labeled data set.The updated labeled data set is used to retrain the two classifiers.This process is repeated until all unlabeled training data are put into the labeled data set.At this time,the training is completed and a co-training classifier is obtained.Finally,the co-training classifier is used to classify the hydrometeor data in the test data set.Test results show that the proposed method uses both labeled data and unlabeled data,which effectively improves the performance of the classifier and realizes accurate classification of hydrometeor.
作者
李海
程新宇
尚金雷
LI Hai;CHENG Xinyu;SHANG Jinlei(Civil Aviation University of China, Tianjin 300300)
出处
《火控雷达技术》
2022年第1期5-13,共9页
Fire Control Radar Technology
基金
工业与信息化部民用飞机专项(MJ-2018-S-28)
天津市自然基金重点项目(20JCZDJC00490)
中央高校基本科研业务费项目(3122015B002)
中国民航大学蓝天教学名师培养经费。
关键词
协同训练
降水粒子分类
双偏振气象雷达
贝叶斯网络
支持向量机
co-training
hydrometeor classification
dual-polarization weather radar
Bayesian network
support vector machine
