| 引用本文: | 李海，孙婷逸，程新宇. 基于增量贝叶斯的双偏振气象雷达降水粒子分类方法[J]. 雷达科学与技术, 2022, 20(3): 319-327.[点击复制] |
| LI Hai, SUN Tingyi, CHENG Xinyu. A Classification Method for Precipitation Particles of Dual-Polarization Weather Radar Based on Incremental Bayes[J]. Radar Science and Technology, 2022, 20(3): 319-327.[点击复制] |
|
| 摘要: |
| 面对降水粒子分类过程中可能存在的样本数不足，样本质量不高的问题，提出一种基于增量贝叶斯的双偏振气象雷达降水粒子分类方法。该方法首先处理有标签的训练数据集，获取属性节点和类节点之间的条件概率表构建朴素贝叶斯分类器；接着使用朴素贝叶斯分类器分类无标签数据，判断类置信度值后将符合条件的数据追加到训练数据集中，最后修正朴素贝叶斯分类器完成增量学习，得到增量贝叶斯分类器实现降水粒子分类。增量贝叶斯分类器不仅能够增加有效的数据样本，还能够及时更新分类器从而提高其泛化性和适应性，分类结果的准确性也得到了一定的改善。 |
| 关键词: 双偏振气象雷达 降水粒子分类 增量学习 朴素贝叶斯分类器 |
| DOI：DOI:10.3969/j.issn.1672-2337.2022.03.011 |
| 分类号:TN959.4 |
| 基金项目:民机项目（No.MJ-2018-S-28）; 天津市自然基金重点项目（No.20JCZDJC00490）; 航空基金项目（No.20182067008）; 中央高校基本科研业务费项目（No.3122018D008）; 中国民航大学蓝天教学名师培养经费 |
|
| A Classification Method for Precipitation Particles of Dual-Polarization Weather Radar Based on Incremental Bayes |
|
LI Hai, SUN Tingyi, CHENG Xinyu
|
|
Tianjin Key Lab for Advanced Signal Processing， Civil Aviation University of China， Tianjin 300300， China
|
| Abstract: |
| Facing the problem of insufficient sample number and low sample quality in the process of precipitation particle classification, a dual-polarization weather radar precipitation particle classification method based on incremental Bayes is proposed. Firstly, the labeled training data set is processed and the conditional probability table between the attribute nodes and the class nodes is obtained to construct a naive Bayes classifier. Then the naive Bayes classifier is used to classify unlabeled data. After judging the class confidence value, the eligible data is appended to the training data set. Finally, the naive Bayes classifier is modified to complete incremental learning, and the incremental Bayes classifier is obtained to realize precipitation particle classification. The incremental Bayes classifier can not only increase effective data samples, but also update the classifier in time to improve its generalization and adaptability. The accuracy of classification results has also been improved. |
| Key words: dual polarization weather radar precipitation particle classification incremental learning naive Bayes classifier |
