基于改进FNN的降水粒子分类方法  

Hydrometeor Classification Method Based on Improved Fuzzy Neural Network(FNN)

作　　者：李海[1] 郭生权 张超群 冯青[1] LI Hai;GUO Shengquan;ZHANG Chaoqun;FENG Qing(Intelligent Signal and Image Processing Key Lab of Tianjin,Civil Aviation University of China,Tianjin 300300)

机构地区：[1]中国民航大学天津市智能信号与图像处理重点实验室,天津300300

出　　处：《火控雷达技术》2022年第4期1-10,共10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。

摘　　要：针对模糊神经网络(Fuzzy Neural Network,FNN)降水粒子分类方法在训练过程中隶属度函数初值不易确定的问题,本文提出了一种改进的模糊神经网络降水粒子分类方法。首先利用K-means++算法对没有标签区域里的双偏振气象雷达数据进行聚类,将聚类结果和另一区域中双偏振气象雷达获取的少量带标签数据进行相关性分析来得到带标签数据,接着对带标签数据进行统计分析来得到模糊神经网络隶属度函数参数初值,将统计好的参数代入模糊神经网络中并对网络进行离线训练,最后使用依据数据自适应调节好的模糊神经网络分类器实现降水粒子的分类。通过对实测数据实验结果表明,该方法能够得到较为合理的降水粒子分类结果。To solve the problem that the initial value of membership function of fuzzy neural network(FNN)hydrometeor classification method is difficult to determine in the training process,an improved fuzzy neural network hydrometeor classification method was proposed in this paper.Firstly,K-means++algorithm was adopted in the no-label-area to cluster dual-polarization weather radar data;clustering result and the other regions of the dual-polarization weather radar for a small amount of labeled data correlation analysis to get labeled data;then the labeled data were analyzed statistically to get the initial parameter of fuzzy neural network function.The statistical parameters were put into the fuzzy neural network and the network was trained off-line.Finally,the hydrometeor classification was realized via fuzzy neural network classifier that was adaptively adjusted.The experimental results of the measured data showed that the method can obtain reasonable hydrometeor classification results.

关 键 词：双偏振气象雷达 降水粒子分类 K-means++算法 相关性分析 模糊神经网络 

分 类 号：TN959.4[电子电信—信号与信息处理]