基于深度卷积神经网络的频高图分类研究  

Research on classification of vertical ionogram based on deep convolution neural network

作　　者：鲁转侠[1] 华彩成[1] 蔚娜[1] 冯静[1] 娄鹏[1] 刘维平 LU ZhuanXia;HUA CaiCheng;WEI Na;FENG Jing;LOU Peng;LIU WeiPing(China Research Institute of Radiowave Propagation,Qingdao 266107,China)

机构地区：[1]中国电波传播研究所,青岛266107

出　　处：《地球物理学进展》2022年第5期1834-1839,共6页Progress in Geophysics

基　　金：中国电波传播研究所稳定支持科研经费资助项目(A132004W10,A132004W11);国家自然科学基金(62031014,62031015)联合资助。

摘　　要：本文提出了一种利用深度卷积神经网络的频高图分类方法,在频高图分类标记的基础上,通过对深度学习经典网络结构的网络层迁移的方式,构建频高图类型识别网络模型,实现基于传播模式分布的频高图自动分类.利用试验获取的大量频高图数据,依据频高图中电离层传播模式分布情况,结合频高图度量基本规则,人工对频高图数据分类标记,生成网络模型样本数据;然后以随机方式,选取样本数据85%的数据作为训练数据,其余数据作为测试数据;经验证训练后的网络模型能够将测试频高图数据自动分为七种类型,频高图类型识别综合准确率高于97%.该方法可为频高图特征参数的自动、精确提取提供重要技术和高质量数据支撑,对电离层结构信息有效获取具有重要意义.A classification method of vertical ionograms using deep convolution neural network is proposed in this paper.The network model of the vertical ionogram type recognition is constructed by means of network layer migrating of deep learning classical network structure on the basis of the type label of the vertical ionogram.The network model can realize automatic classification of vertical ionograms based on propagation mode distribution.A large number of vertical ionograms obtained from the experiment are collected.Operators manually classified to generate the sample data of the network model,according to the distribution of ionospheric propagation modes in vertical ionograms and the basic scaling rules of vertical ionogram.Then,in a random way,85%of the sample data are selected as training data,and the other 15%of the sample data are selected as the test data.It is verified that the trained network model can automatically classify the test vertical ionogram data into seven types.The comprehensive accuracy of vertical ionogram type recognition is higher than 97%.This method can provide important technology support and high-quality data support for automatic and accurate extraction of characteristic parameters of vertical ionogram.The method is the great significance for the effective acquisition of ionospheric structure information.

关 键 词：频高图 深度卷积神经网络 传播模式 

分 类 号：P352[天文地球—空间物理学]