基于深度卷积神经网络的频高图特征提取研究  

Research on characteristic extraction of vertical ionogram based on deep convolution neural network

作　　者：鲁转侠[1] 华彩成[1] 冯健[1] 蔚娜[1] 王岳松[1] 冯静[1] 娄鹏[1] 王严 李春晓 LU ZhuanXia;HUA CaiCheng;FENG Jian;WEI Na;WANG YueSong;FENG Jing;LOU Peng;WANG Yan;Li ChunXiao(China Research Institute of Radio Wave Propagation,Qingdao Shandong 266107,China)

机构地区：[1]中国电波传播研究所,山东青岛266107

出　　处：《地球物理学报》2024年第9期3290-3300,共11页Chinese Journal of Geophysics

基　　金：国家自然科学基金项目(62031014,62031015,U23B2016);中国电波传播研究所稳定支持科研经费项目(A132004W11,A132004W10)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特征提取方法,在频高图不同层回波信息标记的基础上,构建包含降采样部分和上采样部分的频高图回波识别网络模型,实现了频高图不同回波信息自动识别.利用试验获取的频高图数据,通过人工对频高图中电离层不同层的回波信息分别标记,生成网络模型样本数据集.以随机方式,选取样本数据集80%的数据作为训练数据,其余数据作为测试数据.经网络模型训练和测试,结果显示网络模型能够自动有效地识别测试频高图中不同层的回波信息.在此基础上,结合数字图像处理中的腐蚀算法和连通域思想,针对性地设计滤波器,滤除已识别回波信息中的噪声、干扰、多跳回波,能够实现测试频高图特征参数的有效提取.并且通过与传统方法比较,该方法特征提取精度整体上优于传统方法,可为频高图特征的自动、精确提取提供一种新的技术方法.This paper proposes a feature extraction method for vertical ionograms using deep convolution neural networks.Based on the labeling of echo information in different layers of the vertical ionogram,a vertical ionogram echo recognition network model including downsampling parts and upsampling parts is constructed to achieve automatic recognition of different echoe information in the vertical ionogram.Using vertical ionograms obtained from experiments,manually label the echo information of different layers of the ionosphere in the vertical ionogram to generate a network model sample dataset.Randomly select 80%of the sample dataset as training data and the remaining data as testing data.After training and testing the network model,the results showed that the network model can automatically and effectively recognize the echo information of different layers in the vertical ionogram.On this basis,combining the corrosion algorithm and connected domain idea in digital image processing,a targeted filter is designed to filter noise,interference,and multi hop echos in the identified echo information,which can effectively extract the characteristic parameters of the test vertical ionogram.And compared with the traditional method,the overall accuracy of feature extraction in this method is better than that of the traditional method,which can provide a new technical means for automatic and accurate feature extraction of the vertical ionogram.

关 键 词：频高图 深度卷积神经网络 临界频率 

分 类 号：P352[天文地球—空间物理学]