摘要
洞察号火星探测器采集的观测数据中含有大量的瑕疵干扰信号,该信号出现频繁且幅值变化相差较大,其存在严重影响后续有效信号的提取和数据处理研究.传统的时间域信号处理方法在去除瑕疵干扰信号时有处理效率低、需人工调节参数的缺点,考虑到该干扰信号虽然幅值变化较大但在频率域却有一定的相似性,因此本研究试图通过频率域实现对瑕疵干扰信号的高效识别与去除.神经网络具有强大的提取特征的能力并且能够实现对数据的快速处理,因此我们研究了结合神经网络和短时傅里叶变换提取瑕疵信号的方法.本文通过搭建CNN网络,采用有监督训练方法对样本集进行训练,得到了能够有效压制瑕疵信号的神经网络模型.对实际数据进行处理的实验结果表明本文的方法能够有效去除瑕疵信号,处理效率高且对瑕疵的压制效果更好.
The observation data collected by the InSight contains a large number of Glitches,which occur frequently and vary greatly in amplitude.These Glitches seriously affect the subsequent extraction of valuable signals and processing researches.The traditional time domain signal processing methods have the disadvantages of low processing efficiency and needing manual adjustment of parameters when removing Glitches.It is considered that although the amplitude of Glitch changes greatly,it has a certain similarity in the frequency domain.This study attempts to realize efficient identification and removal of Glitches in the frequency domain.Neural network has great feature extraction ability and can realize fast data processing,so we study the method of combining neural network and short-time Fourier transform to extract Glitches.In this paper,by setting up CNN network and using supervised training method to train the model,a neural network model that can effectively suppress Glitches is obtained.The experimental results of actual data processing show that the proposed method can effectively remove the Glitches with high processing efficiency and better suppression effect on the signals.
作者
张雅雯
王晓毅
韩坤
张江杰
ZHANG YaWen;WANG XiaoYi;HAN Kun;ZHANG JiangJie(Key Laboratory of Petroleum Resources Research,Institute of Geology and Geophysics,Chinese Academy of Sciences,Beijing 100029,China;Innovation Academy for Earth Science,Chinese Academy of Sciences,Beijing 100029,China;University of Chinese Academy of Sciences,Beijing 100049,China)
出处
《地球物理学进展》
CSCD
北大核心
2023年第6期2642-2651,共10页
Progress in Geophysics
基金
国家自然科学基金(42074158)资助。
