一种基于神经网络的航磁数据噪声识别和抑制方法  

A neural network-based method for identifying and suppressing noise in aeromagnetic data

作　　者：冯进凯 李姗姗[1] 何兆超 范昊鹏 李新星[1] 范雕 FENG Jinkai;LI Shanshan;HE Zhaochao;FAN Haopeng;LI Xinxing;FAN Diao(Information Engineering University,Zhengzhou 450001,China;92292 Troops,Qindao 266400,China)

机构地区：[1]信息工程大学,郑州450001 [2]92292部队,青岛266400

出　　处：《中国惯性技术学报》2025年第1期18-26,共9页Journal of Chinese Inertial Technology

基　　金：国家自然科学基金项目(42174001,42174007,42174008)。

摘　　要：航空磁力测量极易受到外界因素的干扰,噪声抑制是航磁数据处理中的关键一环。为高效识别和抑制航磁测线中存在的随机噪声,提高航磁测量精度,将神经网路方法引入到航磁测线数据的处理中,搭建了涵盖磁测数据噪声识别和噪声抑制的网络,并提出了一套适配于该网络的数据处理流程。仿真实验表明,所搭建的模型可以实现航磁测线的噪声识别和抑制,模型对验证集中的三种类型的含噪测线识别准确率达到99.85%;针对于不同类型的测线数据,噪声抑制效果相比于传统的中值滤波方法、小波滤波方法和经验模态分解方法均有不同程度的提升。实测数据实验表明,模型对航磁测线的噪声识别率为97.78%,而且能够适配实测数据中的各种噪声类别并达到较好的去噪效果,模型不受输入测线长度限制,使用更加方便灵活。Aeromagnetic measurements are highly susceptible to external interference,making noise suppression a key step in aeromagnetic data processing.To efficiently identify and suppress the random noise present in aeromagnetic survey lines and improve measurement accuracy,a neural network approach is introduced into the processing of aeromagnetic survey data.A network is developed to address both noise identification and suppression in the magnetic survey data,and a data processing workflow is proposed for the network.Simulation experiments demonstrate that the proposed model can effectively identify and suppress noise in aeromagnetic survey lines,achieving an accuracy of 99.85%in identifying noisy lines of three types in the validation set.For different types of survey data,the noise suppression performance of the model shows varying degrees of improvement compared to traditional methods such as median filtering,wavelet filtering,and empirical mode decomposition.The experimental results show that the model achieves a 97.78%noise identification rate for aeromagnetic survey lines,adapts to various noise categories in real-world data,and delivers effective denoising results.Additionally,the model is not restricted by the length of the input survey lines,offering greater flexibility and ease of use.

关 键 词：卷积神经网络 残差卷积神经网络 STFT转换 噪声识别 航磁信号去噪 

分 类 号：P631[天文地球—地质矿产勘探]