基于卷积编解码网络的自由感应衰减信号提取方法  

Free induction decay signal extraction method based on convolutional encoder-decoder network

作　　者：王琦[1] 杜海龙[1] 陈玫玫[1] 韦健[1] WANG Qi;DU Hailong;CHEN Meimei;WEI Jian(College of Communication Engineering,Jilin University,Changchun 130000,China)

机构地区：[1]吉林大学通信工程学院,长春吉林130000

出　　处：《实验技术与管理》2022年第1期59-61,65,共4页Experimental Technology and Management

基　　金：国家自然科学基金(61901187);吉林省自然科学基金(20190201111JC)。

摘　　要：在实际应用中,自由感应衰减信号(FID)数据中通常含有大量噪声干扰,直接影响了检测结果的准确性。该文针对上述问题,设计了一种基于卷积编解码网络的FID信号提取方法。即在卷积神经网络的框架下,通过训练学习复杂噪声背景下FID数据时频谱与纯净FID信号时频谱之间的映射关系,实现FID信号的噪声压制,再经短时傅里叶逆变换完成FID信号的提取。在卷积网络训练中同时考虑了FID信号的实部和虚部特征,更好地保留了其相位信息。实验结果表明:提出的FID信号提取方法可以在多种噪声同时存在的情况下有效地压制噪声,且不损失FID信号。该方法可作为深度学习课程的验证性实验,也可作为创新性实验内容。In practical applications,free induction decay(FID)data usually contains a lot of noise interference,which directly affects the accuracy of the detection results.In response to the above problems,a FID signal extraction method is designed based on a convolutional encoder-decoder network.Under the framework of convolutional neural network,and by training and learning the mapping relationship between complex noise FID data spectrum and pure FID signal spectrum,the noise suppression of FID signal is realized,and the FID extraction is completed by short-time inverse Fourier transform.Furthermore,the real and imaginary features of the FID signal is considered at the same time in the training of the convolutional network,and its phase information is reained.The experimental results show that the FID signal extraction method in this paper can effectively suppress the noise in the presence of multiple noises without losing the FID signal.This method can be used as a confirmatory experiment for deep learning courses,which can also be used as an innovative experiment content.

关 键 词：自由感应衰减 卷积编解码网络 时频谱 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]