基于DnCNN 的侵彻过载时频去噪方法  

Time-Frequency Denoising Method for Penetration Overload Signals Based on Denoising Convolutional Neural Network(DnCNN)

作　　者：郑宏亮 贾森清 郭宇朋 薛颖杰 韩晶[1,2] 赵河明 石志刚[3] ZHENG Hongliang;JIA Senqing;GUO Yupeng;XUE Yingjie;HAN Jing;ZHAO Heming;SHI Zhigang(School of Mechanical and Electrical Engineering,North University of China,Taiyuan 030051,China;Shanxi Key Laboratory of High-end Equipment Reliability Technology,Taiyuan 030051,China;Hubei Space Sanjiang Honglin Detection and Control Co.,Ltd.,Hubei Xiaogan 432000,China;Science and Technology on Electromechanical Dynamic Control Laboratory,Xi'an Institute of Electromechanical Information Technology,Xi'an 710065,China)

机构地区：[1]中北大学机电工程学院,太原030051 [2]山西省高端装备可靠性技术重点实验室,太原030051 [3]湖北三江航天红林探控有限公司,湖北孝感432000 [4]西安机电信息技术研究所机电动态控制重点实验室,西安710065

出　　处：《装备环境工程》2024年第8期17-24,共8页Equipment Environmental Engineering

基　　金：山西省高端装备可靠性技术重点实验室研究基金(446110103)。

摘　　要：目的提高从侵彻过载中准确估计刚体过载信号的能力。方法提出一种基于前馈去噪卷积神经网络(DnCNN)的侵彻过载时频去噪方法,该方法首先应用短时傅里叶变换(STFT)提取侵彻过载信号的时频图像,使DnCNN能够充分利用时频图像信息,估计出刚体过载时频图像。最后,通过逆STFT将时频图像转换回时域,得到估计的刚体过载信号。结果在5-Fold交叉验证中,所提方法在测试集上的平均绝对误差(MAE)为0.968%,Pearson相关系数(r)为90.35%。与低通滤波、总体经验模态分解(EEMD)和小波变换方法相比,所提方法的平均MAE分别降低了1.82%、1.00%、0.75%,平均相关系数r值分别提高了47.81%、17.48%、22.93%。结论所提方法可以从侵彻过载中准确估计出刚体过载信号,在去噪能力上优于低通滤波、EEMD和小波变换方法,且在去噪过程中,无需调整参数,能够自动完成去噪任务。The work aims to enhance the ability to accurately estimate rigid body overload signals from the penetration overload signals.A time-frequency denoising method based on feedforward denoising convolutional neural network(DnCNN)was proposed.In this method,firstly the short-time Fourier transform(STFT)was applied to extract the time-frequency images of the penetration overload signal so that the DnCNN network could make full use of these images to effectively estimate the time-frequency images of the rigid-body overload.Finally,the time-frequency images were converted back to the time domain by inverse STFT to obtain the estimated rigid body overload signal.In the 5-Fold Cross-Validation,the proposed method had a mean absolute error(MAE)of 0.968% and a Pearson correlation coefficient(r)of 90.35%on the test set.Compared with low-pass filtering,ensemble empirical modal decomposition(EEMD)and wavelet transform methods,the proposed method performed better in denoising ability.Specifically,the average MAE of the proposed method was reduced by 1.82%,1.00%,and 0.75%,while the average correlation coefficient r-value was improved by 47.81%,17.48%,and 22.93%,respectively.The proposed method can accurately estimate the rigid body overload signal from the penetration overload and outperform low-pass filtering,EEMD and wavelet transform methods in denoising capability.In the denoising process,there is no need to adjust parameters and the denoising task can be automatically completed.

关 键 词：硬目标侵彻 侵彻过载 前馈去噪卷积神经网络 信号去噪 时频分析 k-Fold交叉验证 

分 类 号：O385[理学—流体力学]