论文详情
瞬变电磁信号易受各种噪声的干扰,尤其是在信号晚期,噪声甚至会淹没有效信号,严重影响了后期的反演。为此,提出了将瞬变电磁信号分为早中期和晚期两部分进行处理的思路,并首次将非负矩阵分解(NMF)的有监督算法应用于受噪声影响较大的瞬变电磁晚期信号的处理。首先,在训练阶段,将纯净信号进行短时傅里叶变换和非负矩阵分解处理,得到表征信号特征的原子字典。然后,在降噪阶段,利用原子字典和降噪模型处理含噪信号,得到初步估计的瞬变电磁信号。最后,多次重复以上步骤,将初步估计的瞬变电磁信号的晚期数据和含噪信号的原始早中期数据分别累加,求各自的算术平均值,再将两者拼接,估计出最终的完整的瞬变电磁信号。仿真实验和实测数据处理结果表明,该算法可以有效改善信噪比和均方根误差,减少波形的失真,降噪效果明显优于小波变换、Hilbert-Huang变换(HHT)和奇异值分解(SVD)等传统降噪方法,为提高后续反演的准确度奠定了基础。
Transient electromagnetic signals are easily disturbed by various types of noise.A transient electromagnetic signal can be divided into early,middle,and late fields.In the late field of a signal,noise may completely mask the effective signal,seriously affecting the inversion process.A supervised algorithm based on non-negative matrix decomposition (NMF) was applied to the late field of a signal to reduce the noise.The method consisted of the following steps.First,in the training phase,the original signal undergoes short-time Fourier transformation and non-negative matrix factorization to yield an atomic dictionary representing the characteristics of the signal.Then,in the noise reduction phase,the signal is processed using the dictionary and the noise reduction model to obtain a preliminary estimate of the transient electromagnetic signal.Subsequently,the late field data of the processed signal and the original early and middle field data are summed separately by repeating the above steps.Then,the arithmetic mean of each portion is calculated,and the portions are finally spliced to reconstruct the complete transient electromagnetic signal.Applications on simulated and measured data showed that the proposed algorithm can effectively improve the signal-to-noise ratio and root mean square error and reduce the distortion of the waveform.The noise reduction effect was also significantly stronger than that provided by traditional noise reduction methods such as the wavelet transform,Hilbert-Huang transform,and singular value decomposition.
国家自然科学基金项目(61961009)、广西自然科学基金重点项目(2016GXNSFDA380018)、桂林电子科技大学认知无线电与信息处理教育部重点实验室基金项目(CRKL160107)和认知无线电与信息处理省部共建教育部重点实验室2017年度主任基金项目(CRKL170108)共同资助。