(1. 江西省防震減災(zāi)與工程地質(zhì)災(zāi)害探測(cè)工程研究中心(東華理工大學(xué))，南昌 330013；
2. 江西省放射性地學(xué)大數(shù)據(jù)技術(shù)工程實(shí)驗(yàn)室(東華理工大學(xué))，南昌 330013；
3. 東華理工大學(xué) 地球物理與測(cè)控技術(shù)學(xué)院，南昌 330013)
摘 要: 為解決大地電磁(Magnetotelluric, MT)常規(guī)的時(shí)間域去噪方法對(duì)于1 Hz附近噪聲壓制的局限性問題，提出了一種基于數(shù)學(xué)形態(tài)濾波(Mathematical morphological filtering, MMF)和K-SVD(K-Singular value decomposition)字典學(xué)習(xí)的新型去噪方法，用于壓制MT信號(hào)中低頻數(shù)據(jù)1Hz附近的強(qiáng)人文噪聲。首先，利用MMF分離出低頻信號(hào)，對(duì)該低頻信號(hào)進(jìn)行保護(hù)以防止信號(hào)丟失；然后，使用K-SVD字典學(xué)習(xí)對(duì)剩余的含噪信號(hào)進(jìn)行處理，通過從觀測(cè)數(shù)據(jù)中自主學(xué)習(xí)獲取噪聲的特征結(jié)構(gòu)，提取噪聲輪廓，達(dá)到去除噪聲的目的。利用一個(gè)合成數(shù)據(jù)集驗(yàn)證算法后，對(duì)兩個(gè)實(shí)測(cè)數(shù)據(jù)進(jìn)行處理，結(jié)果表明：該方法可以在幾乎不損失有效信號(hào)的前提下，消除各種強(qiáng)人文噪聲，信噪比大幅提升，數(shù)據(jù)質(zhì)量得到很大改善，且去噪效果優(yōu)于小波變換等傳統(tǒng)方法。
關(guān)鍵字: 大地電磁；數(shù)學(xué)形態(tài)濾波；K-SVD字典學(xué)習(xí)；強(qiáng)干擾壓制；低頻信號(hào)
（1.Earthquake Prevention and Mitigation and Engineering Geological Disaster Detection Engineering
Research Center in Jiangxi Province (East China University of Technology), Nanchang 330013, China;
2. Jiangxi Engineering Laboratory on Radioactive Geoscience and Big Data Technology (East China University of Technology), Nanchang 330013, China;
3.School of Geophysics and Measurement and Control Technology, East China University of Technology,
Nanchang 330013, China）
Abstract:In order to solve the problem that the conventional magnetotelluric (MT) time-domain denoising methods have the limitation of damageing low-frequency signal, a new method based on mathematical morphological filtering (MMF) and K-singular value decomposition (K-SVD) dictionary learning was proposed to suppress strong cultural noise near 1Hz of MT data. First, MMF was used to separate the low-frequency signal and protect it entirely. Second, K-SVD dictionary learning was used to process residual noisy signals. The field data was used to extract noise contours from auto-learning cultural noise features, which could achieve the purpose of removing noise. The method was verified by testing a synthetic data set and then was used to process two measured data sets. The results show that the proposed method can eliminate all kinds of strong cultural noises without losing useful signals and improve the signal-to-noise ratio and data quality. Moreover, the denoising effect of the proposed method is better than the traditional methods such as wavelet transform.
Key words: magnetotelluric; mathematical morphological filtering; K-SVD dictionary learning; strong interference suppression; low-frequency signal
