【作者】 王亮；

【导师】 田宝凤；

【作者基本信息】 吉林大学 ， 测试计量技术及仪器， 2020， 硕士

【摘要】 磁共振测深技术(Magnetic Resonance Sounding,MRS)作为当前世界上先进的直接探测地下水的地球物理方法,具有直接、定量和无损测量的优点,在水资源勘查与评价、矿井/隧道突涌水等地质灾害隐患探测领域被广泛应用。然而,在实际应用中,由于MRS探测是利用天然地磁场,无法采取屏蔽措施,导致大量的环境噪声和人为噪声等干扰信息一并被仪器系统捕获,严重影响采集数据的质量,不能实现有效信号的提取。如何研究出一种数据采集与信号处理相结合的方法,以直接获取MRS信号为目标,降低环境噪声,实现复杂环境干扰下目标水体的准确检测,是磁共振探测技术在实际应用中所面临的重大难题。压缩感知(Compressed Sensing,CS)作为一种信息采样获取理论,通过信号的稀疏特性,利用数据的冗余信息,通过随机采样的方式获取少量信号的离散样本,最后采用非线性重建算法恢复信号。鉴于此,本文提出开展基于压缩感知的MRS信号重构与噪声压制研究,具有重要的意义和实用价值。本文依据MRS信号和噪声的特点,首先通过比较分析几种常用稀疏基的特征,确定采用K-SVD学习字典对含噪MRS信号进行稀疏分解;其次,开展了不同噪声类型下MRS信号的压缩感知重构和噪声压制算法仿真,探讨了信噪比、测量数量点数,弛豫时间等参数对算法性能的影响;最后,通过实测数据处理实验和与其他算法的对比,验证所提算法的性能。主要的研究成果包括如下:(1)确定压缩感知算法应用中的三要素。通过对比分析傅里叶变换(FFT)、离散余弦变换(DCT)和学习字典(K-SVD)的特性,选择K-SVD学习字典作为稀疏变换基;通过对比高斯随机矩阵、伯努利随机矩阵和部分哈达玛矩阵,选择随机性更强、重构效果更好的高斯随机矩阵作为测量矩阵;通过对比正交匹配追踪和基追踪算法,选择运算速度快、重构精度较高的正交匹配追踪算法进行重构。(2)开展了基于压缩感知的重构与噪声压制算法仿真实验。在重构效果分析中,通过对无噪和含噪MRS信号两种情况下进行仿真分析,实现信号较好的重构;以RMSE和相似系数为评价指标,开展了不同信噪比和不同测量数量点数下的重构效果分析,证明了本文算法的有效性和普适性,并可以依据重构误差RMSE=14.65 n V和相关系数为0.9978来选择合适的测量数量点数。在噪声压制效果分析中,采用预处理后的MRS信号估计值作为K-SVD字典的初始值,使得信号和噪声在稀疏基上稀疏特性不同,实现信噪分离,达到噪声压制的目的。通过对不同测量数量点数、不同信噪比和不同弛豫时间下噪声抑制效果进行分析,证明了算法对噪声抑制的有效性,且信噪比提升最高可达23 d B。(3)开展了基于压缩感知的野外实测数据采集与处理实验。经过压缩感知方法处理后,信噪比提升最高可达15 d B,将本文方法处理后信号与其它方法处理后结果进行对比,验证本文方法的有效性。更多还原

【Abstract】 Magnetic resonance sounding(MRS)technology is an advanced groundwater detection method that can directly and quantitative acquire groundwater information.It is also widely used in the advanced detection of disaster water sources such as landslides and mine water inrush.However,in practical applications,MRS detection uses natural geomagnetic fields,shielding measures cannot be taken,resulting in a large amount of interference information such as environmental noise and man-made noise being captured by the instrument system,which seriously affects the quality of the collected data and cannot achieve effective signals extraction.How to develop a method of combining data acquisition and signal processing,aiming to directly acquire MRS signals,reduce environmental noise,and achieve accurate detection of target water bodies under complex environmental interference is the face of magnetic resonance detection technology in practical applications major problems.Compressed sensing(CS),a kind of information sampling acquisition theory,through the sparse characteristics of the signal,using the redundant information of the data,a small amount of signal is obtained by random sampling discrete samples,and finally use a nonlinear reconstruction algorithm to recover the signal.In view of this,this paper proposes to carry out research on MRS signal reconstruction and noise suppression based on compressed sensing,which has important significance and practical value.Based on the characteristics of MRS signal and noise,this paper first analysis the characteristics of several commonly used sparse bases to determine the sparse decomposition of noisy MRS signal using K-SVD learning dictionary;Secondly,the simulation of MRS signal compressed sensing reconstruction and noise suppression algorithm under different noise was carried out,and the influence of parameters such as signal-to-noise ratio,measured data length,and relaxation time on the performance of the algorithm was discussed;Finally,the performance of the proposed algorithm is verified through measured data processing experiments and comparison with other algorithms.The main research results include the following:(1)Determine the three elements in the application of compressed sensing algorithm.By comparing and analyzing the characteristics of fourier transform(FFT),Discrete Cosine Transform(DCT)and Learning Dictionary(K-SVD),K-SVD learning dictionary is selected as the sparse transform base;By comparing Gaussian random matrix,Bernoulli random matrix and partial Hadamard matrix,the Gaussian random matrix has stronger randomness and better reconstruction effect is selected as the measurement matrix;By comparing the orthogonal matching tracking and base tracking algorithms,the orthogonal matching tracking algorithm with fast calculation speed and high reconstruction accuracy is selected for reconstruction.(2)Carry out simulation experiments of reconstruction and noise suppression algorithms based on compressed sensing.In the analysis of the reconstruction effect,the simulation and analysis of the noiseless and noisy MRS signals are used to achieve a better signal reconstruction;Using RMSE and similarity coefficient as evaluation indicators,the reconstruction effect analysis under different signal-to-noise ratio and different measurement data lengths is carried out,which proves the effectiveness and universality of the algorithm in this paper,and can be based on the reconstruction error RMSE = 14.65 n V and related coefficient is 0.9978 to select the appropriate number of points.In the analysis of the noise suppression effect,the pre-processed noise-free MRS signal estimate is used as the initial value of the K-SVD dictionary,so that the signal and noise have different sparse characteristics on the sparse basis,and the signal-noise separation is achieved to the purpose of noise suppression.By analyzing different measurement data lengths,different signal-to-noise ratios and different relaxation time noise suppression effects,the effectiveness of the algorithm for noise suppression is proved,and the signal-to-noise ratio can be improved up to 23 d B.(3)Conducted field measurement data collection and processing experiments based on compressed sensing.After processing by the compressed sensing method,the signal-to-noise ratio can be improved by up to 15 d B.The signal processed by this method is compared with the results of other methods to verify the effectiveness of this method.更多还原