| Water shortage and how to effectively detect groundwater resource has become one of the burning problems in the world. Magnetic resonance sounding(MRS), as the most efficient and direct geophysical probing technique, has been wildly applied in detecting underground water at present. MRS technology obtains information of water content in the stratum by detecting resonance signal produced by resonant transition of hydrogen proton of water that is excited by artificial magnetic field, which is one kind of efficient technique that is able to receive abundant information in a short time.However, signal detected by MRS technology is extremely weak and only a few tens of nano-volt, so that it is susceptible to environmental noise, such as harmonic interference or some single-frequency interference with fixed-frequency, spike noise,random white noise and so on, leading to the desired signal completely covered by noise and characteristic parameters unable to be extracted accurately, which will affect the subsequent inversion interpretation. So, how to extract accurate characteristic parameters from the noisy MRS signal has become one of hot spots and difficulties in the MRS denoising domain.Independent component analysis(ICA) is a valid method means of data statistics and can separate the observed signals in the case of mixed mechanism of signal and noise unknown. In view of the above mentioned advantages, ICA has been widely applied in the domains of speech signals separation, biosignal processing, image processing and some other fields.On the base of consulting lots of reference and analyzing gaussianity of full-wave MRS signal and common environmental noise by their kurtosis value, this paper determines a nonlinear function suitable for separating noisy MRS signal and researches a noise suppression method of full-wave MRS signal based on a fastfixed-point algorithm for independent component analysis(Fast ICA).Firstly, Aiming at the case of single channel MRS signal, a digital orthogonal method is adopted to construct some extra observed signals combined with the existing single channel MRS signal as the input signal of ICA to solve the undetermined blind source separation which is one of common problems in ICA algorithm. A spectrum correcting method and a prefixing reference method are proposed to settle the problem of variable amplitude of separated signals after ICA that is another common problem of ICA. The results processed by this proposed algorithm(SNR is increased about 29 dB) and traditional filter are contrastively analyzed, and show that the performance of ICA algorithm is superior to of traditional filter.Secondly, for the multichannel MRS signals, a method of increasing signal length is presented to solve the ambiguous problem that MRS signal and white noise are not separated completely. The result is compared with that processed by self-adaptive filtering algorithm shows that ICA has a better performance both on signal wave and improvement of SNR increased about 50 dB.Then, envelope curve of the separated MRS signal is fitted by linear and nonlinear method, at the same time, initial amplitude and relaxation time are extracted and fitting error is calculated. In addition, fitting error of characteristic parameters extracted by this two fitting means in different cases is compared in order to determine a suitable fitting method for full-wave MRS signal.Finally, the proposed algorithm is applied to the processing of lab data and field data. The results indicate that the algorithm in this paper is able to remove environmental noise, and obtain the desired signal which has verified practicability of this proposed algorithm. |
PDF Full Text Request