The Study Of Marine Controlled Source Electromagnetic Method De-noising Based On Feature Learning

The marine controlled-source electromagnetic method(CSEM)is a frequency domain detection technology predominately used for hydrocarbon exploration and reservoir estimation.The magnitude of marine CSEM signal attenuates with the augment of offset between the transmitter and receiver.Especially when the offset is large,the marine CSEM signal is easily contaminated by surrounding noises.De-noising is a crucial procedure for improving data accuracy and credibility.Presently,the studies on marine CSEM de-noising are not sufficient.This paper applies mathematical morphology filtering algorithm to the de-noising procedure based on the features of noises,and it applies Compressive Sensing(CS)theory and dictionary learning algorithm based on the feature of effective signal.Besides,this paper induced Random Forest(RF)algorithm to learn the characteristics of high resistivity layer.Mathematical morphology filtering is based on the similarity between the structural element and the aimed signal.This paper chooses line and triangle elements to suppress the influence caused by impulse noise.And the later numerical experiment proves the effectiveness of this method.CS is a signal recover theory when sampling is insufficient.It includes spare representation and reconfiguration algorithm,and they can be used for de-noising when the noise is not spare in the selected spare space.This paper applies orthogonal matching pursuit algorithm in the simulation signal de-noising experiment.And the simulated signal is affected by four types of noises.The results indicate DST and DST-Wavelet dictionaries are suitable for marine CSEM signal de-noising.Besides,this paper firstly applies dictionary learning in the marine CSEM de-noising numerical experiment.There are two types of learning dictionary based on whether giving an initial dictionary.Then these two dictionaries are tested in the numerical experiment,and the outcome indicates that they are suitable for marine CSEM signal de-noising.The RF algorithm has many advantages such as strong learning ability,high algorithm robustness and high dimensional adaptiveness.It has been successfully applied in many engineering fields.The RF model is generated through large number of numerical simulation signals,and it contains the characters of high resistivity layer.The later numerical experiment proves the accuracy and effectiveness of this algorithm.At last,mathematical morphology filtering,DST dictionary,DST-Wavelet dictionary and these two learning dictionaries are tested in field data de-noising procedure.And their noise suppressing performances are analyzed through MVT plot in this part.The result turns out the effectiveness of CS under the learning dictionaries.However,the performances of morphology filtering are not ideal.Besides,the field data is applied to RF algorithm,too.The comparison with Occam results proves the effectiveness of RF algorithm.