Research On Focus Location Method Of Shallow Seismic Source Based On Multispectral Image Fusion

The location of shallow seismic source is a hot topic in shallow seismic exploration,which is widely used in shallow explosion,engineering blasting,coal exploration and other fields.At present,shallow location mainly refers to deep source location method,which can be divided into three types :（1）Traveltime-based location method;（2）velocity and position union-based Location method;（3）Time-inverse migration-based Location method.Because travel time is not necessary for it,and can directly use all the energy of the wave,it directly uses all the information of waveform,and can realize the location under the condition of low SNR,which is a very popular location method in recent years.However,compared with deep strata,shallow strata are more complex,and their vibration wave transmission is more variable in time and space,so that the detected vibration wave has the characteristics of complex waveform,multifrequency waveform aliasing and severe dispersion,if the deep source location method or the full-frequency signal is used to directly locate the shallow source,it will produces large location error,and it is hard to meet the demand of high precision location of shallow underground source.Aiming at the above problems,this thesis focuses on the high-precision localization method of underground shallow seismic sources using eigenfrequency focused imaging and multispectral fusion.According to the characteristic that the frequency of vibration wave is not affected by the medium when it propagates in the stratum structure,the wide spectrum characteristics of the explosion shock signal and the advantages of using the high-frequency signal for reverse-time imaging to be more focused,In this thesis,on the premise of inverse time-focusing imaging using principal frequency components,deep learning method is adopted to achieve the positioning method of multi-spectral image fusion,which can improve the source location accuracy of shallow underground space.Firstly,the multifrequency principal component decomposition and extraction of sensor information are realized by using variational modal decomposition（VMD）method;Secondly,the reverse-time focusing method is used to perform reverse-time inversion and reconstruction of each modal frequency component in turn to form the corresponding three-dimensional energy field focused image;Finally,using the 3D-UNet deep learning model combined with the attention mechanism to realize the fusion of multi-spectral energy field focused images,and generate a fusion network model that can adaptively adjust the multi-spectral energy field coefficients,and this method is compared with the time-inverse focusing method,QPSO and neural network localization methods under the full frequency signal,and its effectiveness is verified.In this thesis,the method presented has been tested by simulation and field engineering The results show that,based on the main frequency component inverse-time focus imaging,this thesis uses deep learning to perform multi-spectral image fusion to achieve high-precision shallow source location.In the field engineering test of underground explosion in the area of100*100*50m,the RMSE positioning error of the method in this thesis is within 0.5 m.