Seismic Wavelet Classification And Recognition Method Based On Deep Learning In Noisy Environment

High-quality seismic data is an important reference that we know about geological structures,but in the actual collected seismic data,some random noises are often mixed with the effective waves,which greatly limits the extraction of effective geological information.So this link of noise reduction is especially significant and critical.The noise reduction processing of seismic data can be started from many different angles.In this paper,we select the simultaneous noise reduction and the reconstruction of seismic data?the classified identification of seismic subwaves.To complete the simultaneous noise reduction and reconstruction of seismic data,a method of integrating convolution neural noise reducer with convex set projection frame is proposed.Firstly,the simulated seismic record are formed into a data set,and the noise level of convolution neural network noise reducer is obtained by training;Then,combine the denoiser with the convex set projection framework for multiple iteration training;Finally,through comparing and analyzing the pure data,under-sampling data,the POCS algorithm based on f-x,the POCS algorithm based on curvelet and the result map processed by the method,it is concluded that the CNN-POCS method is feasible and practical,and can be used to reconstruct seismic data better while removing noise.In order to complete the research of seismic wavelet classification and recognition,a convolution neural network model is built.First,preprocess the data: segment the noisy seismic wavelet sequence to obtain the complete wavelet,label different types of seismic wavelets,and stack a variety of different noises on the seismic signal to obtain a data set;then,obtain image features by training convolutional neural network;Finally,we use CNN trained network to classify and identify seismic wavelet.The experimental results show that the method achieves high classification rate and high accuracy recognition under low SNR.