Seismic Signal Recognition Based On Deep Neural Network Model

Seismic signal recognition related research has been greatly developed under the continuous improvement of the signal-to-noise ratio of the acquired signals and the promotion of digital recording technology.However,this traditional seismic signal recognition method is based on the causal logic that is known to distinguish the similarities and differences between different signals.It only focuses on the so-called "effective signal",suppresses noise,and selects features that can express most of the seismic signal information.Carrying out signal identification,ignoring the expression of signal integrity.With the development of deep learning and big data technology,through the use of big data-driven deep neural network technology to mine the "don't know" in the seismic signal data,the hidden rich physical parameter information is established to establish a set of seismic signal parameter representations Model,study the difference between seismic signals and other incoherent signals,and realize the recognition of seismic signals becomes possible.A deep neural network is a neural network with a multi-hidden layer structure.It has high-level abstract characterization capabilities and can automatically extract the essential characteristics of data without manual intervention.It has been successfully applied in various fields of production and life,such as speech recognition,Image recognition,automatic driving and target detection,etc.Based on the above advantages of deep neural network,this paper presents a new seismic signal recognition method based on deep neural network model.Compared with traditional seismic signal recognition methods,this method has higher accuracy and stronger model robustness.The specific research content of this article is given below:(1)In order to reduce the impact of manual intervention in the signal feature extraction process,based on the Le Net5 convolutional neural network model,a seismic signal-based convolutional neural network QConv Net(Quake Convolutional Neural Networks)recognition model is proposed.In the experiment,the public seismic signal data set under the earthquake monitoring catalog of the National Geological Survey of the United States was used as the experimental data,and the sequential three-channel seismic waveform data was used as the input of the QConv Net neural network model.The experimental results are compared with the experimental results of traditional machine learning models such as support vector machines,decision trees,and logistic regression.The three evaluation indicators of accuracy,recall and F1-score are more excellent,showing a good two-class classification effect.(2)In order to further improve the performance of the deep neural network model in seismic signal recognition,on the basis of the QConv Net neural network model,by introducing three optimized structures of residual module,multi-scale module and shrinkage module,a hybrid model based on multi-scale module and shrinkage module is proposed.The residual neural network model DRISN(Deep Residual Inception Shrinkage Network).In order to verify the performance of the network model in seismic signal recognition,the experiment uses the seismic signal spectrogram data as input,and compares the deep neural network models such as DRISN with the residual neural network DRSN(Deep Residual Shrinkage Network)introduced into the shrinkage module.The DRISN model has a recognition accuracy of up to 95.62% in the seismic signal recognition experiment,which has better recognition performance and higher accuracy than other experimental models,which shows the effectiveness of this model in seismic signal recognition tasks.