Research Of Seismic Reflection Pattern Recognition Method Based On Intelligent Feature Extraction

In the exploration and prediction research of oil and gas fields,eismic facies are usually used to predict underground geological structures,and well-separated seismic facies maps have high application value in the mining process.Identifying the reflection mode of a seismic signal is one of the main methods for dividing the seismic phase.If the high-dimensional seismic signal is directly processed,it is easy to cause problems such as high computational complexity,dimensionality disaster,and fuzzy division effect.Therefore,the seismic reflection mode can only be characterized by features.However,most of the feature extraction methods of prestack seismic signals only generate a single feature,which can not fully represent the reflection mode of prestack seismic signals.Most post-stack seismic signals need to be manually involved in the feature extraction process,and data-driven intelligent feature learning cannot be achieved.Therefore,based on seismic signal feature extraction,combined with seismic signal preprocessing and classification and recognition algorithms,the seismic reflection pattern recognition method is studied.The main innovations are as follows:First,propose a prestack seismic reflection pattern recognition method based on multi-scale feature fusion(1)Due to the large amount of pre-stack seismic signal data and the scarcity of the label quantity,the deep convolutional generative adversarial network is used to generate a multi-layer feature from the prestack seismic signal.The generative adversarial network model is currently the best performance unsupervised learning model,which can extract a series of features from the low layer to the upper layer from the prestack seismic signal.(2)Based on the improved deep convolutional generative adversarial network,a new feature extraction method based on multi-scale feature fusion is proposed.Through the multi-scale fusion of low-level features and high-level features,the complete characterization of the reflection mode of the prestack seismic signal can be obtained,so that the pre-stack seismic reflection mode can be more accurately identified.Finally,the fuzzy self-organizing map neural network is introduced to classify the obtained fusion features to generate the final prestack seismic phase map.Second,propose a method for post-stack seismic reflection pattern recognition based on intelligent feature learningTraditional methods often use artificially designed or screened indicators to extract the characteristics of post-stack seismic signals.These methods typically require a large amount of work and rely heavily on expert knowledge to automate quantification.Aiming at these problems,this paper first proposes an intelligent feature learning method for post-stack seismic signals based on matrix decomposition.This method can automatically learn the characteristics of post-stack seismic signals under data driving.The fuzzy self-organizing map neural network is introduced to classify the learned features and generate post-stack seismic phase maps.This post-stack seismic reflection pattern recognition method can improve the recognition effect while reducing the workload.