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

The prestack seismic reflection pattern is a unique feature that can distinguish different prestack seismic reflection signals.The process of prestack seismic reflection pattern identification and analysis is to process the prestack seismic reflection signals,extract the prestack seismic reflection features for identification,and finally generate the prestack seismic facies map.Based on the generated prestack seismic facies map,it can predict the distribution of subsurface oil and gas reservoirs,provide a reference basis for oil and gas reservoir exploration,save exploration costs.So it is of great importance.Since the prestack seismic signal is of high dimensionality,the direct recognition of prestack seismic reflection pattern is likely to cause many problems such as dimensional catastrophe,too large computation,and inaccurate recognition effect,so the prestack seismic signal is usually subjected to feature extraction and then recognition.Traditional feature extraction methods such as principal component analysis cannot identify the nonlinear structure in the prestack seismic signal and cannot completely characterize the reflection pattern of the prestack seismic signal.Therefore,this paper is dedicated to the study of intelligent feature extraction methods for prestack seismic signals and proposes a prestack seismic reflection pattern analysis and recognition method based on deep correlation mining and robust adversarial learning,with the following specific innovations.(1)To address the problem that traditional methods cannot completely characterize the reflection patterns of prestack seismic signals,we introduce a deep association mining network,construct pseudo graph loss to explore the correlation between pairs of prestack seismic image samples,construct pseudo label loss to fully utilize the hidden category information of prestack seismic image samples,use local robustness loss to maintain the invariance of geometric transformation of prestack seismic image samples,and introduce triple The four loss functions are combined to explore the information that can fully characterize the reflection patterns of prestack seismic images.In order to enhance the spatial invariance of the prestack seismic features and finally enhance the feature extraction capability.(2)To address the problem of inaccurate prestack seismic facies maps generated by deep association mining networks under the counter-attack perturbation samples,we try to combine the robust counter-attack learning mechanism with the prestack seismic reflection pattern analysis task,and use the counter-attack strategy to learn the perturbation in the prestack feature space to explore those prestack image samples that are easy to generate different clustering results,and propose the corresponding defense algorithm to eliminate the impact of the perturbation.The defense strategy and the adversarial attack strategy can optimize the feature extraction performance of the robust adversarial learning network;to address the problem that there are irrelevant information of other types of geological structures interfering with the recognition of specific geological structures,a prestack attention mechanism is introduced on the basis of the robust adversarial learning network to focus on the key information that can characterize specific prestack seismic reflection patterns,thus enhancing the feature extraction capability of the network.The high-precision prestack seismic facies maps generated on both model data and actual data representing different geological structures demonstrate the feasibility and effectiveness of the two methods proposed in this paper.