Search

CN-121995466-A - Earthquake body wave detection method, equipment and medium based on dual-path deep learning

CN121995466ACN 121995466 ACN121995466 ACN 121995466ACN-121995466-A

Abstract

According to the earthquake body wave detection method, the earthquake body wave detection equipment and the earthquake body wave detection medium based on the double-path deep learning, an initial characteristic projection module is arranged for carrying out initial characteristic conversion and dimension reduction on an input waveform image and an F-K frequency spectrum image; the method comprises the steps of setting a dual-path feature extraction module to synchronously extract deep features from input data of two different forms of waveforms and spectrums, setting a multi-level feature fusion module to effectively integrate the features extracted by dual paths, setting a classification output module to receive fully fused high-level semantic features, firstly compressing a feature map into a feature vector through a global average pooling layer, then calculating through a full-connection layer, and finally outputting probability distribution of the seismic slice belonging to the categories of bulk waves, mixed noise or surface waves to finish automatic identification and classification of signals. The invention realizes high-precision and high-efficiency bulk wave signal detection by a dual-path deep learning architecture and simultaneously utilizing the complementary characteristics of the waveform and the F-K frequency spectrum.

Inventors

  • CHEN MEIWEN
  • FANG XIN
  • WANG ZHIYU
  • MA TIANYI
  • CHENG FAN

Assignees

  • 安徽大学

Dates

Publication Date
20260508
Application Date
20260120

Claims (9)

  1. 1. A seismic body wave detection method based on dual-path deep learning is characterized in that the following steps are executed by computer equipment, Setting an initial characteristic projection module, a dual-path characteristic extraction module, a multi-level characteristic fusion module and a classification output module; The initial characteristic projection module is used for carrying out preliminary characteristic conversion and dimension reduction on the input waveform image and the F-K spectrum image, and laying a foundation for subsequent depth characteristic extraction; The dual-path feature extraction module synchronously extracts complementary deep features with discriminant characteristics from input data of two different forms of waveforms and frequency spectrums; The multi-level feature fusion module is responsible for effectively integrating the features extracted by the dual paths so as to comprehensively utilize the advantages of the two types of features; The classification output module receives the fully fused high-level semantic features, firstly compresses the feature map into a feature vector through the global average pooling layer, then calculates through the full-connection layer, and finally outputs probability distribution of the seismic slice belonging to the categories of bulk waves, mixed noise or surface waves, thereby completing automatic identification and classification of signals.
  2. 2. The dual path deep learning based seismic body wave detection method of claim 1, wherein: The initial feature projection module performs the same operation on both input paths: Firstly, a convolution kernel of 7x7 is used for convolution operation, local features are initially extracted and input channels are mapped to a 64-dimensional feature space, then a maximum pooling operation of 3x3 is used for downsampling, and the space size of a feature map is halved while key features are reserved.
  3. 3. The dual path deep learning based seismic body wave detection method of claim 2, wherein: The dual-path feature extraction module consists of DenseNet paths and ResNet paths which are arranged in parallel and comprises four feature extraction stages which are connected in sequence, wherein each stage internally comprises the two parallel convolution blocks, each convolution block is formed by repeatedly stacking basic convolution layers, the repetition times of each stage are configured as [3,6,12,8], and the number of output channels is sequentially configured as [64,128,256,512], so that the channel amplification and the space dimension reduction of a feature map are realized; The DenseNet paths adopt a dense connection mode, and the output of each layer and the characteristics of all the front layers are spliced in the channel dimension, so that the high reuse of the characteristics and the smooth transmission of the information flow are realized. The basic unit of the system is sequentially formed by 1x1 convolution, 3x3 convolution and 1x1 convolution; The ResNet paths adopt a jump connection mode, the gradient vanishing problem in deep network training is effectively relieved by directly adding the output and the input of the blocks, the training process is stabilized, the model can be effectively learned, and the basic units of the method sequentially comprise 1x1 convolution for dimension reduction and feature fusion, 3x3 convolution for core space feature extraction and 1x1 convolution for channel expansion.
  4. 4. The method for detecting seismic body waves based on dual-path deep learning of claim 3, wherein the multi-level feature fusion module performs feature fusion by using a standard transition layer after the first three feature extraction stages, the transition layer is composed of a layer-by-one convolution and a three-by-three maximum pooling operation sequence, the one-by-one convolution is used for fusing features and controlling the number of channels, and the maximum pooling is used for reducing the space size of a feature map so as to improve the calculation efficiency; after the final fourth feature extraction stage, the standard transition layer is replaced with an attention-based feature fusion module.
  5. 5. A seismic body wave detection method based on dual-path deep learning is characterized by comprising the following steps, S1, data preprocessing, S2, model classification, S3, wavelength reconstruction; In step S1, data preprocessing refers to the preparation and conversion of original continuous passive seismic records; in step S2, model classification refers to that the paired waveform image and the F-K spectrum image obtained after pretreatment are input into a dual-path deep learning model for automatic identification and classification; In step S3, the wave field reconstruction means that all the high signal-to-noise ratio data segments marked as 'body waves' are screened out by utilizing the output result of the model classification stage, a virtual shot set is reconstructed by adopting the seismic interferometry technology based on the high-quality segments, and then the high-resolution seismic reflection section capable of clearly revealing the deep geological structure is finally generated through the conventional seismic imaging processes of velocity analysis, normal time difference correction and superposition.
  6. 6. The method for detecting seismic body waves based on dual-path deep learning of claim 5, wherein S1 comprises, Cutting the continuous record into a plurality of short time window slices of fixed length, performing inter-channel normalization processing on the waveform data in each slice to compensate for amplitude differences between different receivers, applying a band pass filter to suppress low frequency ambient noise and high frequency instrument noise, preserving the effective bulk wave signal band, simultaneously performing space-time Fourier transform on each slice data to generate a corresponding frequency-wave number spectrum (F-K) image, and finally generating paired waveform images and F-K spectrum images for each segment of seismic data.
  7. 7. The method for detecting earthquake body waves based on the dual-path deep learning of claim 5, wherein in S2, the model outputs the probability of the earthquake slice belonging to the categories of body waves, surface waves or noise finally through a series of calculations including initial characteristic projection, dual-path characteristic extraction and multi-level characteristic fusion in the model.
  8. 8. A computer readable storage medium storing a computer program, which when executed by a processor causes the processor to perform the steps of the method according to any one of claims 1 to 7.
  9. 9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the computer program, when executed by the processor, causes the processor to perform the steps of the method according to any of claims 1 to 7.

Description

Earthquake body wave detection method, equipment and medium based on dual-path deep learning Technical Field The invention relates to the technical field of seismic body wave detection, in particular to a dual-path deep learning-based seismic body wave detection method, dual-path deep learning-based seismic body wave detection equipment and a storage medium. Background Seismic imaging is a key technology for researching underground structures, advancing seismology research and improving earthquake prediction capability. Conventional seismic data acquisition methods mainly include Active Source Seismic Exploration (ASSE) and Passive Source Seismic Exploration (PSSE). Compared with ASSE, PSSE has the advantages of low cost, small environmental impact, high safety, capability of generating a large amount of data and the like, and therefore, the PSSE becomes an important direction for future algorithm and technology innovation. In passive source seismic exploration, bulk waves are critical for deep earth imaging because they can penetrate deep layers and provide higher resolution and sensitivity to discontinuities. However, passive source data is often dominated by noise such as surface waves, and the body wave signal is weak and difficult to extract. Conventional bulk wave detection methods rely mainly on seismic interferometry techniques (such as cross-correlation, cross-coherence and multi-dimensional deconvolution) to reconstruct a virtual source, but these methods suffer from the following drawbacks: 1. The quality of the virtual shot set is poor and is limited by uneven source distribution, false reflection caused by artificial noise and surface wave energy interference; 2. Traditional methods (such as illumination diagnosis, F-K filtering and artificial vision inspection) rely on manual parameter adjustment and subjective judgment, and are low in efficiency and fuzzy in result; 3. the existing deep learning method is mostly limited to array data, cannot be suitable for short-section body wave tasks, depends on manual frequency range adjustment, and is low in calculation efficiency and insufficient in data utilization. Therefore, how to realize automatic, high-precision and expandable body wave signal detection to improve the passive source deep reflection imaging quality is a technical problem to be solved in the field. Disclosure of Invention The invention provides a dual-path deep learning-based earthquake body wave detection method, dual-path deep learning-based earthquake body wave detection equipment and a storage medium, which can at least solve one of the technical problems in the background technology. In order to achieve the above purpose, the present invention adopts the following technical scheme: A method for detecting earthquake body wave based on dual-path deep learning comprises the following steps executed by computer equipment, Setting an initial characteristic projection module, a dual-path characteristic extraction module, a multi-level characteristic fusion module and a classification output module; The initial characteristic projection module is used for carrying out preliminary characteristic conversion and dimension reduction on the input waveform image and the F-K spectrum image, and laying a foundation for subsequent depth characteristic extraction; The dual-path feature extraction module synchronously extracts complementary deep features with discriminant characteristics from input data of two different forms of waveforms and frequency spectrums; The multi-level feature fusion module is responsible for effectively integrating the features extracted by the dual paths so as to comprehensively utilize the advantages of the two types of features; The classification output module receives the fully fused high-level semantic features, firstly compresses the feature map into a feature vector through the global average pooling layer, then calculates through the full-connection layer, and finally outputs probability distribution of the seismic slice belonging to the categories of bulk waves, mixed noise or surface waves, thereby completing automatic identification and classification of signals. In yet another aspect, the invention also discloses a computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of the method as described above. In yet another aspect, the invention also discloses a computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of the method as above. According to the technical scheme, the earthquake body wave detection method based on the dual-path deep learning solves the technical problems of low automation degree, poor precision, dependence on manual intervention, insufficient generalization capability of the existing deep learning model and the like in passive earthquake