计算机工程

• •    

一种改进的SPD-UNet地震断层图像识别模型

  

  • 发布日期:2020-12-11

A SPD-UNet image segmentation model for seismic fault recognition

  • Published:2020-12-11

摘要: 断层是控制油气田形成和分布的主要因素,查明断层的形态及分布是油气勘探的重要任务,因此断层识别在油气勘探领域起着重要作用。近年来,深度学习在油气领域的应用越来越多。本文基于Attention-UNet图像分割模型,提出一种改进的网络结构SPD-UNet模型,用于地震断层的图像识别,模型利用神经网络强大的编码和解码功能来实现断层图像的自动识别。SPD-UNet引入了空洞卷积,在扩大感受野的同时可以有效避免信息损失,增强图像特征的提取;同时空洞卷积采用金字塔结构,利用分层的思想进行特征融合以弥补空洞卷积带来的局部信息损失,提升断层识别精度。在胜利油田某区块的验证结果表明,应用SPD-UNet模型进行地震断层的预测,识别效果良好。

Abstract: Fault is the main factor which controls the formation and distribution of oil and gas fields, and finding out the shape and distribution of fault is an important task, so recognition of faults plays an important role in the domain of Oil-Gas exploration. In recent years, deep learning has been applied more and more in oil and gas field. Based on the Attention-UNet model, this paper proposes an improved SPD-UNet model of network structure for image recognition of earthquake faults. The model uses the powerful encoding and decoding functions of neural network to realize automatic recognition of fault images. SPD-UNet introduces dilated convolution, which can effectively avoid information loss and enhance image feature extraction while expanding receptive field. .At the same time, the dilated convolution adopts pyramid structure, and the thought of stratification is used to carry out feature fusion to compensate for the local information loss caused ,then improve the fault identification accuracy. The verification results in a block of Shengli Oilfield show that the SPD-UNet model is applied to the prediction of seismic fault and the identification effect is good.